rdelaage
/
postgresql
mirror of https://git.postgresql.org/git/postgresql.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
postgresql/src/backend/access/heap/heapam.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

10229 lines
315 KiB
C
Raw Normal View History

Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*-------------------------------------------------------------------------
*
Change my-function-name-- to my_function_name, and optimizer renames.
1999-02-14 00:22:53 +01:00
* heapam.c
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
* heap access method code
*
Update copyright for 2023 Backpatch-through: 11
2023-01-02 21:00:37 +01:00
* Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
Add: * Portions Copyright (c) 1996-2000, PostgreSQL, Inc to all files copyright Regents of Berkeley. Man, that's a lot of files.
2000-01-26 06:58:53 +01:00
* Portions Copyright (c) 1994, Regents of the University of California
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*
*
* IDENTIFICATION
Remove cvs keywords from all files.
2010-09-20 22:08:53 +02:00
* src/backend/access/heap/heapam.c
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*
*
* INTERFACE ROUTINES
* heap_beginscan - begin relation scan
* heap_rescan - restart a relation scan
* heap_endscan - end relation scan
* heap_getnext - retrieve next tuple in scan
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
* heap_fetch - retrieve tuple with given tid
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
* heap_insert - insert tuple into a relation
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
* heap_multi_insert - insert multiple tuples into a relation
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
* heap_delete - delete a tuple from a relation
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
* heap_update - replace a tuple in a relation with another tuple
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*
* NOTES
* This file contains the heap_ routines which implement
* the POSTGRES heap access method used for all POSTGRES
* relations.
*
*-------------------------------------------------------------------------
*/
Change #include's to use <> and "" as appropriate.
1999-07-16 01:04:24 +02:00
#include "postgres.h"
First pass at fixing my own mistakes Mainly...fix up the includes I removed, as well as prototypes Pointed out by D'Arcy
1996-10-20 10:32:11 +02:00
Add WAL consistency checking facility. When the new GUC wal_consistency_checking is set to a non-empty value, it triggers recording of additional full-page images, which are compared on the standby against the results of applying the WAL record (without regard to those full-page images). Allowable differences such as hints are masked out, and the resulting pages are compared; any difference results in a FATAL error on the standby. Kuntal Ghosh, based on earlier patches by Michael Paquier and Heikki Linnakangas. Extensively reviewed and revised by Michael Paquier and by me, with additional reviews and comments from Amit Kapila, Álvaro Herrera, Simon Riggs, and Peter Eisentraut.
2017-02-08 21:45:30 +01:00
#include "access/bufmask.h"
Revert "Rename files and headers related to index AM" This follows multiple complains from Peter Geoghegan, Andres Freund and Alvaro Herrera that this issue ought to be dug more before actually happening, if it happens. Discussion: https://postgr.es/m/20191226144606.GA5659@alvherre.pgsql
2019-12-27 00:09:00 +01:00
#include "access/genam.h"
Change #include's to use <> and "" as appropriate.
1999-07-16 01:04:24 +02:00
#include "access/heapam.h"
Split heapam_xlog.h from heapam.h The heapam XLog functions are used by other modules, not all of which are interested in the rest of the heapam API. With this, we let them get just the XLog stuff in which they are interested and not pollute them with unrelated includes. Also, since heapam.h no longer requires xlog.h, many files that do include heapam.h no longer get xlog.h automatically, including a few headers. This is useful because heapam.h is getting pulled in by execnodes.h, which is in turn included by a lot of files.
2012-08-29 01:02:00 +02:00
#include "access/heapam_xlog.h"
Split tuptoaster.c into three separate files. detoast.c/h contain functions required to detoast a datum, partially or completely, plus a few other utility functions for examining the size of toasted datums. toast_internals.c/h contain functions that are used internally to the TOAST subsystem but which (mostly) do not need to be accessed from outside. heaptoast.c/h contains code that is intrinsically specific to the heap AM, either because it operates on HeapTuples or is based on the layout of a heap page. detoast.c and toast_internals.c are placed in src/backend/access/common rather than src/backend/access/heap. At present, both files still have dependencies on the heap, but that will be improved in a future commit. Patch by me, reviewed and tested by Prabhat Sabu, Thomas Munro, Andres Freund, and Álvaro Herrera. Discussion: http://postgr.es/m/CA+TgmoZv-=2iWM4jcw5ZhJeL18HF96+W1yJeYrnGMYdkFFnEpQ@mail.gmail.com
2019-07-08 17:58:05 +02:00
#include "access/heaptoast.h"
Change #include's to use <> and "" as appropriate.
1999-07-16 01:04:24 +02:00
#include "access/hio.h"
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
#include "access/multixact.h"
Create an infrastructure for parallel computation in PostgreSQL. This does four basic things. First, it provides convenience routines to coordinate the startup and shutdown of parallel workers. Second, it synchronizes various pieces of state (e.g. GUCs, combo CID mappings, transaction snapshot) from the parallel group leader to the worker processes. Third, it prohibits various operations that would result in unsafe changes to that state while parallelism is active. Finally, it propagates events that would result in an ErrorResponse, NoticeResponse, or NotifyResponse message being sent to the client from the parallel workers back to the master, from which they can then be sent on to the client. Robert Haas, Amit Kapila, Noah Misch, Rushabh Lathia, Jeevan Chalke. Suggestions and review from Andres Freund, Heikki Linnakangas, Noah Misch, Simon Riggs, Euler Taveira, and Jim Nasby.
2015-04-30 21:02:14 +02:00
#include "access/parallel.h"
Improve our #include situation by moving pointer types away from the corresponding struct definitions. This allows other headers to avoid including certain highly-loaded headers such as rel.h and relscan.h, instead using just relcache.h, heapam.h or genam.h, which are more lightweight and thus cause less unnecessary dependencies.
2008-06-19 02:46:06 +02:00
#include "access/relscan.h"
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
#include "access/subtrans.h"
Move syncscan.c to src/backend/access/common. Since the tableam.c code needs to make use of the syncscan.c routines itself, and since other block-oriented AMs might also want to use it one day, it didn't make sense for it to live under src/backend/access/heap. Reviewed-by: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CA%2BhUKGLCnG%3DNEAByg6bk%2BCT9JZD97Y%3DAxKhh27Su9FeGWOKvDg%40mail.gmail.com
2020-07-29 06:46:58 +02:00
#include "access/syncscan.h"
Restructure some header files a bit, in particular heapam.h, by removing some unnecessary #include lines in it. Also, move some tuple routine prototypes and macros to htup.h, which allows removal of heapam.h inclusion from some .c files. For this to work, a new header file access/sysattr.h needed to be created, initially containing attribute numbers of system columns, for pg_dump usage. While at it, make contrib ltree, intarray and hstore header files more consistent with our header style.
2008-05-12 02:00:54 +02:00
#include "access/sysattr.h"
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
#include "access/tableam.h"
Allow include files to compile own their own. Strip unused include files out unused include files, and add needed includes to C files. The next step is to remove unused include files in C files.
2006-07-13 18:49:20 +02:00
#include "access/transam.h"
Final cleanup.
1999-07-16 07:00:38 +02:00
#include "access/valid.h"
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
#include "access/visibilitymap.h"
Allow include files to compile own their own. Strip unused include files out unused include files, and add needed includes to C files. The next step is to remove unused include files in C files.
2006-07-13 18:49:20 +02:00
#include "access/xact.h"
Move the backup-block logic from XLogInsert to a new file, xloginsert.c. xlog.c is huge, this makes it a little bit smaller, which is nice. Functions related to putting together the WAL record are in xloginsert.c, and the lower level stuff for managing WAL buffers and such are in xlog.c. Also move the definition of XLogRecord to a separate header file. This causes churn in the #includes of all the files that write WAL records, and redo routines, but it avoids pulling in xlog.h into most places. Reviewed by Michael Paquier, Alvaro Herrera, Andres Freund and Amit Kapila.
2014-11-06 12:52:08 +01:00
#include "access/xlog.h"
#include "access/xloginsert.h"
Restructure some header files a bit, in particular heapam.h, by removing some unnecessary #include lines in it. Also, move some tuple routine prototypes and macros to htup.h, which allows removal of heapam.h inclusion from some .c files. For this to work, a new header file access/sysattr.h needed to be created, initially containing attribute numbers of system columns, for pg_dump usage. While at it, make contrib ltree, intarray and hstore header files more consistent with our header style.
2008-05-12 02:00:54 +02:00
#include "access/xlogutils.h"
Change #include's to use <> and "" as appropriate.
1999-07-16 01:04:24 +02:00
#include "catalog/catalog.h"
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
#include "commands/vacuum.h"
Final cleanup.
1999-07-16 07:00:38 +02:00
#include "miscadmin.h"
Revise pgstats stuff to fix the problems with not counting accesses generated by bitmap index scans. Along the way, simplify and speed up the code for counting sequential and index scans; it was both confusing and inefficient to be taking care of that in the per-tuple loops, IMHO. initdb forced because of internal changes in pg_stat view definitions.
2005-10-06 04:29:23 +02:00
#include "pgstat.h"
Use atomic ops to hand out pages to scan in parallel scan. With a lot of CPUs, the spinlock that protects the current scan location in a parallel scan can become a bottleneck. Use an atomic fetch-and-add instruction instead. David Rowley Discussion: https://www.postgresql.org/message-id/CAKJS1f9tgsPhqBcoPjv9_KUPZvTLCZ4jy%3DB%3DbhqgaKn7cYzm-w@mail.gmail.com
2017-08-16 15:18:41 +02:00
#include "port/atomics.h"
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
#include "port/pg_bitutils.h"
Move BufferGetPageSize and BufferGetPage from bufpage.h to bufmgr.h. It is more logical that way, and also it reduces the amount of unnecessary includes in bufpage.h, which is widely used. Zdenek Kotala. My previous patch to bufpage.h should also have credited him as author, but I forgot (sorry about that).
2008-06-09 00:00:48 +02:00
#include "storage/bufmgr.h"
Update FSM on WAL replay. This is a bit limited; the FSM is only updated on non-full-page-image WAL records, and quite arbitrarily, only if there's less than 20% free space on the page after the insert/update (not on HOT updates, though). The 20% cutoff should avoid most of the overhead, when replaying a bulk insertion, for example, while ensuring that pages that are full are marked as full in the FSM. This is mostly to avoid the nasty worst case scenario, where you replay from a PITR archive, and the FSM information in the base backup is really out of date. If there was a lot of pages that the outdated FSM claims to have free space, but don't actually have any, the first unlucky inserter after the recovery would traverse through all those pages, just to find out that they're full. We didn't have this problem with the old FSM implementation, because we simply threw the FSM information away on a non-clean shutdown.
2008-10-31 20:40:27 +01:00
#include "storage/freespace.h"
Restructure some header files a bit, in particular heapam.h, by removing some unnecessary #include lines in it. Also, move some tuple routine prototypes and macros to htup.h, which allows removal of heapam.h inclusion from some .c files. For this to work, a new header file access/sysattr.h needed to be created, initially containing attribute numbers of system columns, for pg_dump usage. While at it, make contrib ltree, intarray and hstore header files more consistent with our header style.
2008-05-12 02:00:54 +02:00
#include "storage/lmgr.h"
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
#include "storage/predicate.h"
Split the shared-memory array of PGPROC pointers out of the sinval communication structure, and make it its own module with its own lock. This should reduce contention at least a little, and it definitely makes the code seem cleaner. Per my recent proposal.
2005-05-19 23:35:48 +02:00
#include "storage/procarray.h"
Prevent WAL logging when COPY is done in the same transation that created it. Simon Riggs
2007-01-25 03:17:26 +01:00
#include "storage/smgr.h"
Add a C API for parallel heap scans. Using this API, one backend can set up a ParallelHeapScanDesc to which multiple backends can then attach. Each tuple in the relation will be returned to exactly one of the scanning backends. Only forward scans are supported, and rescans must be carefully coordinated. This is not exposed to the planner or executor yet. The original version of this code was written by me. Amit Kapila reviewed it, tested it, and improved it, including adding support for synchronized scans, per review comments from Jeff Davis. Extensive testing of this and related patches was performed by Haribabu Kommi. Final cleanup of this patch by me.
2015-10-16 23:25:02 +02:00
#include "storage/spin.h"
Allow read only connections during recovery, known as Hot Standby. Enabled by recovery_connections = on (default) and forcing archive recovery using a recovery.conf. Recovery processing now emulates the original transactions as they are replayed, providing full locking and MVCC behaviour for read only queries. Recovery must enter consistent state before connections are allowed, so there is a delay, typically short, before connections succeed. Replay of recovering transactions can conflict and in some cases deadlock with queries during recovery; these result in query cancellation after max_standby_delay seconds have expired. Infrastructure changes have minor effects on normal running, though introduce four new types of WAL record. New test mode "make standbycheck" allows regression tests of static command behaviour on a standby server while in recovery. Typical and extreme dynamic behaviours have been checked via code inspection and manual testing. Few port specific behaviours have been utilised, though primary testing has been on Linux only so far. This commit is the basic patch. Additional changes will follow in this release to enhance some aspects of behaviour, notably improved handling of conflicts, deadlock detection and query cancellation. Changes to VACUUM FULL are also required. Simon Riggs, with significant and lengthy review by Heikki Linnakangas, including streamlined redesign of snapshot creation and two-phase commit. Important contributions from Florian Pflug, Mark Kirkwood, Merlin Moncure, Greg Stark, Gianni Ciolli, Gabriele Bartolini, Hannu Krosing, Robert Haas, Tatsuo Ishii, Hiroyuki Yamada plus support and feedback from many other community members.
2009-12-19 02:32:45 +01:00
#include "storage/standby.h"
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
#include "utils/datum.h"
Final cleanup.
1999-07-16 07:00:38 +02:00
#include "utils/inval.h"
Change the relation_open protocol so that we obtain lock on a relation (table or index) before trying to open its relcache entry. This fixes race conditions in which someone else commits a change to the relation's catalog entries while we are in process of doing relcache load. Problems of that ilk have been reported sporadically for years, but it was not really practical to fix until recently --- for instance, the recent addition of WAL-log support for in-place updates helped. Along the way, remove pg_am.amconcurrent: all AMs are now expected to support concurrent update.
2006-07-31 22:09:10 +02:00
#include "utils/lsyscache.h"
Final cleanup.
1999-07-16 07:00:38 +02:00
#include "utils/relcache.h"
Rename snapmgmt.c/h to snapmgr.c/h, for consistency with other files. Per complaint from Tom Lane.
2008-03-26 19:48:59 +01:00
#include "utils/snapmgr.h"
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
#include "utils/spccache.h"
Finish reverting "recheck_on_update" patch. This reverts commit c203d6cf8 and some follow-on fixes, completing the task begun in commit 5d28c9bd7. If that feature is ever resurrected, the code will look quite a bit different from this, so it seems best to start from a clean slate. The v11 branch is not touched; in that branch, the recheck_on_update storage option remains present, but nonfunctional and undocumented. Discussion: https://postgr.es/m/20190114223409.3tcvejfhlvbucrv5@alap3.anarazel.de
2019-01-15 18:07:10 +01:00
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
static HeapTuple heap_prepare_insert(Relation relation, HeapTuple tup,
TransactionId xid, CommandId cid, int options);
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
static XLogRecPtr log_heap_update(Relation reln, Buffer oldbuf,
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Buffer newbuf, HeapTuple oldtup,
Fix inconsistencies and typos in the tree, take 9 This addresses more issues with code comments, variable names and unreferenced variables. Author: Alexander Lakhin Discussion: https://postgr.es/m/7ab243e0-116d-3e44-d120-76b3df7abefd@gmail.com
2019-08-05 05:14:58 +02:00
HeapTuple newtup, HeapTuple old_key_tuple,
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
bool all_visible_cleared, bool new_all_visible_cleared);
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
static Bitmapset *HeapDetermineColumnsInfo(Relation relation,
Bitmapset *interesting_cols,
Bitmapset *external_cols,
HeapTuple oldtup, HeapTuple newtup,
bool *has_external);
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
static bool heap_acquire_tuplock(Relation relation, ItemPointer tid,
LockTupleMode mode, LockWaitPolicy wait_policy,
bool *have_tuple_lock);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
static void compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
uint16 old_infomask2, TransactionId add_to_xmax,
LockTupleMode mode, bool is_update,
TransactionId *result_xmax, uint16 *result_infomask,
uint16 *result_infomask2);
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
static TM_Result heap_lock_updated_tuple(Relation rel, HeapTuple tuple,
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
ItemPointer ctid, TransactionId xid,
LockTupleMode mode);
Rename and relocate freeze plan dedup routines. Rename the heapam.c freeze plan deduplication routines added by commit 9e540599 to names that follow conventions for functions in heapam.c. Also relocate the functions so that they're next to their caller, which runs during original execution, when FREEZE_PAGE WAL records are built. The routines were initially placed next to (and followed the naming conventions of) conceptually related REDO routine code, but that scheme turned out to be kind of jarring when considered in a wider context. Author: Peter Geoghegan <pg@bowt.ie> Reported-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/20230109214308.icz26oqvt3k2274c@awork3.anarazel.de
2023-01-12 02:30:42 +01:00
static int heap_log_freeze_plan(HeapTupleFreeze *tuples, int ntuples,
xl_heap_freeze_plan *plans_out,
OffsetNumber *offsets_out);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
static void GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
uint16 *new_infomask2);
static TransactionId MultiXactIdGetUpdateXid(TransactionId xmax,
uint16 t_infomask);
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
static bool DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
LockTupleMode lockmode, bool *current_is_member);
Setup error context callback for transaction lock waits With this in place, a session blocking behind another one because of tuple locks will get a context line mentioning the relation name, tuple TID, and operation being done on tuple. For example: LOG: process 11367 still waiting for ShareLock on transaction 717 after 1000.108 ms DETAIL: Process holding the lock: 11366. Wait queue: 11367. CONTEXT: while updating tuple (0,2) in relation "foo" STATEMENT: UPDATE foo SET value = 3; Most usefully, the new line is displayed by log entries due to log_lock_waits, although of course it will be printed by any other log message as well. Author: Christian Kruse, some tweaks by Álvaro Herrera Reviewed-by: Amit Kapila, Andres Freund, Tom Lane, Robert Haas
2014-03-19 19:10:36 +01:00
static void MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
Relation rel, ItemPointer ctid, XLTW_Oper oper,
int *remaining);
static bool ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
Revert XactLockTableWait context setup in conditional multixact wait There's no point in setting up a context error callback when doing conditional lock acquisition, because we never actually wait and so the user wouldn't be able to see the context message anywhere. In fact, this is more in line with what ConditionalXactLockTableWait is doing. Backpatch to 9.4, where this was added.
2014-08-25 21:33:17 +02:00
uint16 infomask, Relation rel, int *remaining);
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
static void index_delete_sort(TM_IndexDeleteOp *delstate);
static int bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
static XLogRecPtr log_heap_new_cid(Relation relation, HeapTuple tup);
Harmonize heapam and tableam parameter names. Make sure that function declarations use names that exactly match the corresponding names from function definitions. Having parameter names that are reliably consistent in this way will make it easier to reason about groups of related C functions from the same translation unit as a module. It will also make certain refactoring tasks easier. Like other recent commits that cleaned up function parameter names, this commit was written with help from clang-tidy. Later commits will do the same for other parts of the codebase. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: David Rowley <dgrowleyml@gmail.com> Discussion: https://postgr.es/m/CAH2-WznJt9CMM9KJTMjJh_zbL5hD9oX44qdJ4aqZtjFi-zA3Tg@mail.gmail.com
2022-09-20 01:46:23 +02:00
static HeapTuple ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
bool *copy);
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
-Wall'd
1996-10-21 07:59:49 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* Each tuple lock mode has a corresponding heavyweight lock, and one or two
* corresponding MultiXactStatuses (one to merely lock tuples, another one to
* update them). This table (and the macros below) helps us determine the
* heavyweight lock mode and MultiXactStatus values to use for any particular
* tuple lock strength.
Avoid inconsistent type declaration Clang 3.3 correctly complains that a variable of type enum MultiXactStatus cannot hold a value of -1, which makes sense. Change the declared type of the variable to int instead, and apply casting as necessary to avoid the warning. Per notice from Andres Freund
2013-06-25 22:36:29 +02:00
*
* Don't look at lockstatus/updstatus directly! Use get_mxact_status_for_lock
* instead.
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
*/
static const struct
{
LOCKMODE hwlock;
Avoid inconsistent type declaration Clang 3.3 correctly complains that a variable of type enum MultiXactStatus cannot hold a value of -1, which makes sense. Change the declared type of the variable to int instead, and apply casting as necessary to avoid the warning. Per notice from Andres Freund
2013-06-25 22:36:29 +02:00
int lockstatus;
int updstatus;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
pgindent run for release 9.3 This is the first run of the Perl-based pgindent script. Also update pgindent instructions.
2013-05-29 22:58:43 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
tupleLockExtraInfo[MaxLockTupleMode + 1] =
{
{ /* LockTupleKeyShare */
AccessShareLock,
MultiXactStatusForKeyShare,
-1 /* KeyShare does not allow updating tuples */
},
{ /* LockTupleShare */
RowShareLock,
MultiXactStatusForShare,
-1 /* Share does not allow updating tuples */
},
{ /* LockTupleNoKeyExclusive */
ExclusiveLock,
MultiXactStatusForNoKeyUpdate,
MultiXactStatusNoKeyUpdate
},
{ /* LockTupleExclusive */
AccessExclusiveLock,
MultiXactStatusForUpdate,
MultiXactStatusUpdate
}
};
pgindent run for release 9.3 This is the first run of the Perl-based pgindent script. Also update pgindent instructions.
2013-05-29 22:58:43 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/* Get the LOCKMODE for a given MultiXactStatus */
#define LOCKMODE_from_mxstatus(status) \
(tupleLockExtraInfo[TUPLOCK_from_mxstatus((status))].hwlock)
/*
* Acquire heavyweight locks on tuples, using a LockTupleMode strength value.
* This is more readable than having every caller translate it to lock.h's
* LOCKMODE.
*/
#define LockTupleTuplock(rel, tup, mode) \
LockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
#define UnlockTupleTuplock(rel, tup, mode) \
UnlockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
#define ConditionalLockTupleTuplock(rel, tup, mode) \
ConditionalLockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
#ifdef USE_PREFETCH
/*
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
* heap_index_delete_tuples and index_delete_prefetch_buffer use this
* structure to coordinate prefetching activity
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
*/
typedef struct
{
BlockNumber cur_hblkno;
int next_item;
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
int ndeltids;
TM_IndexDelete *deltids;
} IndexDeletePrefetchState;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
#endif
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
/* heap_index_delete_tuples bottom-up index deletion costing constants */
#define BOTTOMUP_MAX_NBLOCKS 6
#define BOTTOMUP_TOLERANCE_NBLOCKS 3
/*
* heap_index_delete_tuples uses this when determining which heap blocks it
* must visit to help its bottom-up index deletion caller
*/
typedef struct IndexDeleteCounts
{
int16 npromisingtids; /* Number of "promising" TIDs in group */
int16 ntids; /* Number of TIDs in group */
int16 ifirsttid; /* Offset to group's first deltid */
} IndexDeleteCounts;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* This table maps tuple lock strength values for each particular
* MultiXactStatus value.
*/
static const int MultiXactStatusLock[MaxMultiXactStatus + 1] =
{
LockTupleKeyShare, /* ForKeyShare */
LockTupleShare, /* ForShare */
LockTupleNoKeyExclusive, /* ForNoKeyUpdate */
LockTupleExclusive, /* ForUpdate */
LockTupleNoKeyExclusive, /* NoKeyUpdate */
LockTupleExclusive /* Update */
};
/* Get the LockTupleMode for a given MultiXactStatus */
#define TUPLOCK_from_mxstatus(status) \
(MultiXactStatusLock[(status)])
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/* ----------------------------------------------------------------
* heap support routines
* ----------------------------------------------------------------
*/
/* ----------------
heap_fetch requires buffer pointer, must be released; heap_getnext no longer returns buffer pointer, can be gotten from scan; descriptor; bootstrap can create multi-key indexes; pg_procname index now is multi-key index; oidint2, oidint4, oidname are gone (must be removed from regression tests); use System Cache rather than sequential scan in many places; heap_modifytuple no longer takes buffer parameter; remove unused buffer parameter in a few other functions; oid8 is not index-able; remove some use of single-character variable names; cleanup Buffer variables usage and scan descriptor looping; cleaned up allocation and freeing of tuples; 18k lines of diff;
1998-08-19 04:04:17 +02:00
* initscan - scan code common to heap_beginscan and heap_rescan
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
* ----------------
*/
static void
Redesign tablesample method API, and do extensive code review. The original implementation of TABLESAMPLE modeled the tablesample method API on index access methods, which wasn't a good choice because, without specialized DDL commands, there's no way to build an extension that can implement a TSM. (Raw inserts into system catalogs are not an acceptable thing to do, because we can't undo them during DROP EXTENSION, nor will pg_upgrade behave sanely.) Instead adopt an API more like procedural language handlers or foreign data wrappers, wherein the only SQL-level support object needed is a single handler function identified by having a special return type. This lets us get rid of the supporting catalog altogether, so that no custom DDL support is needed for the feature. Adjust the API so that it can support non-constant tablesample arguments (the original coding assumed we could evaluate the argument expressions at ExecInitSampleScan time, which is undesirable even if it weren't outright unsafe), and discourage sampling methods from looking at invisible tuples. Make sure that the BERNOULLI and SYSTEM methods are genuinely repeatable within and across queries, as required by the SQL standard, and deal more honestly with methods that can't support that requirement. Make a full code-review pass over the tablesample additions, and fix assorted bugs, omissions, infelicities, and cosmetic issues (such as failure to put the added code stanzas in a consistent ordering). Improve EXPLAIN's output of tablesample plans, too. Back-patch to 9.5 so that we don't have to support the original API in production.
2015-07-25 20:39:00 +02:00
initscan(HeapScanDesc scan, ScanKey key, bool keep_startblock)
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
{
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
ParallelBlockTableScanDesc bpscan = NULL;
Fix CREATE INDEX CONCURRENTLY so that it won't use synchronized scan for its second pass over the table. It has to start at block zero, else the "merge join" logic for detecting which TIDs are already in the index doesn't work. Hence, extend heapam.c's API so that callers can enable or disable syncscan. (I put in an option to disable buffer access strategy, too, just in case somebody needs it.) Per report from Hannes Dorbath.
2008-01-14 02:39:09 +01:00
bool allow_strat;
bool allow_sync;
Rearrange heap_beginscan to eliminate extra call on mdnblocks. Bad enough it does an lseek, but to do it twice for no reason...
1999-10-31 01:10:21 +02:00
/*
Get rid of rd_nblocks field in relcache entries. Turns out this was costing us lots more to maintain than it was worth. On shared tables it was of exactly zero benefit because we couldn't trust it to be up to date. On temp tables it sometimes saved an lseek, but not often enough to be worth getting excited about. And the real problem was that we forced an lseek on every relcache flush in order to update the field. So all in all it seems best to lose the complexity.
2004-05-08 21:09:25 +02:00
* Determine the number of blocks we have to scan.
*
Rearrange heap_beginscan to eliminate extra call on mdnblocks. Bad enough it does an lseek, but to do it twice for no reason...
1999-10-31 01:10:21 +02:00
* It is sufficient to do this once at scan start, since any tuples added
Arrange for large sequential scans to synchronize with each other, so that when multiple backends are scanning the same relation concurrently, each page is (ideally) read only once. Jeff Davis, with review by Heikki and Tom.
2007-06-08 20:23:53 +02:00
* while the scan is in progress will be invisible to my snapshot anyway.
* (That is not true when using a non-MVCC snapshot. However, we couldn't
* guarantee to return tuples added after scan start anyway, since they
* might go into pages we already scanned. To guarantee consistent
* results for a non-MVCC snapshot, the caller must hold some higher-level
* lock that ensures the interesting tuple(s) won't change.)
Rearrange heap_beginscan to eliminate extra call on mdnblocks. Bad enough it does an lseek, but to do it twice for no reason...
1999-10-31 01:10:21 +02:00
*/
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
if (scan->rs_base.rs_parallel != NULL)
{
bpscan = (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel;
scan->rs_nblocks = bpscan->phs_nblocks;
}
Add a C API for parallel heap scans. Using this API, one backend can set up a ParallelHeapScanDesc to which multiple backends can then attach. Each tuple in the relation will be returned to exactly one of the scanning backends. Only forward scans are supported, and rescans must be carefully coordinated. This is not exposed to the planner or executor yet. The original version of this code was written by me. Amit Kapila reviewed it, tested it, and improved it, including adding support for synchronized scans, per review comments from Jeff Davis. Extensive testing of this and related patches was performed by Haribabu Kommi. Final cleanup of this patch by me.
2015-10-16 23:25:02 +02:00
else
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
scan->rs_nblocks = RelationGetNumberOfBlocks(scan->rs_base.rs_rd);
Rearrange heap_beginscan to eliminate extra call on mdnblocks. Bad enough it does an lseek, but to do it twice for no reason...
1999-10-31 01:10:21 +02:00
Make large sequential scans and VACUUMs work in a limited-size "ring" of buffers, rather than blowing out the whole shared-buffer arena. Aside from avoiding cache spoliation, this fixes the problem that VACUUM formerly tended to cause a WAL flush for every page it modified, because we had it hacked to use only a single buffer. Those flushes will now occur only once per ring-ful. The exact ring size, and the threshold for seqscans to switch into the ring usage pattern, remain under debate; but the infrastructure seems done. The key bit of infrastructure is a new optional BufferAccessStrategy object that can be passed to ReadBuffer operations; this replaces the former StrategyHintVacuum API. This patch also changes the buffer usage-count methodology a bit: we now advance usage_count when first pinning a buffer, rather than when last unpinning it. To preserve the behavior that a buffer's lifetime starts to decrease when it's released, the clock sweep code is modified to not decrement usage_count of pinned buffers. Work not done in this commit: teach GiST and GIN indexes to use the vacuum BufferAccessStrategy for vacuum-driven fetches. Original patch by Simon, reworked by Heikki and again by Tom.
2007-05-30 22:12:03 +02:00
/*
* If the table is large relative to NBuffers, use a bulk-read access
Arrange for large sequential scans to synchronize with each other, so that when multiple backends are scanning the same relation concurrently, each page is (ideally) read only once. Jeff Davis, with review by Heikki and Tom.
2007-06-08 20:23:53 +02:00
* strategy and enable synchronized scanning (see syncscan.c). Although
* the thresholds for these features could be different, we make them the
* same so that there are only two behaviors to tune rather than four.
Fix CREATE INDEX CONCURRENTLY so that it won't use synchronized scan for its second pass over the table. It has to start at block zero, else the "merge join" logic for detecting which TIDs are already in the index doesn't work. Hence, extend heapam.c's API so that callers can enable or disable syncscan. (I put in an option to disable buffer access strategy, too, just in case somebody needs it.) Per report from Hannes Dorbath.
2008-01-14 02:39:09 +01:00
* (However, some callers need to be able to disable one or both of these
Add a GUC variable "synchronize_seqscans" to allow clients to disable the new synchronized-scanning behavior, and make pg_dump disable sync scans so that it will reliably preserve row ordering. Per recent discussions.
2008-01-30 19:35:55 +01:00
* behaviors, independently of the size of the table; also there is a GUC
* variable that can disable synchronized scanning.)
Arrange for large sequential scans to synchronize with each other, so that when multiple backends are scanning the same relation concurrently, each page is (ideally) read only once. Jeff Davis, with review by Heikki and Tom.
2007-06-08 20:23:53 +02:00
*
Fix assorted inconsistencies. There were a number of issues in the recent commits which include typos, code and comments mismatch, leftover function declarations. Fix them. Reported-by: Alexander Lakhin Author: Alexander Lakhin, Amit Kapila and Amit Langote Reviewed-by: Amit Kapila Discussion: https://postgr.es/m/ef0c0232-0c1d-3a35-63d4-0ebd06e31387@gmail.com
2019-06-08 04:46:38 +02:00
* Note that table_block_parallelscan_initialize has a very similar test;
* if you change this, consider changing that one, too.
Make large sequential scans and VACUUMs work in a limited-size "ring" of buffers, rather than blowing out the whole shared-buffer arena. Aside from avoiding cache spoliation, this fixes the problem that VACUUM formerly tended to cause a WAL flush for every page it modified, because we had it hacked to use only a single buffer. Those flushes will now occur only once per ring-ful. The exact ring size, and the threshold for seqscans to switch into the ring usage pattern, remain under debate; but the infrastructure seems done. The key bit of infrastructure is a new optional BufferAccessStrategy object that can be passed to ReadBuffer operations; this replaces the former StrategyHintVacuum API. This patch also changes the buffer usage-count methodology a bit: we now advance usage_count when first pinning a buffer, rather than when last unpinning it. To preserve the behavior that a buffer's lifetime starts to decrease when it's released, the clock sweep code is modified to not decrement usage_count of pinned buffers. Work not done in this commit: teach GiST and GIN indexes to use the vacuum BufferAccessStrategy for vacuum-driven fetches. Original patch by Simon, reworked by Heikki and again by Tom.
2007-05-30 22:12:03 +02:00
*/
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
if (!RelationUsesLocalBuffers(scan->rs_base.rs_rd) &&
Teach heapam code to know the difference between a real seqscan and the pseudo HeapScanDesc created for a bitmap heap scan. This avoids some useless overhead during a bitmap scan startup, in particular invoking the syncscan code. (We might someday want to do that, but right now it's merely useless contention for shared memory, to say nothing of possibly pushing useful entries out of syncscan's small LRU list.) This also allows elimination of ugly pgstat_discount_heap_scan() kluge.
2007-06-09 20:49:55 +02:00
scan->rs_nblocks > NBuffers / 4)
Fix CREATE INDEX CONCURRENTLY so that it won't use synchronized scan for its second pass over the table. It has to start at block zero, else the "merge join" logic for detecting which TIDs are already in the index doesn't work. Hence, extend heapam.c's API so that callers can enable or disable syncscan. (I put in an option to disable buffer access strategy, too, just in case somebody needs it.) Per report from Hannes Dorbath.
2008-01-14 02:39:09 +01:00
{
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
allow_strat = (scan->rs_base.rs_flags & SO_ALLOW_STRAT) != 0;
allow_sync = (scan->rs_base.rs_flags & SO_ALLOW_SYNC) != 0;
Fix CREATE INDEX CONCURRENTLY so that it won't use synchronized scan for its second pass over the table. It has to start at block zero, else the "merge join" logic for detecting which TIDs are already in the index doesn't work. Hence, extend heapam.c's API so that callers can enable or disable syncscan. (I put in an option to disable buffer access strategy, too, just in case somebody needs it.) Per report from Hannes Dorbath.
2008-01-14 02:39:09 +01:00
}
else
allow_strat = allow_sync = false;
if (allow_strat)
Make large sequential scans and VACUUMs work in a limited-size "ring" of buffers, rather than blowing out the whole shared-buffer arena. Aside from avoiding cache spoliation, this fixes the problem that VACUUM formerly tended to cause a WAL flush for every page it modified, because we had it hacked to use only a single buffer. Those flushes will now occur only once per ring-ful. The exact ring size, and the threshold for seqscans to switch into the ring usage pattern, remain under debate; but the infrastructure seems done. The key bit of infrastructure is a new optional BufferAccessStrategy object that can be passed to ReadBuffer operations; this replaces the former StrategyHintVacuum API. This patch also changes the buffer usage-count methodology a bit: we now advance usage_count when first pinning a buffer, rather than when last unpinning it. To preserve the behavior that a buffer's lifetime starts to decrease when it's released, the clock sweep code is modified to not decrement usage_count of pinned buffers. Work not done in this commit: teach GiST and GIN indexes to use the vacuum BufferAccessStrategy for vacuum-driven fetches. Original patch by Simon, reworked by Heikki and again by Tom.
2007-05-30 22:12:03 +02:00
{
Add a C API for parallel heap scans. Using this API, one backend can set up a ParallelHeapScanDesc to which multiple backends can then attach. Each tuple in the relation will be returned to exactly one of the scanning backends. Only forward scans are supported, and rescans must be carefully coordinated. This is not exposed to the planner or executor yet. The original version of this code was written by me. Amit Kapila reviewed it, tested it, and improved it, including adding support for synchronized scans, per review comments from Jeff Davis. Extensive testing of this and related patches was performed by Haribabu Kommi. Final cleanup of this patch by me.
2015-10-16 23:25:02 +02:00
/* During a rescan, keep the previous strategy object. */
Make large sequential scans and VACUUMs work in a limited-size "ring" of buffers, rather than blowing out the whole shared-buffer arena. Aside from avoiding cache spoliation, this fixes the problem that VACUUM formerly tended to cause a WAL flush for every page it modified, because we had it hacked to use only a single buffer. Those flushes will now occur only once per ring-ful. The exact ring size, and the threshold for seqscans to switch into the ring usage pattern, remain under debate; but the infrastructure seems done. The key bit of infrastructure is a new optional BufferAccessStrategy object that can be passed to ReadBuffer operations; this replaces the former StrategyHintVacuum API. This patch also changes the buffer usage-count methodology a bit: we now advance usage_count when first pinning a buffer, rather than when last unpinning it. To preserve the behavior that a buffer's lifetime starts to decrease when it's released, the clock sweep code is modified to not decrement usage_count of pinned buffers. Work not done in this commit: teach GiST and GIN indexes to use the vacuum BufferAccessStrategy for vacuum-driven fetches. Original patch by Simon, reworked by Heikki and again by Tom.
2007-05-30 22:12:03 +02:00
if (scan->rs_strategy == NULL)
scan->rs_strategy = GetAccessStrategy(BAS_BULKREAD);
}
else
{
if (scan->rs_strategy != NULL)
FreeAccessStrategy(scan->rs_strategy);
scan->rs_strategy = NULL;
Fix CREATE INDEX CONCURRENTLY so that it won't use synchronized scan for its second pass over the table. It has to start at block zero, else the "merge join" logic for detecting which TIDs are already in the index doesn't work. Hence, extend heapam.c's API so that callers can enable or disable syncscan. (I put in an option to disable buffer access strategy, too, just in case somebody needs it.) Per report from Hannes Dorbath.
2008-01-14 02:39:09 +01:00
}
Arrange for large sequential scans to synchronize with each other, so that when multiple backends are scanning the same relation concurrently, each page is (ideally) read only once. Jeff Davis, with review by Heikki and Tom.
2007-06-08 20:23:53 +02:00
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
if (scan->rs_base.rs_parallel != NULL)
Add a C API for parallel heap scans. Using this API, one backend can set up a ParallelHeapScanDesc to which multiple backends can then attach. Each tuple in the relation will be returned to exactly one of the scanning backends. Only forward scans are supported, and rescans must be carefully coordinated. This is not exposed to the planner or executor yet. The original version of this code was written by me. Amit Kapila reviewed it, tested it, and improved it, including adding support for synchronized scans, per review comments from Jeff Davis. Extensive testing of this and related patches was performed by Haribabu Kommi. Final cleanup of this patch by me.
2015-10-16 23:25:02 +02:00
{
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
/* For parallel scan, believe whatever ParallelTableScanDesc says. */
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
if (scan->rs_base.rs_parallel->phs_syncscan)
scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
else
scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
Add a C API for parallel heap scans. Using this API, one backend can set up a ParallelHeapScanDesc to which multiple backends can then attach. Each tuple in the relation will be returned to exactly one of the scanning backends. Only forward scans are supported, and rescans must be carefully coordinated. This is not exposed to the planner or executor yet. The original version of this code was written by me. Amit Kapila reviewed it, tested it, and improved it, including adding support for synchronized scans, per review comments from Jeff Davis. Extensive testing of this and related patches was performed by Haribabu Kommi. Final cleanup of this patch by me.
2015-10-16 23:25:02 +02:00
}
else if (keep_startblock)
Keep rs_startblock the same during heap_rescan, so that a rescan of a SeqScan node starts from the same place as the first scan did. This avoids surprising behavior of scrollable and WITH HOLD cursors, as seen in Mark Kirkwood's bug report of yesterday. It's not entirely clear whether a rescan should be forced to drop out of the syncscan mode, but for the moment I left the code behaving the same on that point. Any change there would only be a performance and not a correctness issue, anyway. Back-patch to 8.3, since the unstable behavior was created by the syncscan patch.
2009-06-10 20:54:16 +02:00
{
/*
Redesign tablesample method API, and do extensive code review. The original implementation of TABLESAMPLE modeled the tablesample method API on index access methods, which wasn't a good choice because, without specialized DDL commands, there's no way to build an extension that can implement a TSM. (Raw inserts into system catalogs are not an acceptable thing to do, because we can't undo them during DROP EXTENSION, nor will pg_upgrade behave sanely.) Instead adopt an API more like procedural language handlers or foreign data wrappers, wherein the only SQL-level support object needed is a single handler function identified by having a special return type. This lets us get rid of the supporting catalog altogether, so that no custom DDL support is needed for the feature. Adjust the API so that it can support non-constant tablesample arguments (the original coding assumed we could evaluate the argument expressions at ExecInitSampleScan time, which is undesirable even if it weren't outright unsafe), and discourage sampling methods from looking at invisible tuples. Make sure that the BERNOULLI and SYSTEM methods are genuinely repeatable within and across queries, as required by the SQL standard, and deal more honestly with methods that can't support that requirement. Make a full code-review pass over the tablesample additions, and fix assorted bugs, omissions, infelicities, and cosmetic issues (such as failure to put the added code stanzas in a consistent ordering). Improve EXPLAIN's output of tablesample plans, too. Back-patch to 9.5 so that we don't have to support the original API in production.
2015-07-25 20:39:00 +02:00
* When rescanning, we want to keep the previous startblock setting,
* so that rewinding a cursor doesn't generate surprising results.
* Reset the active syncscan setting, though.
Keep rs_startblock the same during heap_rescan, so that a rescan of a SeqScan node starts from the same place as the first scan did. This avoids surprising behavior of scrollable and WITH HOLD cursors, as seen in Mark Kirkwood's bug report of yesterday. It's not entirely clear whether a rescan should be forced to drop out of the syncscan mode, but for the moment I left the code behaving the same on that point. Any change there would only be a performance and not a correctness issue, anyway. Back-patch to 8.3, since the unstable behavior was created by the syncscan patch.
2009-06-10 20:54:16 +02:00
*/
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
if (allow_sync && synchronize_seqscans)
scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
else
scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
Keep rs_startblock the same during heap_rescan, so that a rescan of a SeqScan node starts from the same place as the first scan did. This avoids surprising behavior of scrollable and WITH HOLD cursors, as seen in Mark Kirkwood's bug report of yesterday. It's not entirely clear whether a rescan should be forced to drop out of the syncscan mode, but for the moment I left the code behaving the same on that point. Any change there would only be a performance and not a correctness issue, anyway. Back-patch to 8.3, since the unstable behavior was created by the syncscan patch.
2009-06-10 20:54:16 +02:00
}
else if (allow_sync && synchronize_seqscans)
Fix CREATE INDEX CONCURRENTLY so that it won't use synchronized scan for its second pass over the table. It has to start at block zero, else the "merge join" logic for detecting which TIDs are already in the index doesn't work. Hence, extend heapam.c's API so that callers can enable or disable syncscan. (I put in an option to disable buffer access strategy, too, just in case somebody needs it.) Per report from Hannes Dorbath.
2008-01-14 02:39:09 +01:00
{
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
scan->rs_startblock = ss_get_location(scan->rs_base.rs_rd, scan->rs_nblocks);
Fix CREATE INDEX CONCURRENTLY so that it won't use synchronized scan for its second pass over the table. It has to start at block zero, else the "merge join" logic for detecting which TIDs are already in the index doesn't work. Hence, extend heapam.c's API so that callers can enable or disable syncscan. (I put in an option to disable buffer access strategy, too, just in case somebody needs it.) Per report from Hannes Dorbath.
2008-01-14 02:39:09 +01:00
}
else
{
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
Arrange for large sequential scans to synchronize with each other, so that when multiple backends are scanning the same relation concurrently, each page is (ideally) read only once. Jeff Davis, with review by Heikki and Tom.
2007-06-08 20:23:53 +02:00
scan->rs_startblock = 0;
Make large sequential scans and VACUUMs work in a limited-size "ring" of buffers, rather than blowing out the whole shared-buffer arena. Aside from avoiding cache spoliation, this fixes the problem that VACUUM formerly tended to cause a WAL flush for every page it modified, because we had it hacked to use only a single buffer. Those flushes will now occur only once per ring-ful. The exact ring size, and the threshold for seqscans to switch into the ring usage pattern, remain under debate; but the infrastructure seems done. The key bit of infrastructure is a new optional BufferAccessStrategy object that can be passed to ReadBuffer operations; this replaces the former StrategyHintVacuum API. This patch also changes the buffer usage-count methodology a bit: we now advance usage_count when first pinning a buffer, rather than when last unpinning it. To preserve the behavior that a buffer's lifetime starts to decrease when it's released, the clock sweep code is modified to not decrement usage_count of pinned buffers. Work not done in this commit: teach GiST and GIN indexes to use the vacuum BufferAccessStrategy for vacuum-driven fetches. Original patch by Simon, reworked by Heikki and again by Tom.
2007-05-30 22:12:03 +02:00
}
BRIN: Block Range Indexes BRIN is a new index access method intended to accelerate scans of very large tables, without the maintenance overhead of btrees or other traditional indexes. They work by maintaining "summary" data about block ranges. Bitmap index scans work by reading each summary tuple and comparing them with the query quals; all pages in the range are returned in a lossy TID bitmap if the quals are consistent with the values in the summary tuple, otherwise not. Normal index scans are not supported because these indexes do not store TIDs. As new tuples are added into the index, the summary information is updated (if the block range in which the tuple is added is already summarized) or not; in the latter case, a subsequent pass of VACUUM or the brin_summarize_new_values() function will create the summary information. For data types with natural 1-D sort orders, the summary info consists of the maximum and the minimum values of each indexed column within each page range. This type of operator class we call "Minmax", and we supply a bunch of them for most data types with B-tree opclasses. Since the BRIN code is generalized, other approaches are possible for things such as arrays, geometric types, ranges, etc; even for things such as enum types we could do something different than minmax with better results. In this commit I only include minmax. Catalog version bumped due to new builtin catalog entries. There's more that could be done here, but this is a good step forwards. Loosely based on ideas from Simon Riggs; code mostly by Álvaro Herrera, with contribution by Heikki Linnakangas. Patch reviewed by: Amit Kapila, Heikki Linnakangas, Robert Haas. Testing help from Jeff Janes, Erik Rijkers, Emanuel Calvo. PS: The research leading to these results has received funding from the European Union's Seventh Framework Programme (FP7/2007-2013) under grant agreement n° 318633.
2014-11-07 20:38:14 +01:00
scan->rs_numblocks = InvalidBlockNumber;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
scan->rs_inited = false;
Remove RelationGetBufferWithBuffer(), which is horribly confused about appropriate pin-count manipulation, and instead use ReleaseAndReadBuffer. Make use of the fact that the passed-in buffer (if there is one) must be pinned to avoid grabbing the bufmgr spinlock when we are able to return this same buffer. Eliminate unnecessary 'previous tuple' and 'next tuple' fields of HeapScanDesc and IndexScanDesc, thereby removing a whole lot of bookkeeping from heap_getnext() and related routines.
2001-06-09 20:16:59 +02:00
scan->rs_ctup.t_data = NULL;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
ItemPointerSetInvalid(&scan->rs_ctup.t_self);
Remove RelationGetBufferWithBuffer(), which is horribly confused about appropriate pin-count manipulation, and instead use ReleaseAndReadBuffer. Make use of the fact that the passed-in buffer (if there is one) must be pinned to avoid grabbing the bufmgr spinlock when we are able to return this same buffer. Eliminate unnecessary 'previous tuple' and 'next tuple' fields of HeapScanDesc and IndexScanDesc, thereby removing a whole lot of bookkeeping from heap_getnext() and related routines.
2001-06-09 20:16:59 +02:00
scan->rs_cbuf = InvalidBuffer;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
scan->rs_cblock = InvalidBlockNumber;
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
/* page-at-a-time fields are always invalid when not rs_inited */
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
* copy the scan key, if appropriate
*/
Clean up assorted failures under clang's -fsanitize=undefined checks. Most of these are cases where we could call memcpy() or other libc functions with a NULL pointer and a zero count, which is forbidden by POSIX even though every production version of libc allows it. We've fixed such things before in a piecemeal way, but apparently never made an effort to try to get them all. I don't claim that this patch does so either, but it gets every failure I observe in check-world, using clang 12.0.1 on current RHEL8. numeric.c has a different issue that the sanitizer doesn't like: "ln(-1.0)" will compute log10(0) and then try to assign the resulting -Inf to an integer variable. We don't actually use the result in such a case, so there's no live bug. Back-patch to all supported branches, with the idea that we might start running a buildfarm member that tests this case. This includes back-patching c1132aae3 (Check the size in COPY_POINTER_FIELD), which previously silenced some of these issues in copyfuncs.c. Discussion: https://postgr.es/m/CALNJ-vT9r0DSsAOw9OXVJFxLENoVS_68kJ5x0p44atoYH+H4dg@mail.gmail.com
2022-03-04 00:13:24 +01:00
if (key != NULL && scan->rs_base.rs_nkeys > 0)
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
memcpy(scan->rs_base.rs_key, key, scan->rs_base.rs_nkeys * sizeof(ScanKeyData));
Revise pgstats stuff to fix the problems with not counting accesses generated by bitmap index scans. Along the way, simplify and speed up the code for counting sequential and index scans; it was both confusing and inefficient to be taking care of that in the per-tuple loops, IMHO. initdb forced because of internal changes in pg_stat view definitions.
2005-10-06 04:29:23 +02:00
Teach heapam code to know the difference between a real seqscan and the pseudo HeapScanDesc created for a bitmap heap scan. This avoids some useless overhead during a bitmap scan startup, in particular invoking the syncscan code. (We might someday want to do that, but right now it's merely useless contention for shared memory, to say nothing of possibly pushing useful entries out of syncscan's small LRU list.) This also allows elimination of ugly pgstat_discount_heap_scan() kluge.
2007-06-09 20:49:55 +02:00
/*
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
* Currently, we only have a stats counter for sequential heap scans (but
* e.g for bitmap scans the underlying bitmap index scans will be counted,
* and for sample scans we update stats for tuple fetches).
Teach heapam code to know the difference between a real seqscan and the pseudo HeapScanDesc created for a bitmap heap scan. This avoids some useless overhead during a bitmap scan startup, in particular invoking the syncscan code. (We might someday want to do that, but right now it's merely useless contention for shared memory, to say nothing of possibly pushing useful entries out of syncscan's small LRU list.) This also allows elimination of ugly pgstat_discount_heap_scan() kluge.
2007-06-09 20:49:55 +02:00
*/
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN)
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
pgstat_count_heap_scan(scan->rs_base.rs_rd);
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
}
Fix some oversights in BRIN patch. Remove HeapScanDescData.rs_initblock, which wasn't being used for anything in the final version of the patch. Fix IndexBuildHeapScan so that it supports syncscan again; the patch broke synchronous scanning for index builds by forcing rs_startblk to zero even when the caller did not care about that and had asked for syncscan. Add some commentary and usage defenses to heap_setscanlimits(). Fix heapam so that asking for rs_numblocks == 0 does what you would reasonably expect. As coded it amounted to requesting a whole-table scan, because those "--x <= 0" tests on an unsigned variable would behave surprisingly.
2015-07-21 19:38:24 +02:00
/*
* heap_setscanlimits - restrict range of a heapscan
*
* startBlk is the page to start at
* numBlks is number of pages to scan (InvalidBlockNumber means "all")
*/
BRIN: Block Range Indexes BRIN is a new index access method intended to accelerate scans of very large tables, without the maintenance overhead of btrees or other traditional indexes. They work by maintaining "summary" data about block ranges. Bitmap index scans work by reading each summary tuple and comparing them with the query quals; all pages in the range are returned in a lossy TID bitmap if the quals are consistent with the values in the summary tuple, otherwise not. Normal index scans are not supported because these indexes do not store TIDs. As new tuples are added into the index, the summary information is updated (if the block range in which the tuple is added is already summarized) or not; in the latter case, a subsequent pass of VACUUM or the brin_summarize_new_values() function will create the summary information. For data types with natural 1-D sort orders, the summary info consists of the maximum and the minimum values of each indexed column within each page range. This type of operator class we call "Minmax", and we supply a bunch of them for most data types with B-tree opclasses. Since the BRIN code is generalized, other approaches are possible for things such as arrays, geometric types, ranges, etc; even for things such as enum types we could do something different than minmax with better results. In this commit I only include minmax. Catalog version bumped due to new builtin catalog entries. There's more that could be done here, but this is a good step forwards. Loosely based on ideas from Simon Riggs; code mostly by Álvaro Herrera, with contribution by Heikki Linnakangas. Patch reviewed by: Amit Kapila, Heikki Linnakangas, Robert Haas. Testing help from Jeff Janes, Erik Rijkers, Emanuel Calvo. PS: The research leading to these results has received funding from the European Union's Seventh Framework Programme (FP7/2007-2013) under grant agreement n° 318633.
2014-11-07 20:38:14 +01:00
void
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
heap_setscanlimits(TableScanDesc sscan, BlockNumber startBlk, BlockNumber numBlks)
BRIN: Block Range Indexes BRIN is a new index access method intended to accelerate scans of very large tables, without the maintenance overhead of btrees or other traditional indexes. They work by maintaining "summary" data about block ranges. Bitmap index scans work by reading each summary tuple and comparing them with the query quals; all pages in the range are returned in a lossy TID bitmap if the quals are consistent with the values in the summary tuple, otherwise not. Normal index scans are not supported because these indexes do not store TIDs. As new tuples are added into the index, the summary information is updated (if the block range in which the tuple is added is already summarized) or not; in the latter case, a subsequent pass of VACUUM or the brin_summarize_new_values() function will create the summary information. For data types with natural 1-D sort orders, the summary info consists of the maximum and the minimum values of each indexed column within each page range. This type of operator class we call "Minmax", and we supply a bunch of them for most data types with B-tree opclasses. Since the BRIN code is generalized, other approaches are possible for things such as arrays, geometric types, ranges, etc; even for things such as enum types we could do something different than minmax with better results. In this commit I only include minmax. Catalog version bumped due to new builtin catalog entries. There's more that could be done here, but this is a good step forwards. Loosely based on ideas from Simon Riggs; code mostly by Álvaro Herrera, with contribution by Heikki Linnakangas. Patch reviewed by: Amit Kapila, Heikki Linnakangas, Robert Haas. Testing help from Jeff Janes, Erik Rijkers, Emanuel Calvo. PS: The research leading to these results has received funding from the European Union's Seventh Framework Programme (FP7/2007-2013) under grant agreement n° 318633.
2014-11-07 20:38:14 +01:00
{
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
HeapScanDesc scan = (HeapScanDesc) sscan;
Fix some oversights in BRIN patch. Remove HeapScanDescData.rs_initblock, which wasn't being used for anything in the final version of the patch. Fix IndexBuildHeapScan so that it supports syncscan again; the patch broke synchronous scanning for index builds by forcing rs_startblk to zero even when the caller did not care about that and had asked for syncscan. Add some commentary and usage defenses to heap_setscanlimits(). Fix heapam so that asking for rs_numblocks == 0 does what you would reasonably expect. As coded it amounted to requesting a whole-table scan, because those "--x <= 0" tests on an unsigned variable would behave surprisingly.
2015-07-21 19:38:24 +02:00
Assert(!scan->rs_inited); /* else too late to change */
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
/* else rs_startblock is significant */
Assert(!(scan->rs_base.rs_flags & SO_ALLOW_SYNC));
Fix some oversights in BRIN patch. Remove HeapScanDescData.rs_initblock, which wasn't being used for anything in the final version of the patch. Fix IndexBuildHeapScan so that it supports syncscan again; the patch broke synchronous scanning for index builds by forcing rs_startblk to zero even when the caller did not care about that and had asked for syncscan. Add some commentary and usage defenses to heap_setscanlimits(). Fix heapam so that asking for rs_numblocks == 0 does what you would reasonably expect. As coded it amounted to requesting a whole-table scan, because those "--x <= 0" tests on an unsigned variable would behave surprisingly.
2015-07-21 19:38:24 +02:00
/* Check startBlk is valid (but allow case of zero blocks...) */
Assert(startBlk == 0 || startBlk < scan->rs_nblocks);
BRIN: Block Range Indexes BRIN is a new index access method intended to accelerate scans of very large tables, without the maintenance overhead of btrees or other traditional indexes. They work by maintaining "summary" data about block ranges. Bitmap index scans work by reading each summary tuple and comparing them with the query quals; all pages in the range are returned in a lossy TID bitmap if the quals are consistent with the values in the summary tuple, otherwise not. Normal index scans are not supported because these indexes do not store TIDs. As new tuples are added into the index, the summary information is updated (if the block range in which the tuple is added is already summarized) or not; in the latter case, a subsequent pass of VACUUM or the brin_summarize_new_values() function will create the summary information. For data types with natural 1-D sort orders, the summary info consists of the maximum and the minimum values of each indexed column within each page range. This type of operator class we call "Minmax", and we supply a bunch of them for most data types with B-tree opclasses. Since the BRIN code is generalized, other approaches are possible for things such as arrays, geometric types, ranges, etc; even for things such as enum types we could do something different than minmax with better results. In this commit I only include minmax. Catalog version bumped due to new builtin catalog entries. There's more that could be done here, but this is a good step forwards. Loosely based on ideas from Simon Riggs; code mostly by Álvaro Herrera, with contribution by Heikki Linnakangas. Patch reviewed by: Amit Kapila, Heikki Linnakangas, Robert Haas. Testing help from Jeff Janes, Erik Rijkers, Emanuel Calvo. PS: The research leading to these results has received funding from the European Union's Seventh Framework Programme (FP7/2007-2013) under grant agreement n° 318633.
2014-11-07 20:38:14 +01:00
scan->rs_startblock = startBlk;
scan->rs_numblocks = numBlks;
}
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
/*
* heapgetpage - subroutine for heapgettup()
heap_fetch requires buffer pointer, must be released; heap_getnext no longer returns buffer pointer, can be gotten from scan; descriptor; bootstrap can create multi-key indexes; pg_procname index now is multi-key index; oidint2, oidint4, oidname are gone (must be removed from regression tests); use System Cache rather than sequential scan in many places; heap_modifytuple no longer takes buffer parameter; remove unused buffer parameter in a few other functions; oid8 is not index-able; remove some use of single-character variable names; cleanup Buffer variables usage and scan descriptor looping; cleaned up allocation and freeing of tuples; 18k lines of diff;
1998-08-19 04:04:17 +02:00
*
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
* This routine reads and pins the specified page of the relation.
* In page-at-a-time mode it performs additional work, namely determining
* which tuples on the page are visible.
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*/
TABLESAMPLE, SQL Standard and extensible Add a TABLESAMPLE clause to SELECT statements that allows user to specify random BERNOULLI sampling or block level SYSTEM sampling. Implementation allows for extensible sampling functions to be written, using a standard API. Basic version follows SQLStandard exactly. Usable concrete use cases for the sampling API follow in later commits. Petr Jelinek Reviewed by Michael Paquier and Simon Riggs
2015-05-15 20:37:10 +02:00
void
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
heapgetpage(TableScanDesc sscan, BlockNumber block)
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
{
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
HeapScanDesc scan = (HeapScanDesc) sscan;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
Buffer buffer;
Snapshot snapshot;
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
Page page;
New HeapTuple structure/interface.
1998-11-27 20:52:36 +01:00
int lines;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
int ntup;
New HeapTuple structure/interface.
1998-11-27 20:52:36 +01:00
OffsetNumber lineoff;
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
bool all_visible;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
Assert(block < scan->rs_nblocks);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Make large sequential scans and VACUUMs work in a limited-size "ring" of buffers, rather than blowing out the whole shared-buffer arena. Aside from avoiding cache spoliation, this fixes the problem that VACUUM formerly tended to cause a WAL flush for every page it modified, because we had it hacked to use only a single buffer. Those flushes will now occur only once per ring-ful. The exact ring size, and the threshold for seqscans to switch into the ring usage pattern, remain under debate; but the infrastructure seems done. The key bit of infrastructure is a new optional BufferAccessStrategy object that can be passed to ReadBuffer operations; this replaces the former StrategyHintVacuum API. This patch also changes the buffer usage-count methodology a bit: we now advance usage_count when first pinning a buffer, rather than when last unpinning it. To preserve the behavior that a buffer's lifetime starts to decrease when it's released, the clock sweep code is modified to not decrement usage_count of pinned buffers. Work not done in this commit: teach GiST and GIN indexes to use the vacuum BufferAccessStrategy for vacuum-driven fetches. Original patch by Simon, reworked by Heikki and again by Tom.
2007-05-30 22:12:03 +02:00
/* release previous scan buffer, if any */
if (BufferIsValid(scan->rs_cbuf))
{
ReleaseBuffer(scan->rs_cbuf);
scan->rs_cbuf = InvalidBuffer;
}
Ensure that seqscans check for interrupts at least once per page. If a seqscan encounters many consecutive pages containing only dead tuples, it can remain in the loop in heapgettup for a long time, and there was no CHECK_FOR_INTERRUPTS anywhere in that loop. This meant there were real-world situations where a query would be effectively uncancelable for long stretches. Add a check placed to occur once per page, which should be enough to provide reasonable response time without adding any measurable overhead. Report and patch by Merlin Moncure (though I tweaked it a bit). Back-patch to all supported branches.
2012-05-23 01:42:05 +02:00
/*
* Be sure to check for interrupts at least once per page. Checks at
* higher code levels won't be able to stop a seqscan that encounters many
* pages' worth of consecutive dead tuples.
*/
CHECK_FOR_INTERRUPTS();
Make large sequential scans and VACUUMs work in a limited-size "ring" of buffers, rather than blowing out the whole shared-buffer arena. Aside from avoiding cache spoliation, this fixes the problem that VACUUM formerly tended to cause a WAL flush for every page it modified, because we had it hacked to use only a single buffer. Those flushes will now occur only once per ring-ful. The exact ring size, and the threshold for seqscans to switch into the ring usage pattern, remain under debate; but the infrastructure seems done. The key bit of infrastructure is a new optional BufferAccessStrategy object that can be passed to ReadBuffer operations; this replaces the former StrategyHintVacuum API. This patch also changes the buffer usage-count methodology a bit: we now advance usage_count when first pinning a buffer, rather than when last unpinning it. To preserve the behavior that a buffer's lifetime starts to decrease when it's released, the clock sweep code is modified to not decrement usage_count of pinned buffers. Work not done in this commit: teach GiST and GIN indexes to use the vacuum BufferAccessStrategy for vacuum-driven fetches. Original patch by Simon, reworked by Heikki and again by Tom.
2007-05-30 22:12:03 +02:00
/* read page using selected strategy */
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
scan->rs_cbuf = ReadBufferExtended(scan->rs_base.rs_rd, MAIN_FORKNUM, block,
Unite ReadBufferWithFork, ReadBufferWithStrategy, and ZeroOrReadBuffer functions into one ReadBufferExtended function, that takes the strategy and mode as argument. There's three modes, RBM_NORMAL which is the default used by plain ReadBuffer(), RBM_ZERO, which replaces ZeroOrReadBuffer, and a new mode RBM_ZERO_ON_ERROR, which allows callers to read corrupt pages without throwing an error. The FSM needs the new mode to recover from corrupt pages, which could happend if we crash after extending an FSM file, and the new page is "torn". Add fork number to some error messages in bufmgr.c, that still lacked it.
2008-10-31 16:05:00 +01:00
RBM_NORMAL, scan->rs_strategy);
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
scan->rs_cblock = block;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
if (!(scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE))
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
return;
buffer = scan->rs_cbuf;
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
snapshot = scan->rs_base.rs_snapshot;
Remove RelationGetBufferWithBuffer(), which is horribly confused about appropriate pin-count manipulation, and instead use ReleaseAndReadBuffer. Make use of the fact that the passed-in buffer (if there is one) must be pinned to avoid grabbing the bufmgr spinlock when we are able to return this same buffer. Eliminate unnecessary 'previous tuple' and 'next tuple' fields of HeapScanDesc and IndexScanDesc, thereby removing a whole lot of bookkeeping from heap_getnext() and related routines.
2001-06-09 20:16:59 +02:00
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
/*
* Prune and repair fragmentation for the whole page, if possible.
*/
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
heap_page_prune_opt(scan->rs_base.rs_rd, buffer);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
* We must hold share lock on the buffer content while examining tuple
* visibility. Afterwards, however, the tuples we have found to be
* visible are guaranteed good as long as we hold the buffer pin.
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*/
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
LockBuffer(buffer, BUFFER_LOCK_SHARE);
Error message editing in backend/access.
2003-07-21 22:29:40 +02:00
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
page = BufferGetPage(buffer);
TestForOldSnapshot(snapshot, scan->rs_base.rs_rd, page);
lines = PageGetMaxOffsetNumber(page);
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
ntup = 0;
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
/*
* If the all-visible flag indicates that all tuples on the page are
Update comment in heapgetpage() regarding PD_ALL_VISIBLE vs. Hot Standby. Pavan Deolasee, slightly modified by me
2012-12-14 21:44:38 +01:00
* visible to everyone, we can skip the per-tuple visibility tests.
*
* Note: In hot standby, a tuple that's already visible to all
code: replace 'master' with 'primary' where appropriate. Also changed "in the primary" to "on the primary", and added a few "the" before "primary". Author: Andres Freund Reviewed-By: David Steele Discussion: https://postgr.es/m/20200615182235.x7lch5n6kcjq4aue@alap3.anarazel.de
2020-06-14 23:05:18 +02:00
* transactions on the primary might still be invisible to a read-only
Update comment in heapgetpage() regarding PD_ALL_VISIBLE vs. Hot Standby. Pavan Deolasee, slightly modified by me
2012-12-14 21:44:38 +01:00
* transaction in the standby. We partly handle this problem by tracking
* the minimum xmin of visible tuples as the cut-off XID while marking a
code: replace 'master' with 'primary' where appropriate. Also changed "in the primary" to "on the primary", and added a few "the" before "primary". Author: Andres Freund Reviewed-By: David Steele Discussion: https://postgr.es/m/20200615182235.x7lch5n6kcjq4aue@alap3.anarazel.de
2020-06-14 23:05:18 +02:00
* page all-visible on the primary and WAL log that along with the
Update comment in heapgetpage() regarding PD_ALL_VISIBLE vs. Hot Standby. Pavan Deolasee, slightly modified by me
2012-12-14 21:44:38 +01:00
* visibility map SET operation. In hot standby, we wait for (or abort)
* all transactions that can potentially may not see one or more tuples on
* the page. That's how index-only scans work fine in hot standby. A
* crucial difference between index-only scans and heap scans is that the
* index-only scan completely relies on the visibility map where as heap
* scan looks at the page-level PD_ALL_VISIBLE flag. We are not sure if
* the page-level flag can be trusted in the same way, because it might
* get propagated somehow without being explicitly WAL-logged, e.g. via a
* full page write. Until we can prove that beyond doubt, let's check each
* tuple for visibility the hard way.
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
*/
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
all_visible = PageIsAllVisible(page) && !snapshot->takenDuringRecovery;
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
Push lpp variable closer to usage in heapgetpage() Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-01-02 09:39:04 +01:00
for (lineoff = FirstOffsetNumber; lineoff <= lines; lineoff++)
Remove RelationGetBufferWithBuffer(), which is horribly confused about appropriate pin-count manipulation, and instead use ReleaseAndReadBuffer. Make use of the fact that the passed-in buffer (if there is one) must be pinned to avoid grabbing the bufmgr spinlock when we are able to return this same buffer. Eliminate unnecessary 'previous tuple' and 'next tuple' fields of HeapScanDesc and IndexScanDesc, thereby removing a whole lot of bookkeeping from heap_getnext() and related routines.
2001-06-09 20:16:59 +02:00
{
Push lpp variable closer to usage in heapgetpage() Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-01-02 09:39:04 +01:00
ItemId lpp = PageGetItemId(page, lineoff);
HeapTupleData loctup;
bool valid;
if (!ItemIdIsNormal(lpp))
continue;
loctup.t_tableOid = RelationGetRelid(scan->rs_base.rs_rd);
loctup.t_data = (HeapTupleHeader) PageGetItem(page, lpp);
loctup.t_len = ItemIdGetLength(lpp);
ItemPointerSet(&(loctup.t_self), block, lineoff);
if (all_visible)
valid = true;
else
valid = HeapTupleSatisfiesVisibility(&loctup, snapshot, buffer);
HeapCheckForSerializableConflictOut(valid, scan->rs_base.rs_rd,
&loctup, buffer, snapshot);
if (valid)
scan->rs_vistuples[ntup++] = lineoff;
Remove RelationGetBufferWithBuffer(), which is horribly confused about appropriate pin-count manipulation, and instead use ReleaseAndReadBuffer. Make use of the fact that the passed-in buffer (if there is one) must be pinned to avoid grabbing the bufmgr spinlock when we are able to return this same buffer. Eliminate unnecessary 'previous tuple' and 'next tuple' fields of HeapScanDesc and IndexScanDesc, thereby removing a whole lot of bookkeeping from heap_getnext() and related routines.
2001-06-09 20:16:59 +02:00
}
pgindent run. Make it all clean.
2001-03-22 05:01:46 +01:00
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
Assert(ntup <= MaxHeapTuplesPerPage);
scan->rs_ntuples = ntup;
}
Refactor heapam.c adding heapgettup_initial_block function Here we adjust heapgettup() and heapgettup_pagemode() to move the code that fetches the first block number to scan out into a helper function. This removes some code duplication. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-02 02:17:15 +01:00
/*
* heapgettup_initial_block - return the first BlockNumber to scan
*
* Returns InvalidBlockNumber when there are no blocks to scan. This can
* occur with empty tables and in parallel scans when parallel workers get all
* of the pages before we can get a chance to get our first page.
*/
static BlockNumber
heapgettup_initial_block(HeapScanDesc scan, ScanDirection dir)
{
Assert(!scan->rs_inited);
/* When there are no pages to scan, return InvalidBlockNumber */
if (scan->rs_nblocks == 0 || scan->rs_numblocks == 0)
return InvalidBlockNumber;
if (ScanDirectionIsForward(dir))
{
/* serial scan */
if (scan->rs_base.rs_parallel == NULL)
return scan->rs_startblock;
else
{
/* parallel scan */
table_block_parallelscan_startblock_init(scan->rs_base.rs_rd,
scan->rs_parallelworkerdata,
(ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
/* may return InvalidBlockNumber if there are no more blocks */
return table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
scan->rs_parallelworkerdata,
(ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
}
}
else
{
/* backward parallel scan not supported */
Assert(scan->rs_base.rs_parallel == NULL);
/*
* Disable reporting to syncscan logic in a backwards scan; it's not
* very likely anyone else is doing the same thing at the same time,
* and much more likely that we'll just bollix things for forward
* scanners.
*/
scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
/*
* Start from last page of the scan. Ensure we take into account
* rs_numblocks if it's been adjusted by heap_setscanlimits().
*/
if (scan->rs_numblocks != InvalidBlockNumber)
return (scan->rs_startblock + scan->rs_numblocks - 1) % scan->rs_nblocks;
if (scan->rs_startblock > 0)
return scan->rs_startblock - 1;
return scan->rs_nblocks - 1;
}
}
Add helper functions to simplify heapgettup code Here we add heapgettup_start_page() and heapgettup_continue_page() to simplify the code in the heapgettup() function. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-03 00:38:42 +01:00
/*
* heapgettup_start_page - helper function for heapgettup()
*
* Return the next page to scan based on the scan->rs_cbuf and set *linesleft
* to the number of tuples on this page. Also set *lineoff to the first
* offset to scan with forward scans getting the first offset and backward
* getting the final offset on the page.
*/
static Page
heapgettup_start_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
OffsetNumber *lineoff)
{
Page page;
Assert(scan->rs_inited);
Assert(BufferIsValid(scan->rs_cbuf));
/* Caller is responsible for ensuring buffer is locked if needed */
page = BufferGetPage(scan->rs_cbuf);
TestForOldSnapshot(scan->rs_base.rs_snapshot, scan->rs_base.rs_rd, page);
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
*linesleft = PageGetMaxOffsetNumber(page) - FirstOffsetNumber + 1;
Add helper functions to simplify heapgettup code Here we add heapgettup_start_page() and heapgettup_continue_page() to simplify the code in the heapgettup() function. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-03 00:38:42 +01:00
if (ScanDirectionIsForward(dir))
*lineoff = FirstOffsetNumber;
else
*lineoff = (OffsetNumber) (*linesleft);
/* lineoff now references the physically previous or next tid */
return page;
}
/*
* heapgettup_continue_page - helper function for heapgettup()
*
* Return the next page to scan based on the scan->rs_cbuf and set *linesleft
* to the number of tuples left to scan on this page. Also set *lineoff to
* the next offset to scan according to the ScanDirection in 'dir'.
*/
static inline Page
heapgettup_continue_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
OffsetNumber *lineoff)
{
Page page;
Assert(scan->rs_inited);
Assert(BufferIsValid(scan->rs_cbuf));
/* Caller is responsible for ensuring buffer is locked if needed */
page = BufferGetPage(scan->rs_cbuf);
TestForOldSnapshot(scan->rs_base.rs_snapshot, scan->rs_base.rs_rd, page);
if (ScanDirectionIsForward(dir))
{
*lineoff = OffsetNumberNext(scan->rs_coffset);
*linesleft = PageGetMaxOffsetNumber(page) - (*lineoff) + 1;
}
else
{
/*
* The previous returned tuple may have been vacuumed since the
* previous scan when we use a non-MVCC snapshot, so we must
* re-establish the lineoff <= PageGetMaxOffsetNumber(page) invariant
*/
*lineoff = Min(PageGetMaxOffsetNumber(page), OffsetNumberPrev(scan->rs_coffset));
*linesleft = *lineoff;
}
/* lineoff now references the physically previous or next tid */
return page;
}
Reduce code duplication between heapgettup and heapgettup_pagemode The code to get the next block number was exactly the same between these two functions, so let's just put it into a helper function and call that from both locations. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-03 04:20:43 +01:00
/*
* heapgettup_advance_block - helper for heapgettup() and heapgettup_pagemode()
*
* Given the current block number, the scan direction, and various information
* contained in the scan descriptor, calculate the BlockNumber to scan next
* and return it. If there are no further blocks to scan, return
* InvalidBlockNumber to indicate this fact to the caller.
*
* This should not be called to determine the initial block number -- only for
* subsequent blocks.
*
* This also adjusts rs_numblocks when a limit has been imposed by
* heap_setscanlimits().
*/
static inline BlockNumber
heapgettup_advance_block(HeapScanDesc scan, BlockNumber block, ScanDirection dir)
{
if (ScanDirectionIsForward(dir))
{
if (scan->rs_base.rs_parallel == NULL)
{
block++;
/* wrap back to the start of the heap */
if (block >= scan->rs_nblocks)
block = 0;
/* we're done if we're back at where we started */
if (block == scan->rs_startblock)
return InvalidBlockNumber;
/* check if the limit imposed by heap_setscanlimits() is met */
if (scan->rs_numblocks != InvalidBlockNumber)
{
if (--scan->rs_numblocks == 0)
return InvalidBlockNumber;
}
/*
* Report our new scan position for synchronization purposes. We
* don't do that when moving backwards, however. That would just
* mess up any other forward-moving scanners.
*
* Note: we do this before checking for end of scan so that the
* final state of the position hint is back at the start of the
* rel. That's not strictly necessary, but otherwise when you run
* the same query multiple times the starting position would shift
* a little bit backwards on every invocation, which is confusing.
* We don't guarantee any specific ordering in general, though.
*/
if (scan->rs_base.rs_flags & SO_ALLOW_SYNC)
ss_report_location(scan->rs_base.rs_rd, block);
return block;
}
else
{
return table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
scan->rs_parallelworkerdata, (ParallelBlockTableScanDesc)
scan->rs_base.rs_parallel);
}
}
else
{
/* we're done if the last block is the start position */
if (block == scan->rs_startblock)
return InvalidBlockNumber;
/* check if the limit imposed by heap_setscanlimits() is met */
if (scan->rs_numblocks != InvalidBlockNumber)
{
if (--scan->rs_numblocks == 0)
return InvalidBlockNumber;
}
/* wrap to the end of the heap when the last page was page 0 */
if (block == 0)
block = scan->rs_nblocks;
block--;
return block;
}
}
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
/* ----------------
* heapgettup - fetch next heap tuple
*
* Initialize the scan if not already done; then advance to the next
* tuple as indicated by "dir"; return the next tuple in scan->rs_ctup,
* or set scan->rs_ctup.t_data = NULL if no more tuples.
*
* Note: the reason nkeys/key are passed separately, even though they are
* kept in the scan descriptor, is that the caller may not want us to check
* the scankeys.
*
* Note: when we fall off the end of the scan in either direction, we
* reset rs_inited. This means that a further request with the same
* scan direction will restart the scan, which is a bit odd, but a
* request with the opposite scan direction will start a fresh scan
* in the proper direction. The latter is required behavior for cursors,
* while the former case is generally undefined behavior in Postgres
* so we don't care too much.
* ----------------
*/
static void
heapgettup(HeapScanDesc scan,
Cleanup the usage of ScanDirection: use the symbolic names for the possible ScanDirection alternatives rather than magic numbers (-1, 0, 1). Also, use the ScanDirection macros in a few places rather than directly checking whether `dir == ForwardScanDirection' and the like. Per patch from James William Pye. His patch also changed ScanDirection to be a "char" rather than an enum, which I haven't applied.
2006-02-22 00:01:54 +01:00
ScanDirection dir,
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
int nkeys,
ScanKey key)
{
HeapTuple tuple = &(scan->rs_ctup);
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
BlockNumber block;
Page page;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
OffsetNumber lineoff;
int linesleft;
Further refactor of heapgettup and heapgettup_pagemode Backward and forward scans share much of the same page acquisition code. Here we consolidate that code to reduce some duplication. Additionally, add a new rs_coffset field to HeapScanDescData to track the offset of the current tuple. The new field fits nicely into the padding between a bool and BlockNumber field and saves having to look at the last returned tuple to figure out which offset we should be looking at for the current tuple. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-02 23:48:39 +01:00
if (unlikely(!scan->rs_inited))
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
{
Further refactor of heapgettup and heapgettup_pagemode Backward and forward scans share much of the same page acquisition code. Here we consolidate that code to reduce some duplication. Additionally, add a new rs_coffset field to HeapScanDescData to track the offset of the current tuple. The new field fits nicely into the padding between a bool and BlockNumber field and saves having to look at the last returned tuple to figure out which offset we should be looking at for the current tuple. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-02 23:48:39 +01:00
block = heapgettup_initial_block(scan, dir);
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
/* ensure rs_cbuf is invalid when we get InvalidBlockNumber */
Assert(block != InvalidBlockNumber || !BufferIsValid(scan->rs_cbuf));
Add helper functions to simplify heapgettup code Here we add heapgettup_start_page() and heapgettup_continue_page() to simplify the code in the heapgettup() function. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-03 00:38:42 +01:00
scan->rs_inited = true;
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
}
Remove dead NoMovementScanDirection code Here remove some dead code from heapgettup() and heapgettup_pagemode() which was trying to support NoMovementScanDirection scans. This code can never be reached as standard_ExecutorRun() never calls ExecutePlan with NoMovementScanDirection. Additionally, plans which were scanning an unordered index would use NoMovementScanDirection rather than ForwardScanDirection. There was no real need for this, so here we adjust this so we use ForwardScanDirection for unordered index scans. A comment in pathnodes.h claimed that NoMovementScanDirection was used for PathKey reasons, but if that was true, it no longer is, per code in build_index_paths(). This does change the non-text format of the EXPLAIN output so that unordered index scans now have a "Forward" scan direction rather than "NoMovement". The text format of EXPLAIN has not changed. Author: Melanie Plageman Reviewed-by: Tom Lane, David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-01-31 22:52:41 +01:00
else
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
{
Further refactor of heapgettup and heapgettup_pagemode Backward and forward scans share much of the same page acquisition code. Here we consolidate that code to reduce some duplication. Additionally, add a new rs_coffset field to HeapScanDescData to track the offset of the current tuple. The new field fits nicely into the padding between a bool and BlockNumber field and saves having to look at the last returned tuple to figure out which offset we should be looking at for the current tuple. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-02 23:48:39 +01:00
/* continue from previously returned page/tuple */
block = scan->rs_cblock;
pgindent run for 8.3.
2007-11-15 22:14:46 +01:00
Further refactor of heapgettup and heapgettup_pagemode Backward and forward scans share much of the same page acquisition code. Here we consolidate that code to reduce some duplication. Additionally, add a new rs_coffset field to HeapScanDescData to track the offset of the current tuple. The new field fits nicely into the padding between a bool and BlockNumber field and saves having to look at the last returned tuple to figure out which offset we should be looking at for the current tuple. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-02 23:48:39 +01:00
LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
Add helper functions to simplify heapgettup code Here we add heapgettup_start_page() and heapgettup_continue_page() to simplify the code in the heapgettup() function. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-03 00:38:42 +01:00
page = heapgettup_continue_page(scan, dir, &linesleft, &lineoff);
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
goto continue_page;
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
}
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
* advance the scan until we find a qualifying tuple or run out of stuff
* to scan
*/
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
while (block != InvalidBlockNumber)
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
{
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
heapgetpage((TableScanDesc) scan, block);
LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
page = heapgettup_start_page(scan, dir, &linesleft, &lineoff);
continue_page:
Truncate line pointer array during VACUUM. Teach VACUUM to truncate the line pointer array of each heap page when a contiguous group of LP_UNUSED line pointers appear at the end of the array -- these unused and unreferenced items are excluded. This process occurs during VACUUM's second pass over the heap, right after LP_DEAD line pointers on the page (those encountered/pruned during the first pass) are marked LP_UNUSED. Truncation avoids line pointer bloat with certain workloads, particularly those involving continual range DELETEs and bulk INSERTs against the same table. Also harden heapam code to check for an out-of-range page offset number in places where we weren't already doing so. Author: Matthias van de Meent <boekewurm+postgres@gmail.com> Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAEze2WjgaQc55Y5f5CQd3L=eS5CZcff2Obxp=O6pto8-f0hC4w@mail.gmail.com Discussion: https://postgr.es/m/CAH2-Wzn6a64PJM1Ggzm=uvx2otsopJMhFQj_g1rAj4GWr3ZSzw@mail.gmail.com
2021-04-07 17:47:15 +02:00
/*
* Only continue scanning the page while we have lines left.
*
* Note that this protects us from accessing line pointers past
* PageGetMaxOffsetNumber(); both for forward scans when we resume the
* table scan, and for when we start scanning a new page.
*/
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
for (; linesleft > 0; linesleft--, lineoff += dir)
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
{
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
bool visible;
ItemId lpp = PageGetItemId(page, lineoff);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
if (!ItemIdIsNormal(lpp))
continue;
tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
tuple->t_len = ItemIdGetLength(lpp);
ItemPointerSet(&(tuple->t_self), block, lineoff);
visible = HeapTupleSatisfiesVisibility(tuple,
scan->rs_base.rs_snapshot,
scan->rs_cbuf);
HeapCheckForSerializableConflictOut(visible, scan->rs_base.rs_rd,
tuple, scan->rs_cbuf,
scan->rs_base.rs_snapshot);
/* skip tuples not visible to this snapshot */
if (!visible)
continue;
/* skip any tuples that don't match the scan key */
if (key != NULL &&
!HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
nkeys, key))
continue;
LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
scan->rs_coffset = lineoff;
return;
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
}
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
* if we get here, it means we've exhausted the items on this page and
Remove RelationGetBufferWithBuffer(), which is horribly confused about appropriate pin-count manipulation, and instead use ReleaseAndReadBuffer. Make use of the fact that the passed-in buffer (if there is one) must be pinned to avoid grabbing the bufmgr spinlock when we are able to return this same buffer. Eliminate unnecessary 'previous tuple' and 'next tuple' fields of HeapScanDesc and IndexScanDesc, thereby removing a whole lot of bookkeeping from heap_getnext() and related routines.
2001-06-09 20:16:59 +02:00
* it's time to move to the next.
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
*/
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Reduce code duplication between heapgettup and heapgettup_pagemode The code to get the next block number was exactly the same between these two functions, so let's just put it into a helper function and call that from both locations. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-03 04:20:43 +01:00
/* get the BlockNumber to scan next */
block = heapgettup_advance_block(scan, block, dir);
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
}
Arrange for large sequential scans to synchronize with each other, so that when multiple backends are scanning the same relation concurrently, each page is (ideally) read only once. Jeff Davis, with review by Heikki and Tom.
2007-06-08 20:23:53 +02:00
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
/* end of scan */
if (BufferIsValid(scan->rs_cbuf))
ReleaseBuffer(scan->rs_cbuf);
Remove RelationGetBufferWithBuffer(), which is horribly confused about appropriate pin-count manipulation, and instead use ReleaseAndReadBuffer. Make use of the fact that the passed-in buffer (if there is one) must be pinned to avoid grabbing the bufmgr spinlock when we are able to return this same buffer. Eliminate unnecessary 'previous tuple' and 'next tuple' fields of HeapScanDesc and IndexScanDesc, thereby removing a whole lot of bookkeeping from heap_getnext() and related routines.
2001-06-09 20:16:59 +02:00
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
scan->rs_cbuf = InvalidBuffer;
scan->rs_cblock = InvalidBlockNumber;
tuple->t_data = NULL;
scan->rs_inited = false;
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
}
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
/* ----------------
* heapgettup_pagemode - fetch next heap tuple in page-at-a-time mode
*
* Same API as heapgettup, but used in page-at-a-time mode
*
* The internal logic is much the same as heapgettup's too, but there are some
* differences: we do not take the buffer content lock (that only needs to
* happen inside heapgetpage), and we iterate through just the tuples listed
* in rs_vistuples[] rather than all tuples on the page. Notice that
* lineindex is 0-based, where the corresponding loop variable lineoff in
* heapgettup is 1-based.
* ----------------
*/
static void
heapgettup_pagemode(HeapScanDesc scan,
Cleanup the usage of ScanDirection: use the symbolic names for the possible ScanDirection alternatives rather than magic numbers (-1, 0, 1). Also, use the ScanDirection macros in a few places rather than directly checking whether `dir == ForwardScanDirection' and the like. Per patch from James William Pye. His patch also changed ScanDirection to be a "char" rather than an enum, which I haven't applied.
2006-02-22 00:01:54 +01:00
ScanDirection dir,
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
int nkeys,
ScanKey key)
{
HeapTuple tuple = &(scan->rs_ctup);
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
BlockNumber block;
Page page;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
int lineindex;
int linesleft;
Further refactor of heapgettup and heapgettup_pagemode Backward and forward scans share much of the same page acquisition code. Here we consolidate that code to reduce some duplication. Additionally, add a new rs_coffset field to HeapScanDescData to track the offset of the current tuple. The new field fits nicely into the padding between a bool and BlockNumber field and saves having to look at the last returned tuple to figure out which offset we should be looking at for the current tuple. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-02 23:48:39 +01:00
if (unlikely(!scan->rs_inited))
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
{
Further refactor of heapgettup and heapgettup_pagemode Backward and forward scans share much of the same page acquisition code. Here we consolidate that code to reduce some duplication. Additionally, add a new rs_coffset field to HeapScanDescData to track the offset of the current tuple. The new field fits nicely into the padding between a bool and BlockNumber field and saves having to look at the last returned tuple to figure out which offset we should be looking at for the current tuple. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-02 23:48:39 +01:00
block = heapgettup_initial_block(scan, dir);
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
/* ensure rs_cbuf is invalid when we get InvalidBlockNumber */
Assert(block != InvalidBlockNumber || !BufferIsValid(scan->rs_cbuf));
Further refactor of heapgettup and heapgettup_pagemode Backward and forward scans share much of the same page acquisition code. Here we consolidate that code to reduce some duplication. Additionally, add a new rs_coffset field to HeapScanDescData to track the offset of the current tuple. The new field fits nicely into the padding between a bool and BlockNumber field and saves having to look at the last returned tuple to figure out which offset we should be looking at for the current tuple. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-02 23:48:39 +01:00
scan->rs_inited = true;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
}
Remove dead NoMovementScanDirection code Here remove some dead code from heapgettup() and heapgettup_pagemode() which was trying to support NoMovementScanDirection scans. This code can never be reached as standard_ExecutorRun() never calls ExecutePlan with NoMovementScanDirection. Additionally, plans which were scanning an unordered index would use NoMovementScanDirection rather than ForwardScanDirection. There was no real need for this, so here we adjust this so we use ForwardScanDirection for unordered index scans. A comment in pathnodes.h claimed that NoMovementScanDirection was used for PathKey reasons, but if that was true, it no longer is, per code in build_index_paths(). This does change the non-text format of the EXPLAIN output so that unordered index scans now have a "Forward" scan direction rather than "NoMovement". The text format of EXPLAIN has not changed. Author: Melanie Plageman Reviewed-by: Tom Lane, David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-01-31 22:52:41 +01:00
else
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
{
Further refactor of heapgettup and heapgettup_pagemode Backward and forward scans share much of the same page acquisition code. Here we consolidate that code to reduce some duplication. Additionally, add a new rs_coffset field to HeapScanDescData to track the offset of the current tuple. The new field fits nicely into the padding between a bool and BlockNumber field and saves having to look at the last returned tuple to figure out which offset we should be looking at for the current tuple. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-02 23:48:39 +01:00
/* continue from previously returned page/tuple */
block = scan->rs_cblock; /* current page */
page = BufferGetPage(scan->rs_cbuf);
TestForOldSnapshot(scan->rs_base.rs_snapshot, scan->rs_base.rs_rd, page);
pgindent run for 8.3.
2007-11-15 22:14:46 +01:00
Further refactor of heapgettup and heapgettup_pagemode Backward and forward scans share much of the same page acquisition code. Here we consolidate that code to reduce some duplication. Additionally, add a new rs_coffset field to HeapScanDescData to track the offset of the current tuple. The new field fits nicely into the padding between a bool and BlockNumber field and saves having to look at the last returned tuple to figure out which offset we should be looking at for the current tuple. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-02 23:48:39 +01:00
lineindex = scan->rs_cindex + dir;
if (ScanDirectionIsForward(dir))
linesleft = scan->rs_ntuples - lineindex;
Some marginal additional hacking to shave a few more cycles off heapgettup.
2005-11-26 06:03:06 +01:00
else
Further refactor of heapgettup and heapgettup_pagemode Backward and forward scans share much of the same page acquisition code. Here we consolidate that code to reduce some duplication. Additionally, add a new rs_coffset field to HeapScanDescData to track the offset of the current tuple. The new field fits nicely into the padding between a bool and BlockNumber field and saves having to look at the last returned tuple to figure out which offset we should be looking at for the current tuple. Author: Melanie Plageman Reviewed-by: David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-02 23:48:39 +01:00
linesleft = scan->rs_cindex;
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
/* lineindex now references the next or previous visible tid */
goto continue_page;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
}
/*
* advance the scan until we find a qualifying tuple or run out of stuff
* to scan
*/
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
while (block != InvalidBlockNumber)
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
{
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
heapgetpage((TableScanDesc) scan, block);
page = BufferGetPage(scan->rs_cbuf);
TestForOldSnapshot(scan->rs_base.rs_snapshot, scan->rs_base.rs_rd, page);
linesleft = scan->rs_ntuples;
lineindex = ScanDirectionIsForward(dir) ? 0 : linesleft - 1;
/* lineindex now references the next or previous visible tid */
continue_page:
for (; linesleft > 0; linesleft--, lineindex += dir)
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
{
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
ItemId lpp;
OffsetNumber lineoff;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
lineoff = scan->rs_vistuples[lineindex];
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
lpp = PageGetItemId(page, lineoff);
Redefine the lp_flags field of item pointers as having four states, rather than two independent bits (one of which was never used in heap pages anyway, or at least hadn't been in a very long time). This gives us flexibility to add the HOT notions of redirected and dead item pointers without requiring anything so klugy as magic values of lp_off and lp_len. The state values are chosen so that for the states currently in use (pre-HOT) there is no change in the physical representation.
2007-09-13 00:10:26 +02:00
Assert(ItemIdIsNormal(lpp));
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
tuple->t_len = ItemIdGetLength(lpp);
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
ItemPointerSet(&(tuple->t_self), block, lineoff);
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
/* skip any tuples that don't match the scan key */
if (key != NULL &&
!HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
nkeys, key))
continue;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
scan->rs_cindex = lineindex;
return;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
}
Reduce code duplication between heapgettup and heapgettup_pagemode The code to get the next block number was exactly the same between these two functions, so let's just put it into a helper function and call that from both locations. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_bvkhka0CZQun28KTqhuUh5ZqY=_T8QEqZqOL02rpi2bw@mail.gmail.com
2023-02-03 04:20:43 +01:00
/* get the BlockNumber to scan next */
block = heapgettup_advance_block(scan, block, dir);
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
}
More refactoring of heapgettup() and heapgettup_pagemode() Here we further simplify the code in heapgettup() and heapgettup_pagemode() to make better use of the helper functions added in the previous recent refactors in this area. In passing, remove an unneeded cast added in 8ca6d49f6. Author: Melanie Plageman Reviewed-by: Andres Freund, David Rowley Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-02-07 05:24:07 +01:00
/* end of scan */
if (BufferIsValid(scan->rs_cbuf))
ReleaseBuffer(scan->rs_cbuf);
scan->rs_cbuf = InvalidBuffer;
scan->rs_cblock = InvalidBlockNumber;
tuple->t_data = NULL;
scan->rs_inited = false;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
}
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/* ----------------------------------------------------------------
* heap access method interface
* ----------------------------------------------------------------
*/
Modify heap_open()/heap_openr() API per pghackers discussion of 11 July. These two routines will now ALWAYS elog() on failure, whether you ask for a lock or not. If you really want to get a NULL return on failure, call the new routines heap_open_nofail()/heap_openr_nofail(). By my count there are only about three places that actually want that behavior. There were rather more than three places that were missing the check they needed to make under the old convention :-(.
2000-08-03 21:19:38 +02:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
TableScanDesc
Restructure indexscan API (index_beginscan, index_getnext) per yesterday's proposal to pghackers. Also remove unnecessary parameters to heap_beginscan, heap_rescan. I modified pg_proc.h to reflect the new numbers of parameters for the AM interface routines, but did not force an initdb because nothing actually looks at those fields.
2002-05-21 01:51:44 +02:00
heap_beginscan(Relation relation, Snapshot snapshot,
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
int nkeys, ScanKey key,
ParallelTableScanDesc parallel_scan,
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
uint32 flags)
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
{
heap_fetch requires buffer pointer, must be released; heap_getnext no longer returns buffer pointer, can be gotten from scan; descriptor; bootstrap can create multi-key indexes; pg_procname index now is multi-key index; oidint2, oidint4, oidname are gone (must be removed from regression tests); use System Cache rather than sequential scan in many places; heap_modifytuple no longer takes buffer parameter; remove unused buffer parameter in a few other functions; oid8 is not index-able; remove some use of single-character variable names; cleanup Buffer variables usage and scan descriptor looping; cleaned up allocation and freeing of tuples; 18k lines of diff;
1998-08-19 04:04:17 +02:00
HeapScanDesc scan;
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
* increment relation ref count while scanning relation
Mega-commit to make heap_open/heap_openr/heap_close take an additional argument specifying the kind of lock to acquire/release (or 'NoLock' to do no lock processing). Ensure that all relations are locked with some appropriate lock level before being examined --- this ensures that relevant shared-inval messages have been processed and should prevent problems caused by concurrent VACUUM. Fix several bugs having to do with mismatched increment/decrement of relation ref count and mismatched heap_open/close (which amounts to the same thing). A bogus ref count on a relation doesn't matter much *unless* a SI Inval message happens to arrive at the wrong time, which is probably why we got away with this sloppiness for so long. Repair missing grab of AccessExclusiveLock in DROP TABLE, ALTER/RENAME TABLE, etc, as noted by Hiroshi. Recommend 'make clean all' after pulling this update; I modified the Relation struct layout slightly. Will post further discussion to pghackers list shortly.
1999-09-18 21:08:25 +02:00
*
Make DROP TABLE rollback-able: postpone physical file delete until commit. (WAL logging for this is not done yet, however.) Clean up a number of really crufty things that are no longer needed now that DROP behaves nicely. Make temp table mapper do the right things when drop or rename affecting a temp table is rolled back. Also, remove "relation modified while in use" error check, in favor of locking tables at first reference and holding that lock throughout the statement.
2000-11-08 23:10:03 +01:00
* This is just to make really sure the relcache entry won't go away while
* the scan has a pointer to it. Caller should be holding the rel open
* anyway, so this is redundant in all normal scenarios...
Mega-commit to make heap_open/heap_openr/heap_close take an additional argument specifying the kind of lock to acquire/release (or 'NoLock' to do no lock processing). Ensure that all relations are locked with some appropriate lock level before being examined --- this ensures that relevant shared-inval messages have been processed and should prevent problems caused by concurrent VACUUM. Fix several bugs having to do with mismatched increment/decrement of relation ref count and mismatched heap_open/close (which amounts to the same thing). A bogus ref count on a relation doesn't matter much *unless* a SI Inval message happens to arrive at the wrong time, which is probably why we got away with this sloppiness for so long. Repair missing grab of AccessExclusiveLock in DROP TABLE, ALTER/RENAME TABLE, etc, as noted by Hiroshi. Recommend 'make clean all' after pulling this update; I modified the Relation struct layout slightly. Will post further discussion to pghackers list shortly.
1999-09-18 21:08:25 +02:00
*/
Make DROP TABLE rollback-able: postpone physical file delete until commit. (WAL logging for this is not done yet, however.) Clean up a number of really crufty things that are no longer needed now that DROP behaves nicely. Make temp table mapper do the right things when drop or rename affecting a temp table is rolled back. Also, remove "relation modified while in use" error check, in favor of locking tables at first reference and holding that lock throughout the statement.
2000-11-08 23:10:03 +01:00
RelationIncrementReferenceCount(relation);
Mega-commit to make heap_open/heap_openr/heap_close take an additional argument specifying the kind of lock to acquire/release (or 'NoLock' to do no lock processing). Ensure that all relations are locked with some appropriate lock level before being examined --- this ensures that relevant shared-inval messages have been processed and should prevent problems caused by concurrent VACUUM. Fix several bugs having to do with mismatched increment/decrement of relation ref count and mismatched heap_open/close (which amounts to the same thing). A bogus ref count on a relation doesn't matter much *unless* a SI Inval message happens to arrive at the wrong time, which is probably why we got away with this sloppiness for so long. Repair missing grab of AccessExclusiveLock in DROP TABLE, ALTER/RENAME TABLE, etc, as noted by Hiroshi. Recommend 'make clean all' after pulling this update; I modified the Relation struct layout slightly. Will post further discussion to pghackers list shortly.
1999-09-18 21:08:25 +02:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
* allocate and initialize scan descriptor
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
*/
heap_fetch requires buffer pointer, must be released; heap_getnext no longer returns buffer pointer, can be gotten from scan; descriptor; bootstrap can create multi-key indexes; pg_procname index now is multi-key index; oidint2, oidint4, oidname are gone (must be removed from regression tests); use System Cache rather than sequential scan in many places; heap_modifytuple no longer takes buffer parameter; remove unused buffer parameter in a few other functions; oid8 is not index-able; remove some use of single-character variable names; cleanup Buffer variables usage and scan descriptor looping; cleaned up allocation and freeing of tuples; 18k lines of diff;
1998-08-19 04:04:17 +02:00
scan = (HeapScanDesc) palloc(sizeof(HeapScanDescData));
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
scan->rs_base.rs_rd = relation;
scan->rs_base.rs_snapshot = snapshot;
scan->rs_base.rs_nkeys = nkeys;
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
scan->rs_base.rs_flags = flags;
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
scan->rs_base.rs_parallel = parallel_scan;
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
scan->rs_strategy = NULL; /* set in initscan */
Statistical system views (yet without the config stuff, but it's hard to keep such massive changes in sync with the tree so I need to get it in and work from there now). Jan
2001-06-22 21:16:24 +02:00
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
/*
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
* Disable page-at-a-time mode if it's not a MVCC-safe snapshot.
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
*/
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
if (!(snapshot && IsMVCCSnapshot(snapshot)))
scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
Move the PredicateLockRelation() call from nodeSeqscan.c to heapam.c. It's more consistent that way, since all the other PredicateLock* calls are made in various heapam.c and index AM functions. The call in nodeSeqscan.c was unnecessarily aggressive anyway, there's no need to try to lock the relation every time a tuple is fetched, it's enough to do it once. This has the user-visible effect that if a seq scan is initialized in the executor, but never executed, we now acquire the predicate lock on the heap relation anyway. We could avoid that by taking the lock on the first heap_getnext() call instead, but it doesn't seem worth the trouble given that it feels more natural to do it in heap_beginscan(). Also, remove the retail PredicateLockTuple() calls from heap_getnext(). In a seqscan, started with heap_begin(), we're holding a whole-relation predicate lock on the heap so there's no need to lock the tuples individually. Kevin Grittner and me
2011-06-29 20:43:53 +02:00
/*
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
* For seqscan and sample scans in a serializable transaction, acquire a
* predicate lock on the entire relation. This is required not only to
* lock all the matching tuples, but also to conflict with new insertions
* into the table. In an indexscan, we take page locks on the index pages
* covering the range specified in the scan qual, but in a heap scan there
* is nothing more fine-grained to lock. A bitmap scan is a different
* story, there we have already scanned the index and locked the index
* pages covering the predicate. But in that case we still have to lock
* any matching heap tuples. For sample scan we could optimize the locking
* to be at least page-level granularity, but we'd need to add per-tuple
* locking for that.
Move the PredicateLockRelation() call from nodeSeqscan.c to heapam.c. It's more consistent that way, since all the other PredicateLock* calls are made in various heapam.c and index AM functions. The call in nodeSeqscan.c was unnecessarily aggressive anyway, there's no need to try to lock the relation every time a tuple is fetched, it's enough to do it once. This has the user-visible effect that if a seq scan is initialized in the executor, but never executed, we now acquire the predicate lock on the heap relation anyway. We could avoid that by taking the lock on the first heap_getnext() call instead, but it doesn't seem worth the trouble given that it feels more natural to do it in heap_beginscan(). Also, remove the retail PredicateLockTuple() calls from heap_getnext(). In a seqscan, started with heap_begin(), we're holding a whole-relation predicate lock on the heap so there's no need to lock the tuples individually. Kevin Grittner and me
2011-06-29 20:43:53 +02:00
*/
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
if (scan->rs_base.rs_flags & (SO_TYPE_SEQSCAN | SO_TYPE_SAMPLESCAN))
{
/*
* Ensure a missing snapshot is noticed reliably, even if the
* isolation mode means predicate locking isn't performed (and
* therefore the snapshot isn't used here).
*/
Assert(snapshot);
Move the PredicateLockRelation() call from nodeSeqscan.c to heapam.c. It's more consistent that way, since all the other PredicateLock* calls are made in various heapam.c and index AM functions. The call in nodeSeqscan.c was unnecessarily aggressive anyway, there's no need to try to lock the relation every time a tuple is fetched, it's enough to do it once. This has the user-visible effect that if a seq scan is initialized in the executor, but never executed, we now acquire the predicate lock on the heap relation anyway. We could avoid that by taking the lock on the first heap_getnext() call instead, but it doesn't seem worth the trouble given that it feels more natural to do it in heap_beginscan(). Also, remove the retail PredicateLockTuple() calls from heap_getnext(). In a seqscan, started with heap_begin(), we're holding a whole-relation predicate lock on the heap so there's no need to lock the tuples individually. Kevin Grittner and me
2011-06-29 20:43:53 +02:00
PredicateLockRelation(relation, snapshot);
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
}
Move the PredicateLockRelation() call from nodeSeqscan.c to heapam.c. It's more consistent that way, since all the other PredicateLock* calls are made in various heapam.c and index AM functions. The call in nodeSeqscan.c was unnecessarily aggressive anyway, there's no need to try to lock the relation every time a tuple is fetched, it's enough to do it once. This has the user-visible effect that if a seq scan is initialized in the executor, but never executed, we now acquire the predicate lock on the heap relation anyway. We could avoid that by taking the lock on the first heap_getnext() call instead, but it doesn't seem worth the trouble given that it feels more natural to do it in heap_beginscan(). Also, remove the retail PredicateLockTuple() calls from heap_getnext(). In a seqscan, started with heap_begin(), we're holding a whole-relation predicate lock on the heap so there's no need to lock the tuples individually. Kevin Grittner and me
2011-06-29 20:43:53 +02:00
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
/* we only need to set this up once */
scan->rs_ctup.t_tableOid = RelationGetRelid(relation);
Adjust design of per-worker parallel seqscan data struct The design of the data structures which allow storage of the per-worker memory during parallel seq scans were not ideal. The work done in 56788d215 required an additional data structure to allow workers to remember the range of pages that had been allocated to them for processing during a parallel seqscan. That commit added a void pointer field to TableScanDescData to allow heapam to store the per-worker allocation information. However putting the field there made very little sense given that we have AM specific structs for that, e.g. HeapScanDescData. Here we remove the void pointer field from TableScanDescData and add a dedicated field for this purpose to HeapScanDescData. Previously we also allocated memory for this parallel per-worker data for all scans, regardless if it was a parallel scan or not. This was just a wasted allocation for non-parallel scans, so here we make the allocation conditional on the scan being parallel. Also, add previously missing pfree() to free the per-worker data in heap_endscan(). Reported-by: Andres Freund Reviewed-by: Andres Freund Discussion: https://postgr.es/m/20210317023101.anvejcfotwka6gaa@alap3.anarazel.de
2021-03-29 23:17:09 +02:00
/*
* Allocate memory to keep track of page allocation for parallel workers
* when doing a parallel scan.
*/
if (parallel_scan != NULL)
scan->rs_parallelworkerdata = palloc(sizeof(ParallelBlockTableScanWorkerData));
else
scan->rs_parallelworkerdata = NULL;
Make DROP TABLE rollback-able: postpone physical file delete until commit. (WAL logging for this is not done yet, however.) Clean up a number of really crufty things that are no longer needed now that DROP behaves nicely. Make temp table mapper do the right things when drop or rename affecting a temp table is rolled back. Also, remove "relation modified while in use" error check, in favor of locking tables at first reference and holding that lock throughout the statement.
2000-11-08 23:10:03 +01:00
/*
* we do this here instead of in initscan() because heap_rescan also calls
* initscan() and we don't want to allocate memory again
*/
Restructure indexscan API (index_beginscan, index_getnext) per yesterday's proposal to pghackers. Also remove unnecessary parameters to heap_beginscan, heap_rescan. I modified pg_proc.h to reflect the new numbers of parameters for the AM interface routines, but did not force an initdb because nothing actually looks at those fields.
2002-05-21 01:51:44 +02:00
if (nkeys > 0)
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
scan->rs_base.rs_key = (ScanKey) palloc(sizeof(ScanKeyData) * nkeys);
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
else
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
scan->rs_base.rs_key = NULL;
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Keep rs_startblock the same during heap_rescan, so that a rescan of a SeqScan node starts from the same place as the first scan did. This avoids surprising behavior of scrollable and WITH HOLD cursors, as seen in Mark Kirkwood's bug report of yesterday. It's not entirely clear whether a rescan should be forced to drop out of the syncscan mode, but for the moment I left the code behaving the same on that point. Any change there would only be a performance and not a correctness issue, anyway. Back-patch to 8.3, since the unstable behavior was created by the syncscan patch.
2009-06-10 20:54:16 +02:00
initscan(scan, key, false);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
return (TableScanDesc) scan;
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
}
void
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
heap_rescan(TableScanDesc sscan, ScanKey key, bool set_params,
bool allow_strat, bool allow_sync, bool allow_pagemode)
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
{
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
HeapScanDesc scan = (HeapScanDesc) sscan;
if (set_params)
{
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
if (allow_strat)
scan->rs_base.rs_flags |= SO_ALLOW_STRAT;
else
scan->rs_base.rs_flags &= ~SO_ALLOW_STRAT;
if (allow_sync)
scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
else
scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
if (allow_pagemode && scan->rs_base.rs_snapshot &&
IsMVCCSnapshot(scan->rs_base.rs_snapshot))
scan->rs_base.rs_flags |= SO_ALLOW_PAGEMODE;
else
scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
}
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
* unpin scan buffers
*/
Remove RelationGetBufferWithBuffer(), which is horribly confused about appropriate pin-count manipulation, and instead use ReleaseAndReadBuffer. Make use of the fact that the passed-in buffer (if there is one) must be pinned to avoid grabbing the bufmgr spinlock when we are able to return this same buffer. Eliminate unnecessary 'previous tuple' and 'next tuple' fields of HeapScanDesc and IndexScanDesc, thereby removing a whole lot of bookkeeping from heap_getnext() and related routines.
2001-06-09 20:16:59 +02:00
if (BufferIsValid(scan->rs_cbuf))
ReleaseBuffer(scan->rs_cbuf);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
* reinitialize scan descriptor
*/
Keep rs_startblock the same during heap_rescan, so that a rescan of a SeqScan node starts from the same place as the first scan did. This avoids surprising behavior of scrollable and WITH HOLD cursors, as seen in Mark Kirkwood's bug report of yesterday. It's not entirely clear whether a rescan should be forced to drop out of the syncscan mode, but for the moment I left the code behaving the same on that point. Any change there would only be a performance and not a correctness issue, anyway. Back-patch to 8.3, since the unstable behavior was created by the syncscan patch.
2009-06-10 20:54:16 +02:00
initscan(scan, key, true);
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
}
Redesign tablesample method API, and do extensive code review. The original implementation of TABLESAMPLE modeled the tablesample method API on index access methods, which wasn't a good choice because, without specialized DDL commands, there's no way to build an extension that can implement a TSM. (Raw inserts into system catalogs are not an acceptable thing to do, because we can't undo them during DROP EXTENSION, nor will pg_upgrade behave sanely.) Instead adopt an API more like procedural language handlers or foreign data wrappers, wherein the only SQL-level support object needed is a single handler function identified by having a special return type. This lets us get rid of the supporting catalog altogether, so that no custom DDL support is needed for the feature. Adjust the API so that it can support non-constant tablesample arguments (the original coding assumed we could evaluate the argument expressions at ExecInitSampleScan time, which is undesirable even if it weren't outright unsafe), and discourage sampling methods from looking at invisible tuples. Make sure that the BERNOULLI and SYSTEM methods are genuinely repeatable within and across queries, as required by the SQL standard, and deal more honestly with methods that can't support that requirement. Make a full code-review pass over the tablesample additions, and fix assorted bugs, omissions, infelicities, and cosmetic issues (such as failure to put the added code stanzas in a consistent ordering). Improve EXPLAIN's output of tablesample plans, too. Back-patch to 9.5 so that we don't have to support the original API in production.
2015-07-25 20:39:00 +02:00
void
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
heap_endscan(TableScanDesc sscan)
Redesign tablesample method API, and do extensive code review. The original implementation of TABLESAMPLE modeled the tablesample method API on index access methods, which wasn't a good choice because, without specialized DDL commands, there's no way to build an extension that can implement a TSM. (Raw inserts into system catalogs are not an acceptable thing to do, because we can't undo them during DROP EXTENSION, nor will pg_upgrade behave sanely.) Instead adopt an API more like procedural language handlers or foreign data wrappers, wherein the only SQL-level support object needed is a single handler function identified by having a special return type. This lets us get rid of the supporting catalog altogether, so that no custom DDL support is needed for the feature. Adjust the API so that it can support non-constant tablesample arguments (the original coding assumed we could evaluate the argument expressions at ExecInitSampleScan time, which is undesirable even if it weren't outright unsafe), and discourage sampling methods from looking at invisible tuples. Make sure that the BERNOULLI and SYSTEM methods are genuinely repeatable within and across queries, as required by the SQL standard, and deal more honestly with methods that can't support that requirement. Make a full code-review pass over the tablesample additions, and fix assorted bugs, omissions, infelicities, and cosmetic issues (such as failure to put the added code stanzas in a consistent ordering). Improve EXPLAIN's output of tablesample plans, too. Back-patch to 9.5 so that we don't have to support the original API in production.
2015-07-25 20:39:00 +02:00
{
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
HeapScanDesc scan = (HeapScanDesc) sscan;
Redesign tablesample method API, and do extensive code review. The original implementation of TABLESAMPLE modeled the tablesample method API on index access methods, which wasn't a good choice because, without specialized DDL commands, there's no way to build an extension that can implement a TSM. (Raw inserts into system catalogs are not an acceptable thing to do, because we can't undo them during DROP EXTENSION, nor will pg_upgrade behave sanely.) Instead adopt an API more like procedural language handlers or foreign data wrappers, wherein the only SQL-level support object needed is a single handler function identified by having a special return type. This lets us get rid of the supporting catalog altogether, so that no custom DDL support is needed for the feature. Adjust the API so that it can support non-constant tablesample arguments (the original coding assumed we could evaluate the argument expressions at ExecInitSampleScan time, which is undesirable even if it weren't outright unsafe), and discourage sampling methods from looking at invisible tuples. Make sure that the BERNOULLI and SYSTEM methods are genuinely repeatable within and across queries, as required by the SQL standard, and deal more honestly with methods that can't support that requirement. Make a full code-review pass over the tablesample additions, and fix assorted bugs, omissions, infelicities, and cosmetic issues (such as failure to put the added code stanzas in a consistent ordering). Improve EXPLAIN's output of tablesample plans, too. Back-patch to 9.5 so that we don't have to support the original API in production.
2015-07-25 20:39:00 +02:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/* Note: no locking manipulations needed */
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
* unpin scan buffers
*/
Remove RelationGetBufferWithBuffer(), which is horribly confused about appropriate pin-count manipulation, and instead use ReleaseAndReadBuffer. Make use of the fact that the passed-in buffer (if there is one) must be pinned to avoid grabbing the bufmgr spinlock when we are able to return this same buffer. Eliminate unnecessary 'previous tuple' and 'next tuple' fields of HeapScanDesc and IndexScanDesc, thereby removing a whole lot of bookkeeping from heap_getnext() and related routines.
2001-06-09 20:16:59 +02:00
if (BufferIsValid(scan->rs_cbuf))
ReleaseBuffer(scan->rs_cbuf);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
* decrement relation reference count and free scan descriptor storage
*/
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
RelationDecrementReferenceCount(scan->rs_base.rs_rd);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
if (scan->rs_base.rs_key)
pfree(scan->rs_base.rs_key);
Repair memory leaks that caused CacheCxt to grow without bound. We really ought to fix relcache entry construction so that it does not do so much with CurrentMemoryContext = CacheCxt. As is, relatively harmless leaks in either sequential or index scanning translate to permanent leaks if they occur when called from relcache build. For the moment, however, the path of least resistance is to repair all such leaks...
2000-05-21 04:28:55 +02:00
Make large sequential scans and VACUUMs work in a limited-size "ring" of buffers, rather than blowing out the whole shared-buffer arena. Aside from avoiding cache spoliation, this fixes the problem that VACUUM formerly tended to cause a WAL flush for every page it modified, because we had it hacked to use only a single buffer. Those flushes will now occur only once per ring-ful. The exact ring size, and the threshold for seqscans to switch into the ring usage pattern, remain under debate; but the infrastructure seems done. The key bit of infrastructure is a new optional BufferAccessStrategy object that can be passed to ReadBuffer operations; this replaces the former StrategyHintVacuum API. This patch also changes the buffer usage-count methodology a bit: we now advance usage_count when first pinning a buffer, rather than when last unpinning it. To preserve the behavior that a buffer's lifetime starts to decrease when it's released, the clock sweep code is modified to not decrement usage_count of pinned buffers. Work not done in this commit: teach GiST and GIN indexes to use the vacuum BufferAccessStrategy for vacuum-driven fetches. Original patch by Simon, reworked by Heikki and again by Tom.
2007-05-30 22:12:03 +02:00
if (scan->rs_strategy != NULL)
FreeAccessStrategy(scan->rs_strategy);
Adjust design of per-worker parallel seqscan data struct The design of the data structures which allow storage of the per-worker memory during parallel seq scans were not ideal. The work done in 56788d215 required an additional data structure to allow workers to remember the range of pages that had been allocated to them for processing during a parallel seqscan. That commit added a void pointer field to TableScanDescData to allow heapam to store the per-worker allocation information. However putting the field there made very little sense given that we have AM specific structs for that, e.g. HeapScanDescData. Here we remove the void pointer field from TableScanDescData and add a dedicated field for this purpose to HeapScanDescData. Previously we also allocated memory for this parallel per-worker data for all scans, regardless if it was a parallel scan or not. This was just a wasted allocation for non-parallel scans, so here we make the allocation conditional on the scan being parallel. Also, add previously missing pfree() to free the per-worker data in heap_endscan(). Reported-by: Andres Freund Reviewed-by: Andres Freund Discussion: https://postgr.es/m/20210317023101.anvejcfotwka6gaa@alap3.anarazel.de
2021-03-29 23:17:09 +02:00
if (scan->rs_parallelworkerdata != NULL)
pfree(scan->rs_parallelworkerdata);
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
if (scan->rs_base.rs_flags & SO_TEMP_SNAPSHOT)
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
UnregisterSnapshot(scan->rs_base.rs_snapshot);
Use an MVCC snapshot, rather than SnapshotNow, for catalog scans. SnapshotNow scans have the undesirable property that, in the face of concurrent updates, the scan can fail to see either the old or the new versions of the row. In many cases, we work around this by requiring DDL operations to hold AccessExclusiveLock on the object being modified; in some cases, the existing locking is inadequate and random failures occur as a result. This commit doesn't change anything related to locking, but will hopefully pave the way to allowing lock strength reductions in the future. The major issue has held us back from making this change in the past is that taking an MVCC snapshot is significantly more expensive than using a static special snapshot such as SnapshotNow. However, testing of various worst-case scenarios reveals that this problem is not severe except under fairly extreme workloads. To mitigate those problems, we avoid retaking the MVCC snapshot for each new scan; instead, we take a new snapshot only when invalidation messages have been processed. The catcache machinery already requires that invalidation messages be sent before releasing the related heavyweight lock; else other backends might rely on locally-cached data rather than scanning the catalog at all. Thus, making snapshot reuse dependent on the same guarantees shouldn't break anything that wasn't already subtly broken. Patch by me. Review by Michael Paquier and Andres Freund.
2013-07-02 15:47:01 +02:00
Mega-commit to make heap_open/heap_openr/heap_close take an additional argument specifying the kind of lock to acquire/release (or 'NoLock' to do no lock processing). Ensure that all relations are locked with some appropriate lock level before being examined --- this ensures that relevant shared-inval messages have been processed and should prevent problems caused by concurrent VACUUM. Fix several bugs having to do with mismatched increment/decrement of relation ref count and mismatched heap_open/close (which amounts to the same thing). A bogus ref count on a relation doesn't matter much *unless* a SI Inval message happens to arrive at the wrong time, which is probably why we got away with this sloppiness for so long. Repair missing grab of AccessExclusiveLock in DROP TABLE, ALTER/RENAME TABLE, etc, as noted by Hiroshi. Recommend 'make clean all' after pulling this update; I modified the Relation struct layout slightly. Will post further discussion to pghackers list shortly.
1999-09-18 21:08:25 +02:00
pfree(scan);
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
}
HeapTuple
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
heap_getnext(TableScanDesc sscan, ScanDirection direction)
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
{
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
HeapScanDesc scan = (HeapScanDesc) sscan;
/*
* This is still widely used directly, without going through table AM, so
* add a safety check. It's possible we should, at a later point,
* downgrade this to an assert. The reason for checking the AM routine,
* rather than the AM oid, is that this allows to write regression tests
* that create another AM reusing the heap handler.
*/
if (unlikely(sscan->rs_rd->rd_tableam != GetHeapamTableAmRoutine()))
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
Fix message style Mark one message not for translation, and prefer "cannot" over "may not", per commentary from Robert Haas. Discussion: https://postgr.es/m/20190430145813.GA29872@alvherre.pgsql
2019-06-06 18:18:14 +02:00
errmsg_internal("only heap AM is supported")));
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
Implement streaming mode in ReorderBuffer. Instead of serializing the transaction to disk after reaching the logical_decoding_work_mem limit in memory, we consume the changes we have in memory and invoke stream API methods added by commit 45fdc9738b. However, sometimes if we have incomplete toast or speculative insert we spill to the disk because we can't generate the complete tuple and stream. And, as soon as we get the complete tuple we stream the transaction including the serialized changes. We can do this incremental processing thanks to having assignments (associating subxact with toplevel xacts) in WAL right away, and thanks to logging the invalidation messages at each command end. These features are added by commits 0bead9af48 and c55040ccd0 respectively. Now that we can stream in-progress transactions, the concurrent aborts may cause failures when the output plugin consults catalogs (both system and user-defined). We handle such failures by returning ERRCODE_TRANSACTION_ROLLBACK sqlerrcode from system table scan APIs to the backend or WALSender decoding a specific uncommitted transaction. The decoding logic on the receipt of such a sqlerrcode aborts the decoding of the current transaction and continue with the decoding of other transactions. We have ReorderBufferTXN pointer in each ReorderBufferChange by which we know which xact it belongs to. The output plugin can use this to decide which changes to discard in case of stream_abort_cb (e.g. when a subxact gets discarded). We also provide a new option via SQL APIs to fetch the changes being streamed. Author: Dilip Kumar, Tomas Vondra, Amit Kapila, Nikhil Sontakke Reviewed-by: Amit Kapila, Kuntal Ghosh, Ajin Cherian Tested-by: Neha Sharma, Mahendra Singh Thalor and Ajin Cherian Discussion: https://postgr.es/m/688b0b7f-2f6c-d827-c27b-216a8e3ea700@2ndquadrant.com
2020-08-08 04:04:39 +02:00
/*
* We don't expect direct calls to heap_getnext with valid CheckXidAlive
* for catalog or regular tables. See detailed comments in xact.c where
* these variables are declared. Normally we have such a check at tableam
* level API but this is called from many places so we need to ensure it
* here.
*/
if (unlikely(TransactionIdIsValid(CheckXidAlive) && !bsysscan))
elog(ERROR, "unexpected heap_getnext call during logical decoding");
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/* Note: no locking manipulations needed */
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
if (scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
Cleanup the usage of ScanDirection: use the symbolic names for the possible ScanDirection alternatives rather than magic numbers (-1, 0, 1). Also, use the ScanDirection macros in a few places rather than directly checking whether `dir == ForwardScanDirection' and the like. Per patch from James William Pye. His patch also changed ScanDirection to be a "char" rather than an enum, which I haven't applied.
2006-02-22 00:01:54 +01:00
heapgettup_pagemode(scan, direction,
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
else
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
heapgettup(scan, direction,
scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
if (scan->rs_ctup.t_data == NULL)
Restructure indexscan API (index_beginscan, index_getnext) per yesterday's proposal to pghackers. Also remove unnecessary parameters to heap_beginscan, heap_rescan. I modified pg_proc.h to reflect the new numbers of parameters for the AM interface routines, but did not force an initdb because nothing actually looks at those fields.
2002-05-21 01:51:44 +02:00
return NULL;
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
* if we get here it means we have a new current scan tuple, so point to
* the proper return buffer and return the tuple.
*/
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
pgstat_count_heap_getnext(scan->rs_base.rs_rd);
return &scan->rs_ctup;
}
bool
heap_getnextslot(TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
{
HeapScanDesc scan = (HeapScanDesc) sscan;
/* Note: no locking manipulations needed */
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
else
Don't to predicate lock for analyze scans, refactor scan option passing. Before this commit, when ANALYZE was run on a table and serializable was used (either by virtue of an explicit BEGIN TRANSACTION ISOLATION LEVEL SERIALIZABLE, or default_transaction_isolation being set to serializable) a null pointer dereference lead to a crash. The analyze scan doesn't need a snapshot (nor predicate locking), but before this commit a scan only contained information about being a bitmap or sample scan. Refactor the option passing to the scan_begin callback to use a bitmask instead. Alternatively we could have added a new boolean parameter, but that seems harder to read. Even before this issue various people (Heikki, Tom, Robert) suggested doing so. These changes don't change the scan APIs outside of tableam. The flags argument could be exposed, it's not necessary to fix this problem. Also the wrapper table_beginscan* functions encapsulate most of that complexity. After these changes fixing the bug is trivial, just don't acquire predicate lock for analyze style scans. That was already done for bitmap heap scans. Add an assert that a snapshot is passed when acquiring the predicate lock, so this kind of bug doesn't require running with serializable. Also add a comment about sample scans currently requiring predicate locking the entire relation, that previously wasn't remarked upon. Reported-By: Joe Wildish Author: Andres Freund Discussion: https://postgr.es/m/4EA80A20-E9BF-49F1-9F01-5B66CAB21453@elusive.cx https://postgr.es/m/20190411164947.nkii4gaeilt4bui7@alap3.anarazel.de https://postgr.es/m/20190518203102.g7peu2fianukjuxm@alap3.anarazel.de
2019-05-20 00:10:28 +02:00
heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
if (scan->rs_ctup.t_data == NULL)
{
ExecClearTuple(slot);
return false;
}
/*
* if we get here it means we have a new current scan tuple, so point to
* the proper return buffer and return the tuple.
*/
pgstat_count_heap_getnext(scan->rs_base.rs_rd);
Statistical system views (yet without the config stuff, but it's hard to keep such massive changes in sync with the tree so I need to get it in and work from there now). Jan
2001-06-22 21:16:24 +02:00
tableam: Add and use scan APIs. Too allow table accesses to be not directly dependent on heap, several new abstractions are needed. Specifically: 1) Heap scans need to be generalized into table scans. Do this by introducing TableScanDesc, which will be the "base class" for individual AMs. This contains the AM independent fields from HeapScanDesc. The previous heap_{beginscan,rescan,endscan} et al. have been replaced with a table_ version. There's no direct replacement for heap_getnext(), as that returned a HeapTuple, which is undesirable for a other AMs. Instead there's table_scan_getnextslot(). But note that heap_getnext() lives on, it's still used widely to access catalog tables. This is achieved by new scan_begin, scan_end, scan_rescan, scan_getnextslot callbacks. 2) The portion of parallel scans that's shared between backends need to be able to do so without the user doing per-AM work. To achieve that new parallelscan_{estimate, initialize, reinitialize} callbacks are introduced, which operate on a new ParallelTableScanDesc, which again can be subclassed by AMs. As it is likely that several AMs are going to be block oriented, block oriented callbacks that can be shared between such AMs are provided and used by heap. table_block_parallelscan_{estimate, intiialize, reinitialize} as callbacks, and table_block_parallelscan_{nextpage, init} for use in AMs. These operate on a ParallelBlockTableScanDesc. 3) Index scans need to be able to access tables to return a tuple, and there needs to be state across individual accesses to the heap to store state like buffers. That's now handled by introducing a sort-of-scan IndexFetchTable, which again is intended to be subclassed by individual AMs (for heap IndexFetchHeap). The relevant callbacks for an AM are index_fetch_{end, begin, reset} to create the necessary state, and index_fetch_tuple to retrieve an indexed tuple. Note that index_fetch_tuple implementations need to be smarter than just blindly fetching the tuples for AMs that have optimizations similar to heap's HOT - the currently alive tuple in the update chain needs to be fetched if appropriate. Similar to table_scan_getnextslot(), it's undesirable to continue to return HeapTuples. Thus index_fetch_heap (might want to rename that later) now accepts a slot as an argument. Core code doesn't have a lot of call sites performing index scans without going through the systable_* API (in contrast to loads of heap_getnext calls and working directly with HeapTuples). Index scans now store the result of a search in IndexScanDesc->xs_heaptid, rather than xs_ctup->t_self. As the target is not generally a HeapTuple anymore that seems cleaner. To be able to sensible adapt code to use the above, two further callbacks have been introduced: a) slot_callbacks returns a TupleTableSlotOps* suitable for creating slots capable of holding a tuple of the AMs type. table_slot_callbacks() and table_slot_create() are based upon that, but have additional logic to deal with views, foreign tables, etc. While this change could have been done separately, nearly all the call sites that needed to be adapted for the rest of this commit also would have been needed to be adapted for table_slot_callbacks(), making separation not worthwhile. b) tuple_satisfies_snapshot checks whether the tuple in a slot is currently visible according to a snapshot. That's required as a few places now don't have a buffer + HeapTuple around, but a slot (which in heap's case internally has that information). Additionally a few infrastructure changes were needed: I) SysScanDesc, as used by systable_{beginscan, getnext} et al. now internally uses a slot to keep track of tuples. While systable_getnext() still returns HeapTuples, and will so for the foreseeable future, the index API (see 1) above) now only deals with slots. The remainder, and largest part, of this commit is then adjusting all scans in postgres to use the new APIs. Author: Andres Freund, Haribabu Kommi, Alvaro Herrera Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-11 20:46:41 +01:00
ExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
scan->rs_cbuf);
return true;
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
}
Add TID Range Scans to support efficient scanning ranges of TIDs This adds a new executor node named TID Range Scan. The query planner will generate paths for TID Range scans when quals are discovered on base relations which search for ranges on the table's ctid column. These ranges may be open at either end. For example, WHERE ctid >= '(10,0)'; will return all tuples on page 10 and over. To support this, two new optional callback functions have been added to table AM. scan_set_tidrange is used to set the scan range to just the given range of TIDs. scan_getnextslot_tidrange fetches the next tuple in the given range. For AMs were scanning ranges of TIDs would not make sense, these functions can be set to NULL in the TableAmRoutine. The query planner won't generate TID Range Scan Paths in that case. Author: Edmund Horner, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Tom Lane, Andres Freund, Zhihong Yu Discussion: https://postgr.es/m/CAMyN-kB-nFTkF=VA_JPwFNo08S0d-Yk0F741S2B7LDmYAi8eyA@mail.gmail.com
2021-02-27 10:59:36 +01:00
void
heap_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
ItemPointer maxtid)
{
HeapScanDesc scan = (HeapScanDesc) sscan;
BlockNumber startBlk;
BlockNumber numBlks;
ItemPointerData highestItem;
ItemPointerData lowestItem;
/*
* For relations without any pages, we can simply leave the TID range
* unset. There will be no tuples to scan, therefore no tuples outside
* the given TID range.
*/
if (scan->rs_nblocks == 0)
return;
/*
* Set up some ItemPointers which point to the first and last possible
* tuples in the heap.
*/
ItemPointerSet(&highestItem, scan->rs_nblocks - 1, MaxOffsetNumber);
ItemPointerSet(&lowestItem, 0, FirstOffsetNumber);
/*
* If the given maximum TID is below the highest possible TID in the
* relation, then restrict the range to that, otherwise we scan to the end
* of the relation.
*/
if (ItemPointerCompare(maxtid, &highestItem) < 0)
ItemPointerCopy(maxtid, &highestItem);
/*
* If the given minimum TID is above the lowest possible TID in the
* relation, then restrict the range to only scan for TIDs above that.
*/
if (ItemPointerCompare(mintid, &lowestItem) > 0)
ItemPointerCopy(mintid, &lowestItem);
/*
* Check for an empty range and protect from would be negative results
* from the numBlks calculation below.
*/
if (ItemPointerCompare(&highestItem, &lowestItem) < 0)
{
/* Set an empty range of blocks to scan */
heap_setscanlimits(sscan, 0, 0);
return;
}
/*
* Calculate the first block and the number of blocks we must scan. We
* could be more aggressive here and perform some more validation to try
* and further narrow the scope of blocks to scan by checking if the
Fix typos in comments The changes done in this commit impact comments with no direct user-visible changes, with fixes for incorrect function, variable or structure names. Author: Alexander Lakhin Discussion: https://postgr.es/m/e8c38840-596a-83d6-bd8d-cebc51111572@gmail.com
2023-05-02 05:23:08 +02:00
* lowestItem has an offset above MaxOffsetNumber. In this case, we could
Add TID Range Scans to support efficient scanning ranges of TIDs This adds a new executor node named TID Range Scan. The query planner will generate paths for TID Range scans when quals are discovered on base relations which search for ranges on the table's ctid column. These ranges may be open at either end. For example, WHERE ctid >= '(10,0)'; will return all tuples on page 10 and over. To support this, two new optional callback functions have been added to table AM. scan_set_tidrange is used to set the scan range to just the given range of TIDs. scan_getnextslot_tidrange fetches the next tuple in the given range. For AMs were scanning ranges of TIDs would not make sense, these functions can be set to NULL in the TableAmRoutine. The query planner won't generate TID Range Scan Paths in that case. Author: Edmund Horner, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Tom Lane, Andres Freund, Zhihong Yu Discussion: https://postgr.es/m/CAMyN-kB-nFTkF=VA_JPwFNo08S0d-Yk0F741S2B7LDmYAi8eyA@mail.gmail.com
2021-02-27 10:59:36 +01:00
* advance startBlk by one. Likewise, if highestItem has an offset of 0
* we could scan one fewer blocks. However, such an optimization does not
* seem worth troubling over, currently.
*/
startBlk = ItemPointerGetBlockNumberNoCheck(&lowestItem);
numBlks = ItemPointerGetBlockNumberNoCheck(&highestItem) -
ItemPointerGetBlockNumberNoCheck(&lowestItem) + 1;
/* Set the start block and number of blocks to scan */
heap_setscanlimits(sscan, startBlk, numBlks);
/* Finally, set the TID range in sscan */
ItemPointerCopy(&lowestItem, &sscan->rs_mintid);
ItemPointerCopy(&highestItem, &sscan->rs_maxtid);
}
bool
heap_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
TupleTableSlot *slot)
{
HeapScanDesc scan = (HeapScanDesc) sscan;
ItemPointer mintid = &sscan->rs_mintid;
ItemPointer maxtid = &sscan->rs_maxtid;
/* Note: no locking manipulations needed */
for (;;)
{
if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
else
heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
if (scan->rs_ctup.t_data == NULL)
{
ExecClearTuple(slot);
return false;
}
/*
* heap_set_tidrange will have used heap_setscanlimits to limit the
* range of pages we scan to only ones that can contain the TID range
* we're scanning for. Here we must filter out any tuples from these
Fix various typos and spelling mistakes in code comments Author: Justin Pryzby Discussion: https://postgr.es/m/20220411020336.GB26620@telsasoft.com
2022-04-11 10:49:41 +02:00
* pages that are outside of that range.
Add TID Range Scans to support efficient scanning ranges of TIDs This adds a new executor node named TID Range Scan. The query planner will generate paths for TID Range scans when quals are discovered on base relations which search for ranges on the table's ctid column. These ranges may be open at either end. For example, WHERE ctid >= '(10,0)'; will return all tuples on page 10 and over. To support this, two new optional callback functions have been added to table AM. scan_set_tidrange is used to set the scan range to just the given range of TIDs. scan_getnextslot_tidrange fetches the next tuple in the given range. For AMs were scanning ranges of TIDs would not make sense, these functions can be set to NULL in the TableAmRoutine. The query planner won't generate TID Range Scan Paths in that case. Author: Edmund Horner, David Rowley Reviewed-by: David Rowley, Tomas Vondra, Tom Lane, Andres Freund, Zhihong Yu Discussion: https://postgr.es/m/CAMyN-kB-nFTkF=VA_JPwFNo08S0d-Yk0F741S2B7LDmYAi8eyA@mail.gmail.com
2021-02-27 10:59:36 +01:00
*/
if (ItemPointerCompare(&scan->rs_ctup.t_self, mintid) < 0)
{
ExecClearTuple(slot);
/*
* When scanning backwards, the TIDs will be in descending order.
* Future tuples in this direction will be lower still, so we can
* just return false to indicate there will be no more tuples.
*/
if (ScanDirectionIsBackward(direction))
return false;
continue;
}
/*
* Likewise for the final page, we must filter out TIDs greater than
* maxtid.
*/
if (ItemPointerCompare(&scan->rs_ctup.t_self, maxtid) > 0)
{
ExecClearTuple(slot);
/*
* When scanning forward, the TIDs will be in ascending order.
* Future tuples in this direction will be higher still, so we can
* just return false to indicate there will be no more tuples.
*/
if (ScanDirectionIsForward(direction))
return false;
continue;
}
break;
}
/*
* if we get here it means we have a new current scan tuple, so point to
* the proper return buffer and return the tuple.
*/
pgstat_count_heap_getnext(scan->rs_base.rs_rd);
ExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
return true;
}
Mark index entries "killed" when they are no longer visible to any transaction, so as to avoid returning them out of the index AM. Saves repeated heap_fetch operations on frequently-updated rows. Also detect queries on unique keys (equality to all columns of a unique index), and don't bother continuing scan once we have found first match. Killing is implemented in the btree and hash AMs, but not yet in rtree or gist, because there isn't an equally convenient place to do it in those AMs (the outer amgetnext routine can't do it without re-pinning the index page). Did some small cleanup on APIs of HeapTupleSatisfies, heap_fetch, and index_insert to make this a little easier.
2002-05-24 20:57:57 +02:00
/*
* heap_fetch - retrieve tuple with given tid
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*
Mark index entries "killed" when they are no longer visible to any transaction, so as to avoid returning them out of the index AM. Saves repeated heap_fetch operations on frequently-updated rows. Also detect queries on unique keys (equality to all columns of a unique index), and don't bother continuing scan once we have found first match. Killing is implemented in the btree and hash AMs, but not yet in rtree or gist, because there isn't an equally convenient place to do it in those AMs (the outer amgetnext routine can't do it without re-pinning the index page). Did some small cleanup on APIs of HeapTupleSatisfies, heap_fetch, and index_insert to make this a little easier.
2002-05-24 20:57:57 +02:00
* On entry, tuple->t_self is the TID to fetch. We pin the buffer holding
* the tuple, fill in the remaining fields of *tuple, and check the tuple
* against the specified snapshot.
heap_fetch requires buffer pointer, must be released; heap_getnext no longer returns buffer pointer, can be gotten from scan; descriptor; bootstrap can create multi-key indexes; pg_procname index now is multi-key index; oidint2, oidint4, oidname are gone (must be removed from regression tests); use System Cache rather than sequential scan in many places; heap_modifytuple no longer takes buffer parameter; remove unused buffer parameter in a few other functions; oid8 is not index-able; remove some use of single-character variable names; cleanup Buffer variables usage and scan descriptor looping; cleaned up allocation and freeing of tuples; 18k lines of diff;
1998-08-19 04:04:17 +02:00
*
Wups, managed to break ANALYZE with one aspect of that heap_fetch change.
2002-05-24 21:52:43 +02:00
* If successful (tuple found and passes snapshot time qual), then *userbuf
Change TRUE/FALSE to true/false The lower case spellings are C and C++ standard and are used in most parts of the PostgreSQL sources. The upper case spellings are only used in some files/modules. So standardize on the standard spellings. The APIs for ICU, Perl, and Windows define their own TRUE and FALSE, so those are left as is when using those APIs. In code comments, we use the lower-case spelling for the C concepts and keep the upper-case spelling for the SQL concepts. Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
2017-08-16 06:22:32 +02:00
* is set to the buffer holding the tuple and true is returned. The caller
Wups, managed to break ANALYZE with one aspect of that heap_fetch change.
2002-05-24 21:52:43 +02:00
* must unpin the buffer when done with the tuple.
Restructure indexscan API (index_beginscan, index_getnext) per yesterday's proposal to pghackers. Also remove unnecessary parameters to heap_beginscan, heap_rescan. I modified pg_proc.h to reflect the new numbers of parameters for the AM interface routines, but did not force an initdb because nothing actually looks at those fields.
2002-05-21 01:51:44 +02:00
*
Make heap_fetch API more consistent by having the buffer remain pinned in all cases when keep_buf = true. This allows ANALYZE's inner loop to use heap_release_fetch, which saves multiple buffer lookups for the same page and avoids overestimation of cost by the vacuum cost mechanism.
2004-10-26 18:05:03 +02:00
* If the tuple is not found (ie, item number references a deleted slot),
Prevent access to no-longer-pinned buffer in heapam_tuple_lock(). heap_fetch() used to have a "keep_buf" parameter that told it to return ownership of the buffer pin to the caller after finding that the requested tuple TID exists but is invisible to the specified snapshot. This was thoughtlessly removed in commit 5db6df0c0, which broke heapam_tuple_lock() (formerly EvalPlanQualFetch) because that function needs to do more accesses to the tuple even if it's invisible. The net effect is that we would continue to touch the page for a microsecond or two after releasing pin on the buffer. Usually no harm would result; but if a different session decided to defragment the page concurrently, we could see garbage data and mistakenly conclude that there's no newer tuple version to chain up to. (It's hard to say whether this has happened in the field. The bug was actually found thanks to a later change that allowed valgrind to detect accesses to non-pinned buffers.) The most reasonable way to fix this is to reintroduce keep_buf, although I made it behave slightly differently: buffer ownership is passed back only if there is a valid tuple at the requested TID. In HEAD, we can just add the parameter back to heap_fetch(). To avoid an API break in the back branches, introduce an additional function heap_fetch_extended() in those branches. In HEAD there is an additional, less obvious API change: tuple->t_data will be set to NULL in all cases where buffer ownership is not returned, in particular when the tuple exists but fails the time qual (and !keep_buf). This is to defend against any other callers attempting to access non-pinned buffers. We concluded that making that change in back branches would be more likely to introduce problems than cure any. In passing, remove a comment about heap_fetch that was obsoleted by 9a8ee1dc6. Per bug #17462 from Daniil Anisimov. Back-patch to v12 where the bug was introduced. Discussion: https://postgr.es/m/17462-9c98a0f00df9bd36@postgresql.org
2022-04-13 19:35:02 +02:00
* then tuple->t_data is set to NULL, *userbuf is set to InvalidBuffer,
* and false is returned.
Wups, managed to break ANALYZE with one aspect of that heap_fetch change.
2002-05-24 21:52:43 +02:00
*
Prevent access to no-longer-pinned buffer in heapam_tuple_lock(). heap_fetch() used to have a "keep_buf" parameter that told it to return ownership of the buffer pin to the caller after finding that the requested tuple TID exists but is invisible to the specified snapshot. This was thoughtlessly removed in commit 5db6df0c0, which broke heapam_tuple_lock() (formerly EvalPlanQualFetch) because that function needs to do more accesses to the tuple even if it's invisible. The net effect is that we would continue to touch the page for a microsecond or two after releasing pin on the buffer. Usually no harm would result; but if a different session decided to defragment the page concurrently, we could see garbage data and mistakenly conclude that there's no newer tuple version to chain up to. (It's hard to say whether this has happened in the field. The bug was actually found thanks to a later change that allowed valgrind to detect accesses to non-pinned buffers.) The most reasonable way to fix this is to reintroduce keep_buf, although I made it behave slightly differently: buffer ownership is passed back only if there is a valid tuple at the requested TID. In HEAD, we can just add the parameter back to heap_fetch(). To avoid an API break in the back branches, introduce an additional function heap_fetch_extended() in those branches. In HEAD there is an additional, less obvious API change: tuple->t_data will be set to NULL in all cases where buffer ownership is not returned, in particular when the tuple exists but fails the time qual (and !keep_buf). This is to defend against any other callers attempting to access non-pinned buffers. We concluded that making that change in back branches would be more likely to introduce problems than cure any. In passing, remove a comment about heap_fetch that was obsoleted by 9a8ee1dc6. Per bug #17462 from Daniil Anisimov. Back-patch to v12 where the bug was introduced. Discussion: https://postgr.es/m/17462-9c98a0f00df9bd36@postgresql.org
2022-04-13 19:35:02 +02:00
* If the tuple is found but fails the time qual check, then the behavior
* depends on the keep_buf parameter. If keep_buf is false, the results
* are the same as for the tuple-not-found case. If keep_buf is true,
* then tuple->t_data and *userbuf are returned as for the success case,
* and again the caller must unpin the buffer; but false is returned.
Make heap_fetch API more consistent by having the buffer remain pinned in all cases when keep_buf = true. This allows ANALYZE's inner loop to use heap_release_fetch, which saves multiple buffer lookups for the same page and avoids overestimation of cost by the vacuum cost mechanism.
2004-10-26 18:05:03 +02:00
*
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
* heap_fetch does not follow HOT chains: only the exact TID requested will
* be fetched.
*
Error message editing in backend/access.
2003-07-21 22:29:40 +02:00
* It is somewhat inconsistent that we ereport() on invalid block number but
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
* return false on invalid item number. There are a couple of reasons though.
* One is that the caller can relatively easily check the block number for
* validity, but cannot check the item number without reading the page
* himself. Another is that when we are following a t_ctid link, we can be
* reasonably confident that the page number is valid (since VACUUM shouldn't
* truncate off the destination page without having killed the referencing
* tuple first), but the item number might well not be good.
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*/
Mark index entries "killed" when they are no longer visible to any transaction, so as to avoid returning them out of the index AM. Saves repeated heap_fetch operations on frequently-updated rows. Also detect queries on unique keys (equality to all columns of a unique index), and don't bother continuing scan once we have found first match. Killing is implemented in the btree and hash AMs, but not yet in rtree or gist, because there isn't an equally convenient place to do it in those AMs (the outer amgetnext routine can't do it without re-pinning the index page). Did some small cleanup on APIs of HeapTupleSatisfies, heap_fetch, and index_insert to make this a little easier.
2002-05-24 20:57:57 +02:00
bool
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
heap_fetch(Relation relation,
Use Snapshot in heap access methods.
1998-07-27 21:38:40 +02:00
Snapshot snapshot,
New HeapTuple structure/interface.
1998-11-27 20:52:36 +01:00
HeapTuple tuple,
Prevent access to no-longer-pinned buffer in heapam_tuple_lock(). heap_fetch() used to have a "keep_buf" parameter that told it to return ownership of the buffer pin to the caller after finding that the requested tuple TID exists but is invisible to the specified snapshot. This was thoughtlessly removed in commit 5db6df0c0, which broke heapam_tuple_lock() (formerly EvalPlanQualFetch) because that function needs to do more accesses to the tuple even if it's invisible. The net effect is that we would continue to touch the page for a microsecond or two after releasing pin on the buffer. Usually no harm would result; but if a different session decided to defragment the page concurrently, we could see garbage data and mistakenly conclude that there's no newer tuple version to chain up to. (It's hard to say whether this has happened in the field. The bug was actually found thanks to a later change that allowed valgrind to detect accesses to non-pinned buffers.) The most reasonable way to fix this is to reintroduce keep_buf, although I made it behave slightly differently: buffer ownership is passed back only if there is a valid tuple at the requested TID. In HEAD, we can just add the parameter back to heap_fetch(). To avoid an API break in the back branches, introduce an additional function heap_fetch_extended() in those branches. In HEAD there is an additional, less obvious API change: tuple->t_data will be set to NULL in all cases where buffer ownership is not returned, in particular when the tuple exists but fails the time qual (and !keep_buf). This is to defend against any other callers attempting to access non-pinned buffers. We concluded that making that change in back branches would be more likely to introduce problems than cure any. In passing, remove a comment about heap_fetch that was obsoleted by 9a8ee1dc6. Per bug #17462 from Daniil Anisimov. Back-patch to v12 where the bug was introduced. Discussion: https://postgr.es/m/17462-9c98a0f00df9bd36@postgresql.org
2022-04-13 19:35:02 +02:00
Buffer *userbuf,
bool keep_buf)
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
{
Mark index entries "killed" when they are no longer visible to any transaction, so as to avoid returning them out of the index AM. Saves repeated heap_fetch operations on frequently-updated rows. Also detect queries on unique keys (equality to all columns of a unique index), and don't bother continuing scan once we have found first match. Killing is implemented in the btree and hash AMs, but not yet in rtree or gist, because there isn't an equally convenient place to do it in those AMs (the outer amgetnext routine can't do it without re-pinning the index page). Did some small cleanup on APIs of HeapTupleSatisfies, heap_fetch, and index_insert to make this a little easier.
2002-05-24 20:57:57 +02:00
ItemPointer tid = &(tuple->t_self);
New HeapTuple structure/interface.
1998-11-27 20:52:36 +01:00
ItemId lp;
Buffer buffer;
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
Page page;
New HeapTuple structure/interface.
1998-11-27 20:52:36 +01:00
OffsetNumber offnum;
Mark index entries "killed" when they are no longer visible to any transaction, so as to avoid returning them out of the index AM. Saves repeated heap_fetch operations on frequently-updated rows. Also detect queries on unique keys (equality to all columns of a unique index), and don't bother continuing scan once we have found first match. Killing is implemented in the btree and hash AMs, but not yet in rtree or gist, because there isn't an equally convenient place to do it in those AMs (the outer amgetnext routine can't do it without re-pinning the index page). Did some small cleanup on APIs of HeapTupleSatisfies, heap_fetch, and index_insert to make this a little easier.
2002-05-24 20:57:57 +02:00
bool valid;
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
Remove heap_release_fetch, which is no longer used anywhere; this simplifies heap_fetch a little.
2008-04-03 19:12:27 +02:00
* Fetch and pin the appropriate page of the relation.
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*/
Remove heap_release_fetch, which is no longer used anywhere; this simplifies heap_fetch a little.
2008-04-03 19:12:27 +02:00
buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Mark index entries "killed" when they are no longer visible to any transaction, so as to avoid returning them out of the index AM. Saves repeated heap_fetch operations on frequently-updated rows. Also detect queries on unique keys (equality to all columns of a unique index), and don't bother continuing scan once we have found first match. Killing is implemented in the btree and hash AMs, but not yet in rtree or gist, because there isn't an equally convenient place to do it in those AMs (the outer amgetnext routine can't do it without re-pinning the index page). Did some small cleanup on APIs of HeapTupleSatisfies, heap_fetch, and index_insert to make this a little easier.
2002-05-24 20:57:57 +02:00
/*
* Need share lock on buffer to examine tuple commit status.
*/
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
LockBuffer(buffer, BUFFER_LOCK_SHARE);
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(buffer);
TestForOldSnapshot(snapshot, relation, page);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
First steps towards index scans with heap access decoupled from index access: define new index access method functions 'amgetmulti' that can fetch multiple TIDs per call. (The functions exist but are totally untested as yet.) Since I was modifying pg_am anyway, remove the no-longer-needed 'rel' parameter from amcostestimate functions, and also remove the vestigial amowner column that was creating useless work for Alvaro's shared-object-dependencies project. Initdb forced due to changes in pg_am.
2005-03-28 01:53:05 +02:00
* We'd better check for out-of-range offnum in case of VACUUM since the
* TID was obtained.
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*/
offnum = ItemPointerGetOffsetNumber(tid);
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
First steps towards index scans with heap access decoupled from index access: define new index access method functions 'amgetmulti' that can fetch multiple TIDs per call. (The functions exist but are totally untested as yet.) Since I was modifying pg_am anyway, remove the no-longer-needed 'rel' parameter from amcostestimate functions, and also remove the vestigial amowner column that was creating useless work for Alvaro's shared-object-dependencies project. Initdb forced due to changes in pg_am.
2005-03-28 01:53:05 +02:00
{
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
ReleaseBuffer(buffer);
*userbuf = InvalidBuffer;
First steps towards index scans with heap access decoupled from index access: define new index access method functions 'amgetmulti' that can fetch multiple TIDs per call. (The functions exist but are totally untested as yet.) Since I was modifying pg_am anyway, remove the no-longer-needed 'rel' parameter from amcostestimate functions, and also remove the vestigial amowner column that was creating useless work for Alvaro's shared-object-dependencies project. Initdb forced due to changes in pg_am.
2005-03-28 01:53:05 +02:00
tuple->t_data = NULL;
return false;
}
/*
* get the item line pointer corresponding to the requested tid
*/
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
lp = PageGetItemId(page, offnum);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
First steps towards index scans with heap access decoupled from index access: define new index access method functions 'amgetmulti' that can fetch multiple TIDs per call. (The functions exist but are totally untested as yet.) Since I was modifying pg_am anyway, remove the no-longer-needed 'rel' parameter from amcostestimate functions, and also remove the vestigial amowner column that was creating useless work for Alvaro's shared-object-dependencies project. Initdb forced due to changes in pg_am.
2005-03-28 01:53:05 +02:00
* Must check for deleted tuple.
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*/
Redefine the lp_flags field of item pointers as having four states, rather than two independent bits (one of which was never used in heap pages anyway, or at least hadn't been in a very long time). This gives us flexibility to add the HOT notions of redirected and dead item pointers without requiring anything so klugy as magic values of lp_off and lp_len. The state values are chosen so that for the states currently in use (pre-HOT) there is no change in the physical representation.
2007-09-13 00:10:26 +02:00
if (!ItemIdIsNormal(lp))
Tid access method feature from Hiroshi Inoue, Inoue@tpf.co.jp
1999-11-23 21:07:06 +01:00
{
unlock buffer before releasing in heap_insert +unlock buffer in heap_fetch under !ItemIdIsUsed().
2000-07-04 03:39:24 +02:00
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
ReleaseBuffer(buffer);
*userbuf = InvalidBuffer;
Wups, managed to break ANALYZE with one aspect of that heap_fetch change.
2002-05-24 21:52:43 +02:00
tuple->t_data = NULL;
return false;
Tid access method feature from Hiroshi Inoue, Inoue@tpf.co.jp
1999-11-23 21:07:06 +01:00
}
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Mark index entries "killed" when they are no longer visible to any transaction, so as to avoid returning them out of the index AM. Saves repeated heap_fetch operations on frequently-updated rows. Also detect queries on unique keys (equality to all columns of a unique index), and don't bother continuing scan once we have found first match. Killing is implemented in the btree and hash AMs, but not yet in rtree or gist, because there isn't an equally convenient place to do it in those AMs (the outer amgetnext routine can't do it without re-pinning the index page). Did some small cleanup on APIs of HeapTupleSatisfies, heap_fetch, and index_insert to make this a little easier.
2002-05-24 20:57:57 +02:00
/*
* fill in *tuple fields
*/
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
New HeapTuple structure/interface.
1998-11-27 20:52:36 +01:00
tuple->t_len = ItemIdGetLength(lp);
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
tuple->t_tableOid = RelationGetRelid(relation);
New HeapTuple structure/interface.
1998-11-27 20:52:36 +01:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
/*
Rephrase references to "time qualification". Now that the relevant code has, for other reasons, moved out of tqual.[ch], it seems time to refer to visiblity rather than time qualification. Author: Andres Freund Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-01-22 02:03:15 +01:00
* check tuple visibility, then release lock
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*/
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
valid = HeapTupleSatisfiesVisibility(tuple, snapshot, buffer);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
if (valid)
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
PredicateLockTID(relation, &(tuple->t_self), snapshot,
HeapTupleHeaderGetXmin(tuple->t_data));
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
HeapCheckForSerializableConflictOut(valid, relation, tuple, buffer, snapshot);
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
You must hold a lock on the heap page when you call CheckForSerializableConflictOut(), because it can set hint bits. YAMAMOTO Takashi
2011-03-04 14:37:09 +01:00
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
Mark index entries "killed" when they are no longer visible to any transaction, so as to avoid returning them out of the index AM. Saves repeated heap_fetch operations on frequently-updated rows. Also detect queries on unique keys (equality to all columns of a unique index), and don't bother continuing scan once we have found first match. Killing is implemented in the btree and hash AMs, but not yet in rtree or gist, because there isn't an equally convenient place to do it in those AMs (the outer amgetnext routine can't do it without re-pinning the index page). Did some small cleanup on APIs of HeapTupleSatisfies, heap_fetch, and index_insert to make this a little easier.
2002-05-24 20:57:57 +02:00
if (valid)
Several changes here, not very related but touching some of the same files. * Buffer refcount cleanup (per my "progress report" to pghackers, 9/22). * Add links to backend PROC structs to sinval's array of per-backend info, and use these links for routines that need to check the state of all backends (rather than the slow, complicated search of the ShmemIndex hashtable that was used before). Add databaseOID to PROC structs. * Use this to implement an interlock that prevents DESTROY DATABASE of a database containing running backends. (It's a little tricky to prevent a concurrently-starting backend from getting in there, since the new backend is not able to lock anything at the time it tries to look up its database in pg_database. My solution is to recheck that the DB is OK at the end of InitPostgres. It may not be a 100% solution, but it's a lot better than no interlock at all...) * In ALTER TABLE RENAME, flush buffers for the relation before doing the rename of the physical files, to ensure we don't get failures later from mdblindwrt(). * Update TRUNCATE patch so that it actually compiles against current sources :-(. You should do "make clean all" after pulling these changes.
1999-09-24 02:25:33 +02:00
{
/*
* All checks passed, so return the tuple as valid. Caller is now
* responsible for releasing the buffer.
*/
*userbuf = buffer;
Statistical system views (yet without the config stuff, but it's hard to keep such massive changes in sync with the tree so I need to get it in and work from there now). Jan
2001-06-22 21:16:24 +02:00
Mark index entries "killed" when they are no longer visible to any transaction, so as to avoid returning them out of the index AM. Saves repeated heap_fetch operations on frequently-updated rows. Also detect queries on unique keys (equality to all columns of a unique index), and don't bother continuing scan once we have found first match. Killing is implemented in the btree and hash AMs, but not yet in rtree or gist, because there isn't an equally convenient place to do it in those AMs (the outer amgetnext routine can't do it without re-pinning the index page). Did some small cleanup on APIs of HeapTupleSatisfies, heap_fetch, and index_insert to make this a little easier.
2002-05-24 20:57:57 +02:00
return true;
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
}
Mark index entries "killed" when they are no longer visible to any transaction, so as to avoid returning them out of the index AM. Saves repeated heap_fetch operations on frequently-updated rows. Also detect queries on unique keys (equality to all columns of a unique index), and don't bother continuing scan once we have found first match. Killing is implemented in the btree and hash AMs, but not yet in rtree or gist, because there isn't an equally convenient place to do it in those AMs (the outer amgetnext routine can't do it without re-pinning the index page). Did some small cleanup on APIs of HeapTupleSatisfies, heap_fetch, and index_insert to make this a little easier.
2002-05-24 20:57:57 +02:00
Prevent access to no-longer-pinned buffer in heapam_tuple_lock(). heap_fetch() used to have a "keep_buf" parameter that told it to return ownership of the buffer pin to the caller after finding that the requested tuple TID exists but is invisible to the specified snapshot. This was thoughtlessly removed in commit 5db6df0c0, which broke heapam_tuple_lock() (formerly EvalPlanQualFetch) because that function needs to do more accesses to the tuple even if it's invisible. The net effect is that we would continue to touch the page for a microsecond or two after releasing pin on the buffer. Usually no harm would result; but if a different session decided to defragment the page concurrently, we could see garbage data and mistakenly conclude that there's no newer tuple version to chain up to. (It's hard to say whether this has happened in the field. The bug was actually found thanks to a later change that allowed valgrind to detect accesses to non-pinned buffers.) The most reasonable way to fix this is to reintroduce keep_buf, although I made it behave slightly differently: buffer ownership is passed back only if there is a valid tuple at the requested TID. In HEAD, we can just add the parameter back to heap_fetch(). To avoid an API break in the back branches, introduce an additional function heap_fetch_extended() in those branches. In HEAD there is an additional, less obvious API change: tuple->t_data will be set to NULL in all cases where buffer ownership is not returned, in particular when the tuple exists but fails the time qual (and !keep_buf). This is to defend against any other callers attempting to access non-pinned buffers. We concluded that making that change in back branches would be more likely to introduce problems than cure any. In passing, remove a comment about heap_fetch that was obsoleted by 9a8ee1dc6. Per bug #17462 from Daniil Anisimov. Back-patch to v12 where the bug was introduced. Discussion: https://postgr.es/m/17462-9c98a0f00df9bd36@postgresql.org
2022-04-13 19:35:02 +02:00
/* Tuple failed time qual, but maybe caller wants to see it anyway. */
if (keep_buf)
*userbuf = buffer;
else
{
ReleaseBuffer(buffer);
*userbuf = InvalidBuffer;
tuple->t_data = NULL;
}
Mark index entries "killed" when they are no longer visible to any transaction, so as to avoid returning them out of the index AM. Saves repeated heap_fetch operations on frequently-updated rows. Also detect queries on unique keys (equality to all columns of a unique index), and don't bother continuing scan once we have found first match. Killing is implemented in the btree and hash AMs, but not yet in rtree or gist, because there isn't an equally convenient place to do it in those AMs (the outer amgetnext routine can't do it without re-pinning the index page). Did some small cleanup on APIs of HeapTupleSatisfies, heap_fetch, and index_insert to make this a little easier.
2002-05-24 20:57:57 +02:00
return false;
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
}
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
/*
* heap_hot_search_buffer - search HOT chain for tuple satisfying snapshot
*
* On entry, *tid is the TID of a tuple (either a simple tuple, or the root
* of a HOT chain), and buffer is the buffer holding this tuple. We search
* for the first chain member satisfying the given snapshot. If one is
* found, we update *tid to reference that tuple's offset number, and
Change TRUE/FALSE to true/false The lower case spellings are C and C++ standard and are used in most parts of the PostgreSQL sources. The upper case spellings are only used in some files/modules. So standardize on the standard spellings. The APIs for ICU, Perl, and Windows define their own TRUE and FALSE, so those are left as is when using those APIs. In code comments, we use the lower-case spelling for the C concepts and keep the upper-case spelling for the SQL concepts. Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
2017-08-16 06:22:32 +02:00
* return true. If no match, return false without modifying *tid.
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
*
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
* heapTuple is a caller-supplied buffer. When a match is found, we return
* the tuple here, in addition to updating *tid. If no match is found, the
* contents of this buffer on return are undefined.
*
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
* If all_dead is not NULL, we check non-visible tuples to see if they are
Change TRUE/FALSE to true/false The lower case spellings are C and C++ standard and are used in most parts of the PostgreSQL sources. The upper case spellings are only used in some files/modules. So standardize on the standard spellings. The APIs for ICU, Perl, and Windows define their own TRUE and FALSE, so those are left as is when using those APIs. In code comments, we use the lower-case spelling for the C concepts and keep the upper-case spelling for the SQL concepts. Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
2017-08-16 06:22:32 +02:00
* globally dead; *all_dead is set true if all members of the HOT chain
* are vacuumable, false if not.
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
*
* Unlike heap_fetch, the caller must already have pin and (at least) share
Prevent access to no-longer-pinned buffer in heapam_tuple_lock(). heap_fetch() used to have a "keep_buf" parameter that told it to return ownership of the buffer pin to the caller after finding that the requested tuple TID exists but is invisible to the specified snapshot. This was thoughtlessly removed in commit 5db6df0c0, which broke heapam_tuple_lock() (formerly EvalPlanQualFetch) because that function needs to do more accesses to the tuple even if it's invisible. The net effect is that we would continue to touch the page for a microsecond or two after releasing pin on the buffer. Usually no harm would result; but if a different session decided to defragment the page concurrently, we could see garbage data and mistakenly conclude that there's no newer tuple version to chain up to. (It's hard to say whether this has happened in the field. The bug was actually found thanks to a later change that allowed valgrind to detect accesses to non-pinned buffers.) The most reasonable way to fix this is to reintroduce keep_buf, although I made it behave slightly differently: buffer ownership is passed back only if there is a valid tuple at the requested TID. In HEAD, we can just add the parameter back to heap_fetch(). To avoid an API break in the back branches, introduce an additional function heap_fetch_extended() in those branches. In HEAD there is an additional, less obvious API change: tuple->t_data will be set to NULL in all cases where buffer ownership is not returned, in particular when the tuple exists but fails the time qual (and !keep_buf). This is to defend against any other callers attempting to access non-pinned buffers. We concluded that making that change in back branches would be more likely to introduce problems than cure any. In passing, remove a comment about heap_fetch that was obsoleted by 9a8ee1dc6. Per bug #17462 from Daniil Anisimov. Back-patch to v12 where the bug was introduced. Discussion: https://postgr.es/m/17462-9c98a0f00df9bd36@postgresql.org
2022-04-13 19:35:02 +02:00
* lock on the buffer; it is still pinned/locked at exit.
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
*/
bool
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
heap_hot_search_buffer(ItemPointer tid, Relation relation, Buffer buffer,
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
Snapshot snapshot, HeapTuple heapTuple,
bool *all_dead, bool first_call)
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
{
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
Page page = BufferGetPage(buffer);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
TransactionId prev_xmax = InvalidTransactionId;
Fix predicate-locking of HOT updated rows. In serializable mode, heap_hot_search_buffer() incorrectly acquired a predicate lock on the root tuple, not the returned tuple that satisfied the visibility checks. As explained in README-SSI, the predicate lock does not need to be copied or extended to other tuple versions, but for that to work, the correct, visible, tuple version must be locked in the first place. The original SSI commit had this bug in it, but it was fixed back in 2013, in commit 81fbbfe335. But unfortunately, it was reintroduced a few months later in commit b89e151054. Wising up from that, add a regression test to cover this, so that it doesn't get reintroduced again. Also, move the code that sets 't_self', so that it happens at the same time that the other HeapTuple fields are set, to make it more clear that all the code in the loop operate on the "current" tuple in the chain, not the root tuple. Bug spotted by Andres Freund, analysis and original fix by Thomas Munro, test case and some additional changes to the fix by Heikki Linnakangas. Backpatch to all supported versions (9.4). Discussion: https://www.postgresql.org/message-id/20190731210630.nqhszuktygwftjty%40alap3.anarazel.de
2019-08-07 11:40:49 +02:00
BlockNumber blkno;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
OffsetNumber offnum;
bool at_chain_start;
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
bool valid;
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
bool skip;
snapshot scalability: Don't compute global horizons while building snapshots. To make GetSnapshotData() more scalable, it cannot not look at at each proc's xmin: While snapshot contents do not need to change whenever a read-only transaction commits or a snapshot is released, a proc's xmin is modified in those cases. The frequency of xmin modifications leads to, particularly on higher core count systems, many cache misses inside GetSnapshotData(), despite the data underlying a snapshot not changing. That is the most significant source of GetSnapshotData() scaling poorly on larger systems. Without accessing xmins, GetSnapshotData() cannot calculate accurate horizons / thresholds as it has so far. But we don't really have to: The horizons don't actually change that much between GetSnapshotData() calls. Nor are the horizons actually used every time a snapshot is built. The trick this commit introduces is to delay computation of accurate horizons until there use and using horizon boundaries to determine whether accurate horizons need to be computed. The use of RecentGlobal[Data]Xmin to decide whether a row version could be removed has been replaces with new GlobalVisTest* functions. These use two thresholds to determine whether a row can be pruned: 1) definitely_needed, indicating that rows deleted by XIDs >= definitely_needed are definitely still visible. 2) maybe_needed, indicating that rows deleted by XIDs < maybe_needed can definitely be removed GetSnapshotData() updates definitely_needed to be the xmin of the computed snapshot. When testing whether a row can be removed (with GlobalVisTestIsRemovableXid()) and the tested XID falls in between the two (i.e. XID >= maybe_needed && XID < definitely_needed) the boundaries can be recomputed to be more accurate. As it is not cheap to compute accurate boundaries, we limit the number of times that happens in short succession. As the boundaries used by GlobalVisTestIsRemovableXid() are never reset (with maybe_needed updated by GetSnapshotData()), it is likely that further test can benefit from an earlier computation of accurate horizons. To avoid regressing performance when old_snapshot_threshold is set (as that requires an accurate horizon to be computed), heap_page_prune_opt() doesn't unconditionally call TransactionIdLimitedForOldSnapshots() anymore. Both the computation of the limited horizon, and the triggering of errors (with SetOldSnapshotThresholdTimestamp()) is now only done when necessary to remove tuples. This commit just removes the accesses to PGXACT->xmin from GetSnapshotData(), but other members of PGXACT residing in the same cache line are accessed. Therefore this in itself does not result in a significant improvement. Subsequent commits will take advantage of the fact that GetSnapshotData() now does not need to access xmins anymore. Note: This contains a workaround in heap_page_prune_opt() to keep the snapshot_too_old tests working. While that workaround is ugly, the tests currently are not meaningful, and it seems best to address them separately. Author: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Reviewed-By: Thomas Munro <thomas.munro@gmail.com> Reviewed-By: David Rowley <dgrowleyml@gmail.com> Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
2020-08-13 01:03:49 +02:00
GlobalVisState *vistest = NULL;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
/* If this is not the first call, previous call returned a (live!) tuple */
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
if (all_dead)
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
*all_dead = first_call;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
Fix predicate-locking of HOT updated rows. In serializable mode, heap_hot_search_buffer() incorrectly acquired a predicate lock on the root tuple, not the returned tuple that satisfied the visibility checks. As explained in README-SSI, the predicate lock does not need to be copied or extended to other tuple versions, but for that to work, the correct, visible, tuple version must be locked in the first place. The original SSI commit had this bug in it, but it was fixed back in 2013, in commit 81fbbfe335. But unfortunately, it was reintroduced a few months later in commit b89e151054. Wising up from that, add a regression test to cover this, so that it doesn't get reintroduced again. Also, move the code that sets 't_self', so that it happens at the same time that the other HeapTuple fields are set, to make it more clear that all the code in the loop operate on the "current" tuple in the chain, not the root tuple. Bug spotted by Andres Freund, analysis and original fix by Thomas Munro, test case and some additional changes to the fix by Heikki Linnakangas. Backpatch to all supported versions (9.4). Discussion: https://www.postgresql.org/message-id/20190731210630.nqhszuktygwftjty%40alap3.anarazel.de
2019-08-07 11:40:49 +02:00
blkno = ItemPointerGetBlockNumber(tid);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
offnum = ItemPointerGetOffsetNumber(tid);
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
at_chain_start = first_call;
skip = !first_call;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
snapshot scalability: Don't compute global horizons while building snapshots. To make GetSnapshotData() more scalable, it cannot not look at at each proc's xmin: While snapshot contents do not need to change whenever a read-only transaction commits or a snapshot is released, a proc's xmin is modified in those cases. The frequency of xmin modifications leads to, particularly on higher core count systems, many cache misses inside GetSnapshotData(), despite the data underlying a snapshot not changing. That is the most significant source of GetSnapshotData() scaling poorly on larger systems. Without accessing xmins, GetSnapshotData() cannot calculate accurate horizons / thresholds as it has so far. But we don't really have to: The horizons don't actually change that much between GetSnapshotData() calls. Nor are the horizons actually used every time a snapshot is built. The trick this commit introduces is to delay computation of accurate horizons until there use and using horizon boundaries to determine whether accurate horizons need to be computed. The use of RecentGlobal[Data]Xmin to decide whether a row version could be removed has been replaces with new GlobalVisTest* functions. These use two thresholds to determine whether a row can be pruned: 1) definitely_needed, indicating that rows deleted by XIDs >= definitely_needed are definitely still visible. 2) maybe_needed, indicating that rows deleted by XIDs < maybe_needed can definitely be removed GetSnapshotData() updates definitely_needed to be the xmin of the computed snapshot. When testing whether a row can be removed (with GlobalVisTestIsRemovableXid()) and the tested XID falls in between the two (i.e. XID >= maybe_needed && XID < definitely_needed) the boundaries can be recomputed to be more accurate. As it is not cheap to compute accurate boundaries, we limit the number of times that happens in short succession. As the boundaries used by GlobalVisTestIsRemovableXid() are never reset (with maybe_needed updated by GetSnapshotData()), it is likely that further test can benefit from an earlier computation of accurate horizons. To avoid regressing performance when old_snapshot_threshold is set (as that requires an accurate horizon to be computed), heap_page_prune_opt() doesn't unconditionally call TransactionIdLimitedForOldSnapshots() anymore. Both the computation of the limited horizon, and the triggering of errors (with SetOldSnapshotThresholdTimestamp()) is now only done when necessary to remove tuples. This commit just removes the accesses to PGXACT->xmin from GetSnapshotData(), but other members of PGXACT residing in the same cache line are accessed. Therefore this in itself does not result in a significant improvement. Subsequent commits will take advantage of the fact that GetSnapshotData() now does not need to access xmins anymore. Note: This contains a workaround in heap_page_prune_opt() to keep the snapshot_too_old tests working. While that workaround is ugly, the tests currently are not meaningful, and it seems best to address them separately. Author: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Reviewed-By: Thomas Munro <thomas.munro@gmail.com> Reviewed-By: David Rowley <dgrowleyml@gmail.com> Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
2020-08-13 01:03:49 +02:00
/* XXX: we should assert that a snapshot is pushed or registered */
Assert(TransactionIdIsValid(RecentXmin));
Fix predicate-locking of HOT updated rows. In serializable mode, heap_hot_search_buffer() incorrectly acquired a predicate lock on the root tuple, not the returned tuple that satisfied the visibility checks. As explained in README-SSI, the predicate lock does not need to be copied or extended to other tuple versions, but for that to work, the correct, visible, tuple version must be locked in the first place. The original SSI commit had this bug in it, but it was fixed back in 2013, in commit 81fbbfe335. But unfortunately, it was reintroduced a few months later in commit b89e151054. Wising up from that, add a regression test to cover this, so that it doesn't get reintroduced again. Also, move the code that sets 't_self', so that it happens at the same time that the other HeapTuple fields are set, to make it more clear that all the code in the loop operate on the "current" tuple in the chain, not the root tuple. Bug spotted by Andres Freund, analysis and original fix by Thomas Munro, test case and some additional changes to the fix by Heikki Linnakangas. Backpatch to all supported versions (9.4). Discussion: https://www.postgresql.org/message-id/20190731210630.nqhszuktygwftjty%40alap3.anarazel.de
2019-08-07 11:40:49 +02:00
Assert(BufferGetBlockNumber(buffer) == blkno);
Fix bugs in SSI tuple locking. 1. In heap_hot_search_buffer(), the PredicateLockTuple() call is passed wrong offset number. heapTuple->t_self is set to the tid of the first tuple in the chain that's visited, not the one actually being read. 2. CheckForSerializableConflictIn() uses the tuple's t_ctid field instead of t_self to check for exiting predicate locks on the tuple. If the tuple was updated, but the updater rolled back, t_ctid points to the aborted dead tuple. Reported by Hannu Krosing. Backpatch to 9.1.
2013-10-07 22:57:40 +02:00
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
/* Scan through possible multiple members of HOT-chain */
for (;;)
{
ItemId lp;
/* check for bogus TID */
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
break;
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
lp = PageGetItemId(page, offnum);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
/* check for unused, dead, or redirected items */
if (!ItemIdIsNormal(lp))
{
/* We should only see a redirect at start of chain */
if (ItemIdIsRedirected(lp) && at_chain_start)
{
/* Follow the redirect */
offnum = ItemIdGetRedirect(lp);
at_chain_start = false;
continue;
}
/* else must be end of chain */
break;
}
Fix predicate-locking of HOT updated rows. In serializable mode, heap_hot_search_buffer() incorrectly acquired a predicate lock on the root tuple, not the returned tuple that satisfied the visibility checks. As explained in README-SSI, the predicate lock does not need to be copied or extended to other tuple versions, but for that to work, the correct, visible, tuple version must be locked in the first place. The original SSI commit had this bug in it, but it was fixed back in 2013, in commit 81fbbfe335. But unfortunately, it was reintroduced a few months later in commit b89e151054. Wising up from that, add a regression test to cover this, so that it doesn't get reintroduced again. Also, move the code that sets 't_self', so that it happens at the same time that the other HeapTuple fields are set, to make it more clear that all the code in the loop operate on the "current" tuple in the chain, not the root tuple. Bug spotted by Andres Freund, analysis and original fix by Thomas Munro, test case and some additional changes to the fix by Heikki Linnakangas. Backpatch to all supported versions (9.4). Discussion: https://www.postgresql.org/message-id/20190731210630.nqhszuktygwftjty%40alap3.anarazel.de
2019-08-07 11:40:49 +02:00
/*
* Update heapTuple to point to the element of the HOT chain we're
* currently investigating. Having t_self set correctly is important
* because the SSI checks and the *Satisfies routine for historical
* MVCC snapshots need the correct tid to decide about the visibility.
*/
Variable renaming in preparation for refactoring Rename page -> block and dp -> page where appropriate. The old naming mixed up block and page in confusing ways. Author: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://www.postgresql.org/message-id/flat/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2022-11-16 16:17:18 +01:00
heapTuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
heapTuple->t_len = ItemIdGetLength(lp);
Adjust HeapTupleSatisfies* routines to take a HeapTuple. Previously, these functions took a HeapTupleHeader, but upcoming patches for logical replication will introduce new a new snapshot type under which the tuple's TID will be used to lookup (CMIN, CMAX) for visibility determination purposes. This makes that information available. Code churn is minimal since HeapTupleSatisfiesVisibility took the HeapTuple anyway, and deferenced it before calling the satisfies function. Independently of logical replication, this allows t_tableOid and t_self to be cross-checked via assertions in tqual.c. This seems like a useful way to make sure that all callers are setting these values properly, which has been previously put forward as desirable. Andres Freund, reviewed by Álvaro Herrera
2013-07-22 19:26:33 +02:00
heapTuple->t_tableOid = RelationGetRelid(relation);
Fix predicate-locking of HOT updated rows. In serializable mode, heap_hot_search_buffer() incorrectly acquired a predicate lock on the root tuple, not the returned tuple that satisfied the visibility checks. As explained in README-SSI, the predicate lock does not need to be copied or extended to other tuple versions, but for that to work, the correct, visible, tuple version must be locked in the first place. The original SSI commit had this bug in it, but it was fixed back in 2013, in commit 81fbbfe335. But unfortunately, it was reintroduced a few months later in commit b89e151054. Wising up from that, add a regression test to cover this, so that it doesn't get reintroduced again. Also, move the code that sets 't_self', so that it happens at the same time that the other HeapTuple fields are set, to make it more clear that all the code in the loop operate on the "current" tuple in the chain, not the root tuple. Bug spotted by Andres Freund, analysis and original fix by Thomas Munro, test case and some additional changes to the fix by Heikki Linnakangas. Backpatch to all supported versions (9.4). Discussion: https://www.postgresql.org/message-id/20190731210630.nqhszuktygwftjty%40alap3.anarazel.de
2019-08-07 11:40:49 +02:00
ItemPointerSet(&heapTuple->t_self, blkno, offnum);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
/*
* Shouldn't see a HEAP_ONLY tuple at chain start.
*/
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
if (at_chain_start && HeapTupleIsHeapOnly(heapTuple))
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
break;
/*
* The xmin should match the previous xmax value, else chain is
* broken.
*/
if (TransactionIdIsValid(prev_xmax) &&
Revert bogus fixes of HOT-freezing bug It turns out we misdiagnosed what the real problem was. Revert the previous changes, because they may have worse consequences going forward. A better fix is forthcoming. The simplistic test case is kept, though disabled. Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de
2017-11-02 15:51:05 +01:00
!TransactionIdEquals(prev_xmax,
HeapTupleHeaderGetXmin(heapTuple->t_data)))
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
break;
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
/*
Lots of doc corrections. Josh Kupershmidt
2012-04-24 04:43:09 +02:00
* When first_call is true (and thus, skip is initially false) we'll
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
* return the first tuple we find. But on later passes, heapTuple
* will initially be pointing to the tuple we returned last time.
* Returning it again would be incorrect (and would loop forever), so
* we skip it and return the next match we find.
*/
if (!skip)
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
{
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
/* If it's visible per the snapshot, we must return it */
valid = HeapTupleSatisfiesVisibility(heapTuple, snapshot, buffer);
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
HeapCheckForSerializableConflictOut(valid, relation, heapTuple,
buffer, snapshot);
Introduce logical decoding. This feature, building on previous commits, allows the write-ahead log stream to be decoded into a series of logical changes; that is, inserts, updates, and deletes and the transactions which contain them. It is capable of handling decoding even across changes to the schema of the effected tables. The output format is controlled by a so-called "output plugin"; an example is included. To make use of this in a real replication system, the output plugin will need to be modified to produce output in the format appropriate to that system, and to perform filtering. Currently, information can be extracted from the logical decoding system only via SQL; future commits will add the ability to stream changes via walsender. Andres Freund, with review and other contributions from many other people, including Álvaro Herrera, Abhijit Menon-Sen, Peter Gheogegan, Kevin Grittner, Robert Haas, Heikki Linnakangas, Fujii Masao, Abhijit Menon-Sen, Michael Paquier, Simon Riggs, Craig Ringer, and Steve Singer.
2014-03-03 22:32:18 +01:00
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
if (valid)
{
ItemPointerSetOffsetNumber(tid, offnum);
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
PredicateLockTID(relation, &heapTuple->t_self, snapshot,
HeapTupleHeaderGetXmin(heapTuple->t_data));
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
if (all_dead)
*all_dead = false;
return true;
}
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
}
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
skip = false;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
/*
* If we can't see it, maybe no one else can either. At caller
* request, check whether all chain members are dead to all
* transactions.
Improve performance of get_actual_variable_range with recently-dead tuples. In commit fccebe421, we hacked get_actual_variable_range() to scan the index with SnapshotDirty, so that if there are many uncommitted tuples at the end of the index range, it wouldn't laboriously scan through all of them looking for a live value to return. However, that didn't fix it for the case of many recently-dead tuples at the end of the index; SnapshotDirty recognizes those as committed dead and so we're back to the same problem. To improve the situation, invent a "SnapshotNonVacuumable" snapshot type and use that instead. The reason this helps is that, if the snapshot rejects a given index entry, we know that the indexscan will mark that index entry as killed. This means the next get_actual_variable_range() scan will proceed past that entry without visiting the heap, making the scan a lot faster. We may end up accepting a recently-dead tuple as being the estimated extremal value, but that doesn't seem much worse than the compromise we made before to accept not-yet-committed extremal values. The cost of the scan is still proportional to the number of dead index entries at the end of the range, so in the interval after a mass delete but before VACUUM's cleaned up the mess, it's still possible for get_actual_variable_range() to take a noticeable amount of time, if you've got enough such dead entries. But the constant factor is much much better than before, since all we need to do with each index entry is test its "killed" bit. We chose to back-patch commit fccebe421 at the time, but I'm hesitant to do so here, because this form of the problem seems to affect many fewer people. Also, even when it happens, it's less bad than the case fixed by commit fccebe421 because we don't get the contention effects from expensive TransactionIdIsInProgress tests. Dmitriy Sarafannikov, reviewed by Andrey Borodin Discussion: https://postgr.es/m/05C72CF7-B5F6-4DB9-8A09-5AC897653113@yandex.ru
2017-09-08 01:41:51 +02:00
*
* Note: if you change the criterion here for what is "dead", fix the
* planner's get_actual_variable_range() function to match.
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
*/
snapshot scalability: Don't compute global horizons while building snapshots. To make GetSnapshotData() more scalable, it cannot not look at at each proc's xmin: While snapshot contents do not need to change whenever a read-only transaction commits or a snapshot is released, a proc's xmin is modified in those cases. The frequency of xmin modifications leads to, particularly on higher core count systems, many cache misses inside GetSnapshotData(), despite the data underlying a snapshot not changing. That is the most significant source of GetSnapshotData() scaling poorly on larger systems. Without accessing xmins, GetSnapshotData() cannot calculate accurate horizons / thresholds as it has so far. But we don't really have to: The horizons don't actually change that much between GetSnapshotData() calls. Nor are the horizons actually used every time a snapshot is built. The trick this commit introduces is to delay computation of accurate horizons until there use and using horizon boundaries to determine whether accurate horizons need to be computed. The use of RecentGlobal[Data]Xmin to decide whether a row version could be removed has been replaces with new GlobalVisTest* functions. These use two thresholds to determine whether a row can be pruned: 1) definitely_needed, indicating that rows deleted by XIDs >= definitely_needed are definitely still visible. 2) maybe_needed, indicating that rows deleted by XIDs < maybe_needed can definitely be removed GetSnapshotData() updates definitely_needed to be the xmin of the computed snapshot. When testing whether a row can be removed (with GlobalVisTestIsRemovableXid()) and the tested XID falls in between the two (i.e. XID >= maybe_needed && XID < definitely_needed) the boundaries can be recomputed to be more accurate. As it is not cheap to compute accurate boundaries, we limit the number of times that happens in short succession. As the boundaries used by GlobalVisTestIsRemovableXid() are never reset (with maybe_needed updated by GetSnapshotData()), it is likely that further test can benefit from an earlier computation of accurate horizons. To avoid regressing performance when old_snapshot_threshold is set (as that requires an accurate horizon to be computed), heap_page_prune_opt() doesn't unconditionally call TransactionIdLimitedForOldSnapshots() anymore. Both the computation of the limited horizon, and the triggering of errors (with SetOldSnapshotThresholdTimestamp()) is now only done when necessary to remove tuples. This commit just removes the accesses to PGXACT->xmin from GetSnapshotData(), but other members of PGXACT residing in the same cache line are accessed. Therefore this in itself does not result in a significant improvement. Subsequent commits will take advantage of the fact that GetSnapshotData() now does not need to access xmins anymore. Note: This contains a workaround in heap_page_prune_opt() to keep the snapshot_too_old tests working. While that workaround is ugly, the tests currently are not meaningful, and it seems best to address them separately. Author: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Reviewed-By: Thomas Munro <thomas.munro@gmail.com> Reviewed-By: David Rowley <dgrowleyml@gmail.com> Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
2020-08-13 01:03:49 +02:00
if (all_dead && *all_dead)
{
if (!vistest)
vistest = GlobalVisTestFor(relation);
if (!HeapTupleIsSurelyDead(heapTuple, vistest))
*all_dead = false;
}
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
/*
* Check to see if HOT chain continues past this tuple; if so fetch
* the next offnum and loop around.
*/
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
if (HeapTupleIsHotUpdated(heapTuple))
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
{
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
Assert(ItemPointerGetBlockNumber(&heapTuple->t_data->t_ctid) ==
Fix predicate-locking of HOT updated rows. In serializable mode, heap_hot_search_buffer() incorrectly acquired a predicate lock on the root tuple, not the returned tuple that satisfied the visibility checks. As explained in README-SSI, the predicate lock does not need to be copied or extended to other tuple versions, but for that to work, the correct, visible, tuple version must be locked in the first place. The original SSI commit had this bug in it, but it was fixed back in 2013, in commit 81fbbfe335. But unfortunately, it was reintroduced a few months later in commit b89e151054. Wising up from that, add a regression test to cover this, so that it doesn't get reintroduced again. Also, move the code that sets 't_self', so that it happens at the same time that the other HeapTuple fields are set, to make it more clear that all the code in the loop operate on the "current" tuple in the chain, not the root tuple. Bug spotted by Andres Freund, analysis and original fix by Thomas Munro, test case and some additional changes to the fix by Heikki Linnakangas. Backpatch to all supported versions (9.4). Discussion: https://www.postgresql.org/message-id/20190731210630.nqhszuktygwftjty%40alap3.anarazel.de
2019-08-07 11:40:49 +02:00
blkno);
Avoid having two copies of the HOT-chain search logic. It's been like this since HOT was originally introduced, but the logic is complex enough that this is a recipe for bugs, as we've already found out with SSI. So refactor heap_hot_search_buffer() so that it can satisfy the needs of index_getnext(), and make index_getnext() use that rather than duplicating the logic. This change was originally proposed by Heikki Linnakangas as part of a larger refactoring oriented towards allowing index-only scans. I extracted and adjusted this part, since it seems to have independent merit. Review by Jeff Davis.
2011-06-27 16:27:17 +02:00
offnum = ItemPointerGetOffsetNumber(&heapTuple->t_data->t_ctid);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
at_chain_start = false;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
prev_xmax = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
}
else
break; /* end of chain */
}
The row-version chaining in Serializable Snapshot Isolation was still wrong. On further analysis, it turns out that it is not needed to duplicate predicate locks to the new row version at update, the lock on the version that the transaction saw as visible is enough. However, there was a different bug in the code that checks for dangerous structures when a new rw-conflict happens. Fix that bug, and remove all the row-version chaining related code. Kevin Grittner & Dan Ports, with some comment editorialization by me.
2011-05-30 19:42:16 +02:00
return false;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
}
Mark index entries "killed" when they are no longer visible to any transaction, so as to avoid returning them out of the index AM. Saves repeated heap_fetch operations on frequently-updated rows. Also detect queries on unique keys (equality to all columns of a unique index), and don't bother continuing scan once we have found first match. Killing is implemented in the btree and hash AMs, but not yet in rtree or gist, because there isn't an equally convenient place to do it in those AMs (the outer amgetnext routine can't do it without re-pinning the index page). Did some small cleanup on APIs of HeapTupleSatisfies, heap_fetch, and index_insert to make this a little easier.
2002-05-24 20:57:57 +02:00
/*
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
* heap_get_latest_tid - get the latest tid of a specified tuple
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
*
tableam: Avoid relying on relation size to determine validity of tids. Instead add a tableam callback to do so. To avoid adding per validation overhead, pass a scan to tuple_tid_valid. In heap's case we'd otherwise incurred a RelationGetNumberOfBlocks() call for each tid - which'd have added noticable overhead to nodeTidscan.c. Author: Andres Freund Reviewed-By: Ashwin Agrawal Discussion: https://postgr.es/m/20190515185447.gno2jtqxyktylyvs@alap3.anarazel.de
2019-05-18 03:52:01 +02:00
* Actually, this gets the latest version that is visible according to the
* scan's snapshot. Create a scan using SnapshotDirty to get the very latest,
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
* possibly uncommitted version.
*
* *tid is both an input and an output parameter: it is updated to
* show the latest version of the row. Note that it will not be changed
* if no version of the row passes the snapshot test.
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
*/
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
void
tableam: Avoid relying on relation size to determine validity of tids. Instead add a tableam callback to do so. To avoid adding per validation overhead, pass a scan to tuple_tid_valid. In heap's case we'd otherwise incurred a RelationGetNumberOfBlocks() call for each tid - which'd have added noticable overhead to nodeTidscan.c. Author: Andres Freund Reviewed-By: Ashwin Agrawal Discussion: https://postgr.es/m/20190515185447.gno2jtqxyktylyvs@alap3.anarazel.de
2019-05-18 03:52:01 +02:00
heap_get_latest_tid(TableScanDesc sscan,
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
ItemPointer tid)
{
tableam: Avoid relying on relation size to determine validity of tids. Instead add a tableam callback to do so. To avoid adding per validation overhead, pass a scan to tuple_tid_valid. In heap's case we'd otherwise incurred a RelationGetNumberOfBlocks() call for each tid - which'd have added noticable overhead to nodeTidscan.c. Author: Andres Freund Reviewed-By: Ashwin Agrawal Discussion: https://postgr.es/m/20190515185447.gno2jtqxyktylyvs@alap3.anarazel.de
2019-05-18 03:52:01 +02:00
Relation relation = sscan->rs_rd;
Snapshot snapshot = sscan->rs_snapshot;
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
ItemPointerData ctid;
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
TransactionId priorXmax;
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
/*
Fix comments related to table AMs Incorrect function names were referenced. As this fixes some portions of tableam.h, that is mentioned in the docs as something to look at when implementing a table AM, backpatch down to 12 where this has been introduced. Author: Hironobu Suzuki Discussion: https://postgr.es/m/8fe6d672-28dd-3f1d-7aed-ac2f6d599d3f@interdb.jp Backpatch-through: 12
2020-07-14 06:17:11 +02:00
* table_tuple_get_latest_tid() verified that the passed in tid is valid.
* Assume that t_ctid links are valid however - there shouldn't be invalid
* ones in the table.
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
*/
tableam: Avoid relying on relation size to determine validity of tids. Instead add a tableam callback to do so. To avoid adding per validation overhead, pass a scan to tuple_tid_valid. In heap's case we'd otherwise incurred a RelationGetNumberOfBlocks() call for each tid - which'd have added noticable overhead to nodeTidscan.c. Author: Andres Freund Reviewed-By: Ashwin Agrawal Discussion: https://postgr.es/m/20190515185447.gno2jtqxyktylyvs@alap3.anarazel.de
2019-05-18 03:52:01 +02:00
Assert(ItemPointerIsValid(tid));
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
/*
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
* Loop to chase down t_ctid links. At top of loop, ctid is the tuple we
* need to examine, and *tid is the TID we will return if ctid turns out
* to be bogus.
*
* Note that we will loop until we reach the end of the t_ctid chain.
* Depending on the snapshot passed, there might be at most one visible
* version of the row, but we don't try to optimize for that.
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
*/
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
ctid = *tid;
priorXmax = InvalidTransactionId; /* cannot check first XMIN */
for (;;)
{
Buffer buffer;
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
Page page;
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
OffsetNumber offnum;
ItemId lp;
HeapTupleData tp;
bool valid;
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
/*
* Read, pin, and lock the page.
*/
buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&ctid));
LockBuffer(buffer, BUFFER_LOCK_SHARE);
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(buffer);
TestForOldSnapshot(snapshot, relation, page);
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
/*
* Check for bogus item number. This is not treated as an error
* condition because it can happen while following a t_ctid link. We
* just assume that the prior tid is OK and return it unchanged.
*/
offnum = ItemPointerGetOffsetNumber(&ctid);
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
{
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
UnlockReleaseBuffer(buffer);
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
break;
}
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
lp = PageGetItemId(page, offnum);
Redefine the lp_flags field of item pointers as having four states, rather than two independent bits (one of which was never used in heap pages anyway, or at least hadn't been in a very long time). This gives us flexibility to add the HOT notions of redirected and dead item pointers without requiring anything so klugy as magic values of lp_off and lp_len. The state values are chosen so that for the states currently in use (pre-HOT) there is no change in the physical representation.
2007-09-13 00:10:26 +02:00
if (!ItemIdIsNormal(lp))
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
{
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
UnlockReleaseBuffer(buffer);
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
break;
}
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
/* OK to access the tuple */
tp.t_self = ctid;
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
tp.t_len = ItemIdGetLength(lp);
Adjust HeapTupleSatisfies* routines to take a HeapTuple. Previously, these functions took a HeapTupleHeader, but upcoming patches for logical replication will introduce new a new snapshot type under which the tuple's TID will be used to lookup (CMIN, CMAX) for visibility determination purposes. This makes that information available. Code churn is minimal since HeapTupleSatisfiesVisibility took the HeapTuple anyway, and deferenced it before calling the satisfies function. Independently of logical replication, this allows t_tableOid and t_self to be cross-checked via assertions in tqual.c. This seems like a useful way to make sure that all callers are setting these values properly, which has been previously put forward as desirable. Andres Freund, reviewed by Álvaro Herrera
2013-07-22 19:26:33 +02:00
tp.t_tableOid = RelationGetRelid(relation);
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
/*
* After following a t_ctid link, we might arrive at an unrelated
* tuple. Check for XMIN match.
*/
if (TransactionIdIsValid(priorXmax) &&
Revert bogus fixes of HOT-freezing bug It turns out we misdiagnosed what the real problem was. Revert the previous changes, because they may have worse consequences going forward. A better fix is forthcoming. The simplistic test case is kept, though disabled. Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de
2017-11-02 15:51:05 +01:00
!TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(tp.t_data)))
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
{
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
UnlockReleaseBuffer(buffer);
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
break;
}
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
/*
Rephrase references to "time qualification". Now that the relevant code has, for other reasons, moved out of tqual.[ch], it seems time to refer to visiblity rather than time qualification. Author: Andres Freund Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-01-22 02:03:15 +01:00
* Check tuple visibility; if visible, set it as the new result
* candidate.
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
*/
Change seqscan logic so that we check visibility of all tuples on a page when we first read the page, rather than checking them one at a time. This allows us to take and release the buffer content lock just once per page, instead of once per tuple. Since it's a shared lock the contention penalty for holding the lock longer shouldn't be too bad. We can safely do this only when using an MVCC snapshot; else the assumption that visibility won't change over time is uncool. Therefore there are now two code paths depending on the snapshot type. I also made the same change in nodeBitmapHeapscan.c, where it can be done always because we only support MVCC snapshots for bitmap scans anyway. Also make some incidental cleanups in the APIs of these functions. Per a suggestion from Qingqing Zhou.
2005-11-26 04:03:07 +01:00
valid = HeapTupleSatisfiesVisibility(&tp, snapshot, buffer);
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
HeapCheckForSerializableConflictOut(valid, relation, &tp, buffer, snapshot);
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
if (valid)
*tid = ctid;
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
/*
* If there's a valid t_ctid link, follow it, else we're done.
*/
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
HeapTupleHeaderIsOnlyLocked(tp.t_data) ||
Raise error when affecting tuple moved into different partition. When an update moves a row between partitions (supported since 2f178441044b), our normal logic for following update chains in READ COMMITTED mode doesn't work anymore. Cross partition updates are modeled as an delete from the old and insert into the new partition. No ctid chain exists across partitions, and there's no convenient space to introduce that link. Not throwing an error in a partitioned context when one would have been thrown without partitioning is obviously problematic. This commit introduces infrastructure to detect when a tuple has been moved, not just plainly deleted. That allows to throw an error when encountering a deletion that's actually a move, while attempting to following a ctid chain. The row deleted as part of a cross partition update is marked by pointing it's t_ctid to an invalid block, instead of self as a normal update would. That was deemed to be the least invasive and most future proof way to represent the knowledge, given how few infomask bits are there to be recycled (there's also some locking issues with using infomask bits). External code following ctid chains should be updated to check for moved tuples. The most likely consequence of not doing so is a missed error. Author: Amul Sul, editorialized by me Reviewed-By: Amit Kapila, Pavan Deolasee, Andres Freund, Robert Haas Discussion: http://postgr.es/m/CAAJ_b95PkwojoYfz0bzXU8OokcTVGzN6vYGCNVUukeUDrnF3dw@mail.gmail.com
2018-04-07 22:24:10 +02:00
HeapTupleHeaderIndicatesMovedPartitions(tp.t_data) ||
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
{
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
UnlockReleaseBuffer(buffer);
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
break;
}
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
ctid = tp.t_data->t_ctid;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
priorXmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
UnlockReleaseBuffer(buffer);
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
} /* end of loop */
The 1st step to implement new type of scan,TidScan. Now WHERE restriction on ctid is allowed though it is sequentially scanned.
1999-10-11 08:28:29 +02:00
}
Fix oversight in async-commit patch: there were some places in heapam.c that still thought they could set HEAP_XMAX_COMMITTED immediately after seeing the other transaction commit. Make them use the same logic as tqual.c does to determine if the hint bit can be set yet.
2007-08-14 19:35:18 +02:00
/*
* UpdateXmaxHintBits - update tuple hint bits after xmax transaction ends
*
* This is called after we have waited for the XMAX transaction to terminate.
* If the transaction aborted, we guarantee the XMAX_INVALID hint bit will
* be set on exit. If the transaction committed, we set the XMAX_COMMITTED
* hint bit if possible --- but beware that that may not yet be possible,
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* if the transaction committed asynchronously.
*
* Note that if the transaction was a locker only, we set HEAP_XMAX_INVALID
* even if it commits.
*
* Hence callers should look only at XMAX_INVALID.
*
* Note this is not allowed for tuples whose xmax is a multixact.
Fix oversight in async-commit patch: there were some places in heapam.c that still thought they could set HEAP_XMAX_COMMITTED immediately after seeing the other transaction commit. Make them use the same logic as tqual.c does to determine if the hint bit can be set yet.
2007-08-14 19:35:18 +02:00
*/
static void
UpdateXmaxHintBits(HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
{
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Assert(TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple), xid));
Assert(!(tuple->t_infomask & HEAP_XMAX_IS_MULTI));
Fix oversight in async-commit patch: there were some places in heapam.c that still thought they could set HEAP_XMAX_COMMITTED immediately after seeing the other transaction commit. Make them use the same logic as tqual.c does to determine if the hint bit can be set yet.
2007-08-14 19:35:18 +02:00
if (!(tuple->t_infomask & (HEAP_XMAX_COMMITTED | HEAP_XMAX_INVALID)))
{
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask) &&
TransactionIdDidCommit(xid))
Fix oversight in async-commit patch: there were some places in heapam.c that still thought they could set HEAP_XMAX_COMMITTED immediately after seeing the other transaction commit. Make them use the same logic as tqual.c does to determine if the hint bit can be set yet.
2007-08-14 19:35:18 +02:00
HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_COMMITTED,
xid);
else
HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_INVALID,
InvalidTransactionId);
}
}
Improve bulk-insert performance by keeping the current target buffer pinned (but not locked, as that would risk deadlocks). Also, make it work in a small ring of buffers to avoid having bulk inserts trash the whole buffer arena. Robert Haas, after an idea of Simon Riggs'.
2008-11-06 21:51:15 +01:00
/*
* GetBulkInsertState - prepare status object for a bulk insert
*/
BulkInsertState
GetBulkInsertState(void)
{
BulkInsertState bistate;
bistate = (BulkInsertState) palloc(sizeof(BulkInsertStateData));
bistate->strategy = GetAccessStrategy(BAS_BULKWRITE);
bistate->current_buf = InvalidBuffer;
hio: Use ExtendBufferedRelBy() to extend tables more efficiently While we already had some form of bulk extension for relations, it was fairly limited. It only amortized the cost of acquiring the extension lock, the relation itself was still extended one-by-one. Bulk extension was also solely triggered by contention, not by the amount of data inserted. To address this, use ExtendBufferedRelBy(), introduced in 31966b151e6, to extend the relation. We try to extend the relation by multiple blocks in two situations: 1) The caller tells RelationGetBufferForTuple() that it will need multiple pages. For now that's only used by heap_multi_insert(), see commit FIXME. 2) If there is contention on the extension lock, use the number of waiters for the lock as a multiplier for the number of blocks to extend by. This is similar to what we already did. Previously we additionally multiplied the numbers of waiters by 20, but with the new relation extension infrastructure I could not see a benefit in doing so. Using the freespacemap to provide empty pages can cause significant contention, and adds measurable overhead, even if there is no contention. To reduce that, remember the blocks the relation was extended by in the BulkInsertState, in the extending backend. In case 1) from above, the blocks the extending backend needs are not entered into the FSM, as we know that we will need those blocks. One complication with using the FSM to record empty pages, is that we need to insert blocks into the FSM, when we already hold a buffer content lock. To avoid doing IO while holding a content lock, release the content lock before recording free space. Currently that opens a small window in which another backend could fill the block, if a concurrent VACUUM records the free space. If that happens, we retry, similar to the already existing case when otherBuffer is provided. In the future it might be worth closing the race by preventing VACUUM from recording the space in newly extended pages. This change provides very significant wins (3x at 16 clients, on my workstation) for concurrent COPY into a single relation. Even single threaded COPY is measurably faster, primarily due to not dirtying pages while extending, if supported by the operating system (see commit 4d330a61bb1). Even single-row INSERTs benefit, although to a much smaller degree, as the relation extension lock rarely is the primary bottleneck. Reviewed-by: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://postgr.es/m/20221029025420.eplyow6k7tgu6he3@awork3.anarazel.de
2023-04-07 01:35:21 +02:00
bistate->next_free = InvalidBlockNumber;
bistate->last_free = InvalidBlockNumber;
Improve bulk-insert performance by keeping the current target buffer pinned (but not locked, as that would risk deadlocks). Also, make it work in a small ring of buffers to avoid having bulk inserts trash the whole buffer arena. Robert Haas, after an idea of Simon Riggs'.
2008-11-06 21:51:15 +01:00
return bistate;
}
/*
* FreeBulkInsertState - clean up after finishing a bulk insert
*/
void
FreeBulkInsertState(BulkInsertState bistate)
{
if (bistate->current_buf != InvalidBuffer)
ReleaseBuffer(bistate->current_buf);
FreeAccessStrategy(bistate->strategy);
pfree(bistate);
}
Fix interaction of partitioned tables with BulkInsertState. When copying into a partitioned table, the target heap may change from one tuple to next. We must ask ReadBufferBI() to get a new buffer every time such change occurs. To do that, use new function ReleaseBulkInsertStatePin(). This fixes the bug that tuples ended up being inserted into the wrong partition, which occurred exactly because the wrong buffer was used. Amit Langote, per a suggestion from Robert Haas. Some cosmetic adjustments by me. Reports by 高增琦 (Gao Zengqi), Venkata B Nagothi, and Ragnar Ouchterlony. Discussion: http://postgr.es/m/CAFmBtr32FDOqofo8yG-4mjzL1HnYHxXK5S9OGFJ%3D%3DcJpgEW4vA%40mail.gmail.com Discussion: http://postgr.es/m/CAEyp7J9WiX0L3DoiNcRrY-9iyw%3DqP%2Bj%3DDLsAnNFF1xT2J1ggfQ%40mail.gmail.com Discussion: http://postgr.es/m/16d73804-c9cd-14c5-463e-5caad563ff77%40agama.tv Discussion: http://postgr.es/m/CA+TgmoaiZpDVUUN8LZ4jv1qFE_QyR+H9ec+79f5vNczYarg5Zg@mail.gmail.com
2017-01-24 14:50:16 +01:00
/*
* ReleaseBulkInsertStatePin - release a buffer currently held in bistate
*/
void
ReleaseBulkInsertStatePin(BulkInsertState bistate)
{
if (bistate->current_buf != InvalidBuffer)
ReleaseBuffer(bistate->current_buf);
bistate->current_buf = InvalidBuffer;
}
Improve bulk-insert performance by keeping the current target buffer pinned (but not locked, as that would risk deadlocks). Also, make it work in a small ring of buffers to avoid having bulk inserts trash the whole buffer arena. Robert Haas, after an idea of Simon Riggs'.
2008-11-06 21:51:15 +01:00
Remove global variable scanCommandId in favor of storing a command ID in snapshots, per my proposal of a few days ago. Also, tweak heapam.c routines (heap_insert, heap_update, heap_delete, heap_mark4update) to be passed the command ID to use, instead of doing GetCurrentCommandID. For catalog updates they'll still get passed current command ID, but for updates generated from the main executor they'll get passed the command ID saved in the snapshot the query is using. This should fix some corner cases associated with functions and triggers that advance current command ID while an outer query is still in progress.
2002-05-22 00:05:55 +02:00
/*
* heap_insert - insert tuple into a heap
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*
Remove global variable scanCommandId in favor of storing a command ID in snapshots, per my proposal of a few days ago. Also, tweak heapam.c routines (heap_insert, heap_update, heap_delete, heap_mark4update) to be passed the command ID to use, instead of doing GetCurrentCommandID. For catalog updates they'll still get passed current command ID, but for updates generated from the main executor they'll get passed the command ID saved in the snapshot the query is using. This should fix some corner cases associated with functions and triggers that advance current command ID while an outer query is still in progress.
2002-05-22 00:05:55 +02:00
* The new tuple is stamped with current transaction ID and the specified
* command ID.
Avoid WAL-logging individual tuple insertions during CREATE TABLE AS (a/k/a SELECT INTO). Instead, flush and fsync the whole relation before committing. We do still need the WAL log when PITR is active, however. Simon Riggs and Tom Lane.
2005-06-20 20:37:02 +02:00
*
tableam: Rename wrapper functions to match callback names. Some of the wrapper functions didn't match the callback names. Many of them due to staying "consistent" with historic naming of the wrapped functionality. We decided that for most cases it's more important to be for tableam to be consistent going forward, than with the past. The one exception is beginscan/endscan/... because it'd have looked odd to have systable_beginscan/endscan/... with a different naming scheme, and changing the systable_* APIs would have caused way too much churn (including breaking a lot of external users). Author: Ashwin Agrawal, with some small additions by Andres Freund Reviewed-By: Andres Freund Discussion: https://postgr.es/m/CALfoeiugyrXZfX7n0ORCa4L-m834dzmaE8eFdbNR6PMpetU4Ww@mail.gmail.com
2019-05-24 01:25:48 +02:00
* See table_tuple_insert for comments about most of the input flags, except
* that this routine directly takes a tuple rather than a slot.
Fix logical decoding error when system table w/ toast is repeatedly rewritten. Repeatedly rewriting a mapped catalog table with VACUUM FULL or CLUSTER could cause logical decoding to fail with: ERROR, "could not map filenode \"%s\" to relation OID" To trigger the problem the rewritten catalog had to have live tuples with toasted columns. The problem was triggered as during catalog table rewrites the heap_insert() check that prevents logical decoding information to be emitted for system catalogs, failed to treat the new heap's toast table as a system catalog (because the new heap is not recognized as a catalog table via RelationIsLogicallyLogged()). The relmapper, in contrast to the normal catalog contents, does not contain historical information. After a single rewrite of a mapped table the new relation is known to the relmapper, but if the table is rewritten twice before logical decoding occurs, the relfilenode cannot be mapped to a relation anymore. Which then leads us to error out. This only happens for toast tables, because the main table contents aren't re-inserted with heap_insert(). The fix is simple, add a new heap_insert() flag that prevents logical decoding information from being emitted, and accept during decoding that there might not be tuple data for toast tables. Unfortunately that does not fix pre-existing logical decoding errors. Doing so would require not throwing an error when a filenode cannot be mapped to a relation during decoding, and that seems too likely to hide bugs. If it's crucial to fix decoding for an existing slot, temporarily changing the ERROR in ReorderBufferCommit() to a WARNING appears to be the best fix. Author: Andres Freund Discussion: https://postgr.es/m/20180914021046.oi7dm4ra3ot2g2kt@alap3.anarazel.de Backpatch: 9.4-, where logical decoding was introduced
2018-10-10 22:53:02 +02:00
*
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
* There's corresponding HEAP_INSERT_ options to all the TABLE_INSERT_
* options, and there additionally is HEAP_INSERT_SPECULATIVE which is used to
tableam: Rename wrapper functions to match callback names. Some of the wrapper functions didn't match the callback names. Many of them due to staying "consistent" with historic naming of the wrapped functionality. We decided that for most cases it's more important to be for tableam to be consistent going forward, than with the past. The one exception is beginscan/endscan/... because it'd have looked odd to have systable_beginscan/endscan/... with a different naming scheme, and changing the systable_* APIs would have caused way too much churn (including breaking a lot of external users). Author: Ashwin Agrawal, with some small additions by Andres Freund Reviewed-By: Andres Freund Discussion: https://postgr.es/m/CALfoeiugyrXZfX7n0ORCa4L-m834dzmaE8eFdbNR6PMpetU4Ww@mail.gmail.com
2019-05-24 01:25:48 +02:00
* implement table_tuple_insert_speculative().
Teach CLUSTER to skip writing WAL if not needed (ie, not using archiving) --- Simon. Also, code review and cleanup for the previous COPY-no-WAL patches --- Tom.
2007-03-29 02:15:39 +02:00
*
Adjust some more comments for WITH OIDS removal. I missed these in 578b229718e8f. Author: Andres Freund
2019-01-21 18:05:51 +01:00
* On return the header fields of *tup are updated to match the stored tuple;
* in particular tup->t_self receives the actual TID where the tuple was
* stored. But note that any toasting of fields within the tuple data is NOT
* reflected into *tup.
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*/
Remove WITH OIDS support, change oid catalog column visibility. Previously tables declared WITH OIDS, including a significant fraction of the catalog tables, stored the oid column not as a normal column, but as part of the tuple header. This special column was not shown by default, which was somewhat odd, as it's often (consider e.g. pg_class.oid) one of the more important parts of a row. Neither pg_dump nor COPY included the contents of the oid column by default. The fact that the oid column was not an ordinary column necessitated a significant amount of special case code to support oid columns. That already was painful for the existing, but upcoming work aiming to make table storage pluggable, would have required expanding and duplicating that "specialness" significantly. WITH OIDS has been deprecated since 2005 (commit ff02d0a05280e0). Remove it. Removing includes: - CREATE TABLE and ALTER TABLE syntax for declaring the table to be WITH OIDS has been removed (WITH (oids[ = true]) will error out) - pg_dump does not support dumping tables declared WITH OIDS and will issue a warning when dumping one (and ignore the oid column). - restoring an pg_dump archive with pg_restore will warn when restoring a table with oid contents (and ignore the oid column) - COPY will refuse to load binary dump that includes oids. - pg_upgrade will error out when encountering tables declared WITH OIDS, they have to be altered to remove the oid column first. - Functionality to access the oid of the last inserted row (like plpgsql's RESULT_OID, spi's SPI_lastoid, ...) has been removed. The syntax for declaring a table WITHOUT OIDS (or WITH (oids = false) for CREATE TABLE) is still supported. While that requires a bit of support code, it seems unnecessary to break applications / dumps that do not use oids, and are explicit about not using them. The biggest user of WITH OID columns was postgres' catalog. This commit changes all 'magic' oid columns to be columns that are normally declared and stored. To reduce unnecessary query breakage all the newly added columns are still named 'oid', even if a table's column naming scheme would indicate 'reloid' or such. This obviously requires adapting a lot code, mostly replacing oid access via HeapTupleGetOid() with access to the underlying Form_pg_*->oid column. The bootstrap process now assigns oids for all oid columns in genbki.pl that do not have an explicit value (starting at the largest oid previously used), only oids assigned later by oids will be above FirstBootstrapObjectId. As the oid column now is a normal column the special bootstrap syntax for oids has been removed. Oids are not automatically assigned during insertion anymore, all backend code explicitly assigns oids with GetNewOidWithIndex(). For the rare case that insertions into the catalog via SQL are called for the new pg_nextoid() function can be used (which only works on catalog tables). The fact that oid columns on system tables are now normal columns means that they will be included in the set of columns expanded by * (i.e. SELECT * FROM pg_class will now include the table's oid, previously it did not). It'd not technically be hard to hide oid column by default, but that'd mean confusing behavior would either have to be carried forward forever, or it'd cause breakage down the line. While it's not unlikely that further adjustments are needed, the scope/invasiveness of the patch makes it worthwhile to get merge this now. It's painful to maintain externally, too complicated to commit after the code code freeze, and a dependency of a number of other patches. Catversion bump, for obvious reasons. Author: Andres Freund, with contributions by John Naylor Discussion: https://postgr.es/m/20180930034810.ywp2c7awz7opzcfr@alap3.anarazel.de
2018-11-21 00:36:57 +01:00
void
Avoid WAL-logging individual tuple insertions during CREATE TABLE AS (a/k/a SELECT INTO). Instead, flush and fsync the whole relation before committing. We do still need the WAL log when PITR is active, however. Simon Riggs and Tom Lane.
2005-06-20 20:37:02 +02:00
heap_insert(Relation relation, HeapTuple tup, CommandId cid,
Improve bulk-insert performance by keeping the current target buffer pinned (but not locked, as that would risk deadlocks). Also, make it work in a small ring of buffers to avoid having bulk inserts trash the whole buffer arena. Robert Haas, after an idea of Simon Riggs'.
2008-11-06 21:51:15 +01:00
int options, BulkInsertState bistate)
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
{
Restructure subtransaction handling to reduce resource consumption, as per recent discussions. Invent SubTransactionIds that are managed like CommandIds (ie, counter is reset at start of each top transaction), and use these instead of TransactionIds to keep track of subtransaction status in those modules that need it. This means that a subtransaction does not need an XID unless it actually inserts/modifies rows in the database. Accordingly, don't assign it an XID nor take a lock on the XID until it tries to do that. This saves a lot of overhead for subtransactions that are only used for error recovery (eg plpgsql exceptions). Also, arrange to release a subtransaction's XID lock as soon as the subtransaction exits, in both the commit and abort cases. This avoids holding many unique locks after a long series of subtransactions. The price is some additional overhead in XactLockTableWait, but that seems acceptable. Finally, restructure the state machine in xact.c to have a more orthogonal set of states for subtransactions.
2004-09-16 18:58:44 +02:00
TransactionId xid = GetCurrentTransactionId();
Modify tuptoaster's API so that it does not try to modify the passed tuple in-place, but instead passes back an all-new tuple structure if any changes are needed. This is a much cleaner and more robust solution for the bug discovered by Alexey Beschiokov; accordingly, revert the quick hack I installed yesterday. With this change, HeapTupleData.t_datamcxt is no longer needed; will remove it in a separate commit in HEAD only.
2005-11-20 19:38:20 +01:00
HeapTuple heaptup;
heap' logging
2000-07-03 04:54:21 +02:00
Buffer buffer;
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Buffer vmbuffer = InvalidBuffer;
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
bool all_visible_cleared = false;
heap' logging
2000-07-03 04:54:21 +02:00
Fix mishandling of resjunk columns in ON CONFLICT ... UPDATE tlists. It's unusual to have any resjunk columns in an ON CONFLICT ... UPDATE list, but it can happen when MULTIEXPR_SUBLINK SubPlans are present. If it happens, the ON CONFLICT UPDATE code path would end up storing tuples that include the values of the extra resjunk columns. That's fairly harmless in the short run, but if new columns are added to the table then the values would become accessible, possibly leading to malfunctions if they don't match the datatypes of the new columns. This had escaped notice through a confluence of missing sanity checks, including * There's no cross-check that a tuple presented to heap_insert or heap_update matches the table rowtype. While it's difficult to check that fully at reasonable cost, we can easily add assertions that there aren't too many columns. * The output-column-assignment cases in execExprInterp.c lacked any sanity checks on the output column numbers, which seems like an oversight considering there are plenty of assertion checks on input column numbers. Add assertions there too. * We failed to apply nodeModifyTable's ExecCheckPlanOutput() to the ON CONFLICT UPDATE tlist. That wouldn't have caught this specific error, since that function is chartered to ignore resjunk columns; but it sure seems like a bad omission now that we've seen this bug. In HEAD, the right way to fix this is to make the processing of ON CONFLICT UPDATE tlists work the same as regular UPDATE tlists now do, that is don't add "SET x = x" entries, and use ExecBuildUpdateProjection to evaluate the tlist and combine it with old values of the not-set columns. This adds a little complication to ExecBuildUpdateProjection, but allows removal of a comparable amount of now-dead code from the planner. In the back branches, the most expedient solution seems to be to (a) use an output slot for the ON CONFLICT UPDATE projection that actually matches the target table, and then (b) invent a variant of ExecBuildProjectionInfo that can be told to not store values resulting from resjunk columns, so it doesn't try to store into nonexistent columns of the output slot. (We can't simply ignore the resjunk columns altogether; they have to be evaluated for MULTIEXPR_SUBLINK to work.) This works back to v10. In 9.6, projections work much differently and we can't cheaply give them such an option. The 9.6 version of this patch works by inserting a JunkFilter when it's necessary to get rid of resjunk columns. In addition, v11 and up have the reverse problem when trying to perform ON CONFLICT UPDATE on a partitioned table. Through a further oversight, adjust_partition_tlist() discarded resjunk columns when re-ordering the ON CONFLICT UPDATE tlist to match a partition. This accidentally prevented the storing-bogus-tuples problem, but at the cost that MULTIEXPR_SUBLINK cases didn't work, typically crashing if more than one row has to be updated. Fix by preserving resjunk columns in that routine. (I failed to resist the temptation to add more assertions there too, and to do some minor code beautification.) Per report from Andres Freund. Back-patch to all supported branches. Security: CVE-2021-32028
2021-05-10 17:02:29 +02:00
/* Cheap, simplistic check that the tuple matches the rel's rowtype. */
Assert(HeapTupleHeaderGetNatts(tup->t_data) <=
RelationGetNumberOfAttributes(relation));
TOAST WARNING: This is actually broken - we have self-deadlocks due to concurrent changes in buffer management. Vadim and me are working on it. Jan
2000-07-04 01:10:14 +02:00
/*
Adjust some more comments for WITH OIDS removal. I missed these in 578b229718e8f. Author: Andres Freund
2019-01-21 18:05:51 +01:00
* Fill in tuple header fields and toast the tuple if necessary.
Modify tuptoaster's API so that it does not try to modify the passed tuple in-place, but instead passes back an all-new tuple structure if any changes are needed. This is a much cleaner and more robust solution for the bug discovered by Alexey Beschiokov; accordingly, revert the quick hack I installed yesterday. With this change, HeapTupleData.t_datamcxt is no longer needed; will remove it in a separate commit in HEAD only.
2005-11-20 19:38:20 +01:00
*
* Note: below this point, heaptup is the data we actually intend to store
* into the relation; tup is the caller's original untoasted data.
TOAST WARNING: This is actually broken - we have self-deadlocks due to concurrent changes in buffer management. Vadim and me are working on it. Jan
2000-07-04 01:10:14 +02:00
*/
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
heaptup = heap_prepare_insert(relation, tup, xid, cid, options);
TOAST WARNING: This is actually broken - we have self-deadlocks due to concurrent changes in buffer management. Vadim and me are working on it. Jan
2000-07-04 01:10:14 +02:00
Fix serialization anomalies due to race conditions on INSERT. On insert the CheckForSerializableConflictIn() test was performed before the page(s) which were going to be modified had been locked (with an exclusive buffer content lock). If another process acquired a relation SIReadLock on the heap and scanned to a page on which an insert was going to occur before the page was so locked, a rw-conflict would be missed, which could allow a serialization anomaly to be missed. The window between the check and the page lock was small, so the bug was generally not noticed unless there was high concurrency with multiple processes inserting into the same table. This was reported by Peter Bailis as bug #11732, by Sean Chittenden as bug #13667, and by others. The race condition was eliminated in heap_insert() by moving the check down below the acquisition of the buffer lock, which had been the very next statement. Because of the loop locking and unlocking multiple buffers in heap_multi_insert() a check was added after all inserts were completed. The check before the start of the inserts was left because it might avoid a large amount of work to detect a serialization anomaly before performing the all of the inserts and the related WAL logging. While investigating this bug, other SSI bugs which were even harder to hit in practice were noticed and fixed, an unnecessary check (covered by another check, so redundant) was removed from heap_update(), and comments were improved. Back-patch to all supported branches. Kevin Grittner and Thomas Munro
2015-10-31 20:43:34 +01:00
/*
* Find buffer to insert this tuple into. If the page is all visible,
* this will also pin the requisite visibility map page.
*/
buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
InvalidBuffer, options, bistate,
heapam: Pass number of required pages to RelationGetBufferForTuple() A future commit will use this information to determine how aggressively to extend the relation by. In heap_multi_insert() we know accurately how many pages we need once we need to extend the relation, providing an accurate lower bound for how much to extend. Reviewed-by: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://postgr.es/m/20221029025420.eplyow6k7tgu6he3@awork3.anarazel.de
2023-04-06 23:18:24 +02:00
&vmbuffer, NULL,
0);
Fix serialization anomalies due to race conditions on INSERT. On insert the CheckForSerializableConflictIn() test was performed before the page(s) which were going to be modified had been locked (with an exclusive buffer content lock). If another process acquired a relation SIReadLock on the heap and scanned to a page on which an insert was going to occur before the page was so locked, a rw-conflict would be missed, which could allow a serialization anomaly to be missed. The window between the check and the page lock was small, so the bug was generally not noticed unless there was high concurrency with multiple processes inserting into the same table. This was reported by Peter Bailis as bug #11732, by Sean Chittenden as bug #13667, and by others. The race condition was eliminated in heap_insert() by moving the check down below the acquisition of the buffer lock, which had been the very next statement. Because of the loop locking and unlocking multiple buffers in heap_multi_insert() a check was added after all inserts were completed. The check before the start of the inserts was left because it might avoid a large amount of work to detect a serialization anomaly before performing the all of the inserts and the related WAL logging. While investigating this bug, other SSI bugs which were even harder to hit in practice were noticed and fixed, an unnecessary check (covered by another check, so redundant) was removed from heap_update(), and comments were improved. Back-patch to all supported branches. Kevin Grittner and Thomas Munro
2015-10-31 20:43:34 +01:00
Avoid unnecessary page-level SSI lock check in heap_insert(). As observed by Heikki, we need not conflict on heap page locks during an insert; heap page locks are only aggregated tuple locks, they don't imply locking "gaps" as index page locks do. So we can avoid some unnecessary conflicts, and also do the SSI check while not holding exclusive lock on the target buffer. Kevin Grittner, reviewed by Jeff Davis. Back-patch to 9.1.
2011-09-16 20:47:20 +02:00
/*
* We're about to do the actual insert -- but check for conflict first, to
* avoid possibly having to roll back work we've just done.
*
Fix serialization anomalies due to race conditions on INSERT. On insert the CheckForSerializableConflictIn() test was performed before the page(s) which were going to be modified had been locked (with an exclusive buffer content lock). If another process acquired a relation SIReadLock on the heap and scanned to a page on which an insert was going to occur before the page was so locked, a rw-conflict would be missed, which could allow a serialization anomaly to be missed. The window between the check and the page lock was small, so the bug was generally not noticed unless there was high concurrency with multiple processes inserting into the same table. This was reported by Peter Bailis as bug #11732, by Sean Chittenden as bug #13667, and by others. The race condition was eliminated in heap_insert() by moving the check down below the acquisition of the buffer lock, which had been the very next statement. Because of the loop locking and unlocking multiple buffers in heap_multi_insert() a check was added after all inserts were completed. The check before the start of the inserts was left because it might avoid a large amount of work to detect a serialization anomaly before performing the all of the inserts and the related WAL logging. While investigating this bug, other SSI bugs which were even harder to hit in practice were noticed and fixed, an unnecessary check (covered by another check, so redundant) was removed from heap_update(), and comments were improved. Back-patch to all supported branches. Kevin Grittner and Thomas Munro
2015-10-31 20:43:34 +01:00
* This is safe without a recheck as long as there is no possibility of
* another process scanning the page between this check and the insert
* being visible to the scan (i.e., an exclusive buffer content lock is
* continuously held from this point until the tuple insert is visible).
*
Avoid unnecessary page-level SSI lock check in heap_insert(). As observed by Heikki, we need not conflict on heap page locks during an insert; heap page locks are only aggregated tuple locks, they don't imply locking "gaps" as index page locks do. So we can avoid some unnecessary conflicts, and also do the SSI check while not holding exclusive lock on the target buffer. Kevin Grittner, reviewed by Jeff Davis. Back-patch to 9.1.
2011-09-16 20:47:20 +02:00
* For a heap insert, we only need to check for table-level SSI locks. Our
* new tuple can't possibly conflict with existing tuple locks, and heap
* page locks are only consolidated versions of tuple locks; they do not
Fix serialization anomalies due to race conditions on INSERT. On insert the CheckForSerializableConflictIn() test was performed before the page(s) which were going to be modified had been locked (with an exclusive buffer content lock). If another process acquired a relation SIReadLock on the heap and scanned to a page on which an insert was going to occur before the page was so locked, a rw-conflict would be missed, which could allow a serialization anomaly to be missed. The window between the check and the page lock was small, so the bug was generally not noticed unless there was high concurrency with multiple processes inserting into the same table. This was reported by Peter Bailis as bug #11732, by Sean Chittenden as bug #13667, and by others. The race condition was eliminated in heap_insert() by moving the check down below the acquisition of the buffer lock, which had been the very next statement. Because of the loop locking and unlocking multiple buffers in heap_multi_insert() a check was added after all inserts were completed. The check before the start of the inserts was left because it might avoid a large amount of work to detect a serialization anomaly before performing the all of the inserts and the related WAL logging. While investigating this bug, other SSI bugs which were even harder to hit in practice were noticed and fixed, an unnecessary check (covered by another check, so redundant) was removed from heap_update(), and comments were improved. Back-patch to all supported branches. Kevin Grittner and Thomas Munro
2015-10-31 20:43:34 +01:00
* lock "gaps" as index page locks do. So we don't need to specify a
* buffer when making the call, which makes for a faster check.
Avoid unnecessary page-level SSI lock check in heap_insert(). As observed by Heikki, we need not conflict on heap page locks during an insert; heap page locks are only aggregated tuple locks, they don't imply locking "gaps" as index page locks do. So we can avoid some unnecessary conflicts, and also do the SSI check while not holding exclusive lock on the target buffer. Kevin Grittner, reviewed by Jeff Davis. Back-patch to 9.1.
2011-09-16 20:47:20 +02:00
*/
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
Avoid unnecessary page-level SSI lock check in heap_insert(). As observed by Heikki, we need not conflict on heap page locks during an insert; heap page locks are only aggregated tuple locks, they don't imply locking "gaps" as index page locks do. So we can avoid some unnecessary conflicts, and also do the SSI check while not holding exclusive lock on the target buffer. Kevin Grittner, reviewed by Jeff Davis. Back-patch to 9.1.
2011-09-16 20:47:20 +02:00
Error message editing in backend/access.
2003-07-21 22:29:40 +02:00
/* NO EREPORT(ERROR) from here till changes are logged */
Add more critical-section calls: all code sections that hold spinlocks are now critical sections, so as to ensure die() won't interrupt us while we are munging shared-memory data structures. Avoid insecure intermediate states in some code that proc_exit will call, like palloc/pfree. Rename START/END_CRIT_CODE to START/END_CRIT_SECTION, since that seems to be what people tend to call them anyway, and make them be called with () like a function call, in hopes of not confusing pg_indent. I doubt that this is sufficient to make SIGTERM safe anywhere; there's just too much code that could get invoked during proc_exit().
2001-01-12 22:54:01 +01:00
START_CRIT_SECTION();
Make sure that t_ctid is reset to equal t_self in heap_delete and heap_mark4update; this avoids situations where a deleted tuple might look like it is chained to something else. Also, cause all the WAL redo routines to set t_ctid to equal t_self, rather than leaving it undefined as before. Make heap_xlog_clean set the page's LSN and SUI correctly. All per past discussions in pghackers, ranging back to last December.
2002-08-13 22:11:03 +02:00
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
RelationPutHeapTuple(relation, buffer, heaptup,
(options & HEAP_INSERT_SPECULATIVE) != 0);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Revert most of 39b66a91bd Reverts most of commit 39b66a91bd, which was found to cause significant regression for REFRESH MATERIALIZED VIEW. This means only rows inserted by heap_multi_insert will benefit from the optimization, implemented in commit 7db0cd2145. Reported-by: Masahiko Sawada Discussion: https://postgr.es/m/CAD21AoA%3D%3Df2VSw3c-Cp_y%3DWLKHMKc1D6s7g3YWsCOvgaYPpJcg%40mail.gmail.com
2021-06-03 00:06:42 +02:00
if (PageIsAllVisible(BufferGetPage(buffer)))
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
{
all_visible_cleared = true;
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
PageClearAllVisible(BufferGetPage(buffer));
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
visibilitymap_clear(relation,
ItemPointerGetBlockNumber(&(heaptup->t_self)),
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
vmbuffer, VISIBILITYMAP_VALID_BITS);
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
}
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
/*
* XXX Should we set PageSetPrunable on this page ?
*
* The inserting transaction may eventually abort thus making this tuple
* DEAD and hence available for pruning. Though we don't want to optimize
* for aborts, if no other tuple in this page is UPDATEd/DELETEd, the
* aborted tuple will never be pruned until next vacuum is triggered.
*
* If you do add PageSetPrunable here, add it in heap_xlog_insert too.
*/
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
MarkBufferDirty(buffer);
heap' xlog records
2000-06-02 12:20:27 +02:00
/* XLOG stuff */
Skip WAL for new relfilenodes, under wal_level=minimal. Until now, only selected bulk operations (e.g. COPY) did this. If a given relfilenode received both a WAL-skipping COPY and a WAL-logged operation (e.g. INSERT), recovery could lose tuples from the COPY. See src/backend/access/transam/README section "Skipping WAL for New RelFileNode" for the new coding rules. Maintainers of table access methods should examine that section. To maintain data durability, just before commit, we choose between an fsync of the relfilenode and copying its contents to WAL. A new GUC, wal_skip_threshold, guides that choice. If this change slows a workload that creates small, permanent relfilenodes under wal_level=minimal, try adjusting wal_skip_threshold. Users setting a timeout on COMMIT may need to adjust that timeout, and log_min_duration_statement analysis will reflect time consumption moving to COMMIT from commands like COPY. Internally, this requires a reliable determination of whether RollbackAndReleaseCurrentSubTransaction() would unlink a relation's current relfilenode. Introduce rd_firstRelfilenodeSubid. Amend the specification of rd_createSubid such that the field is zero when a new rel has an old rd_node. Make relcache.c retain entries for certain dropped relations until end of transaction. Bump XLOG_PAGE_MAGIC, since this introduces XLOG_GIST_ASSIGN_LSN. Future servers accept older WAL, so this bump is discretionary. Kyotaro Horiguchi, reviewed (in earlier, similar versions) by Robert Haas. Heikki Linnakangas and Michael Paquier implemented earlier designs that materially clarified the problem. Reviewed, in earlier designs, by Andrew Dunstan, Andres Freund, Alvaro Herrera, Tom Lane, Fujii Masao, and Simon Riggs. Reported by Martijn van Oosterhout. Discussion: https://postgr.es/m/20150702220524.GA9392@svana.org
2020-04-04 21:25:34 +02:00
if (RelationNeedsWAL(relation))
heap' xlog records
2000-06-02 12:20:27 +02:00
{
xl_heap_insert xlrec;
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
xl_heap_header xlhdr;
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
XLogRecPtr recptr;
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
Page page = BufferGetPage(buffer);
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
uint8 info = XLOG_HEAP_INSERT;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
int bufflags = 0;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
/*
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
* If this is a catalog, we need to transmit combo CIDs to properly
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
* decode, so log that as well.
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
*/
if (RelationIsAccessibleInLogicalDecoding(relation))
log_heap_new_cid(relation, heaptup);
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
Fix problems with loss of tuple commit status bits during WAL redo of VACUUM FULL tuple moves. Store full-width t_infomask in WAL, rather than storing low 8 bits and expecting to be able to reconstruct upper bits. While at it, remove redundant t_oid field from WAL headers (the OID, if present, is now recorded in the data portion of the tuple). WAL version number bumped --- this does not force an initdb, you can instead run pg_resetxlog after a clean shutdown of the old postmaster.
2002-09-27 00:46:29 +02:00
/*
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
* If this is the single and first tuple on page, we can reinit the
* page instead of restoring the whole thing. Set flag, and hide
* buffer references from XLogInsert.
Fix problems with loss of tuple commit status bits during WAL redo of VACUUM FULL tuple moves. Store full-width t_infomask in WAL, rather than storing low 8 bits and expecting to be able to reconstruct upper bits. While at it, remove redundant t_oid field from WAL headers (the OID, if present, is now recorded in the data portion of the tuple). WAL version number bumped --- this does not force an initdb, you can instead run pg_resetxlog after a clean shutdown of the old postmaster.
2002-09-27 00:46:29 +02:00
*/
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (ItemPointerGetOffsetNumber(&(heaptup->t_self)) == FirstOffsetNumber &&
PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
{
info |= XLOG_HEAP_INIT_PAGE;
bufflags |= REGBUF_WILL_INIT;
}
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
xlrec.offnum = ItemPointerGetOffsetNumber(&heaptup->t_self);
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
xlrec.flags = 0;
if (all_visible_cleared)
xlrec.flags |= XLH_INSERT_ALL_VISIBLE_CLEARED;
if (options & HEAP_INSERT_SPECULATIVE)
xlrec.flags |= XLH_INSERT_IS_SPECULATIVE;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
Assert(ItemPointerGetBlockNumber(&heaptup->t_self) == BufferGetBlockNumber(buffer));
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
/*
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
* For logical decoding, we need the tuple even if we're doing a full
* page write, so make sure it's included even if we take a full-page
* image. (XXX We could alternatively store a pointer into the FPW).
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
*/
Fix logical decoding error when system table w/ toast is repeatedly rewritten. Repeatedly rewriting a mapped catalog table with VACUUM FULL or CLUSTER could cause logical decoding to fail with: ERROR, "could not map filenode \"%s\" to relation OID" To trigger the problem the rewritten catalog had to have live tuples with toasted columns. The problem was triggered as during catalog table rewrites the heap_insert() check that prevents logical decoding information to be emitted for system catalogs, failed to treat the new heap's toast table as a system catalog (because the new heap is not recognized as a catalog table via RelationIsLogicallyLogged()). The relmapper, in contrast to the normal catalog contents, does not contain historical information. After a single rewrite of a mapped table the new relation is known to the relmapper, but if the table is rewritten twice before logical decoding occurs, the relfilenode cannot be mapped to a relation anymore. Which then leads us to error out. This only happens for toast tables, because the main table contents aren't re-inserted with heap_insert(). The fix is simple, add a new heap_insert() flag that prevents logical decoding information from being emitted, and accept during decoding that there might not be tuple data for toast tables. Unfortunately that does not fix pre-existing logical decoding errors. Doing so would require not throwing an error when a filenode cannot be mapped to a relation during decoding, and that seems too likely to hide bugs. If it's crucial to fix decoding for an existing slot, temporarily changing the ERROR in ReorderBufferCommit() to a WARNING appears to be the best fix. Author: Andres Freund Discussion: https://postgr.es/m/20180914021046.oi7dm4ra3ot2g2kt@alap3.anarazel.de Backpatch: 9.4-, where logical decoding was introduced
2018-10-10 22:53:02 +02:00
if (RelationIsLogicallyLogged(relation) &&
!(options & HEAP_INSERT_NO_LOGICAL))
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
{
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
xlrec.flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
bufflags |= REGBUF_KEEP_DATA;
Implement streaming mode in ReorderBuffer. Instead of serializing the transaction to disk after reaching the logical_decoding_work_mem limit in memory, we consume the changes we have in memory and invoke stream API methods added by commit 45fdc9738b. However, sometimes if we have incomplete toast or speculative insert we spill to the disk because we can't generate the complete tuple and stream. And, as soon as we get the complete tuple we stream the transaction including the serialized changes. We can do this incremental processing thanks to having assignments (associating subxact with toplevel xacts) in WAL right away, and thanks to logging the invalidation messages at each command end. These features are added by commits 0bead9af48 and c55040ccd0 respectively. Now that we can stream in-progress transactions, the concurrent aborts may cause failures when the output plugin consults catalogs (both system and user-defined). We handle such failures by returning ERRCODE_TRANSACTION_ROLLBACK sqlerrcode from system table scan APIs to the backend or WALSender decoding a specific uncommitted transaction. The decoding logic on the receipt of such a sqlerrcode aborts the decoding of the current transaction and continue with the decoding of other transactions. We have ReorderBufferTXN pointer in each ReorderBufferChange by which we know which xact it belongs to. The output plugin can use this to decide which changes to discard in case of stream_abort_cb (e.g. when a subxact gets discarded). We also provide a new option via SQL APIs to fetch the changes being streamed. Author: Dilip Kumar, Tomas Vondra, Amit Kapila, Nikhil Sontakke Reviewed-by: Amit Kapila, Kuntal Ghosh, Ajin Cherian Tested-by: Neha Sharma, Mahendra Singh Thalor and Ajin Cherian Discussion: https://postgr.es/m/688b0b7f-2f6c-d827-c27b-216a8e3ea700@2ndquadrant.com
2020-08-08 04:04:39 +02:00
if (IsToastRelation(relation))
xlrec.flags |= XLH_INSERT_ON_TOAST_RELATION;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
}
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHeapInsert);
xlhdr.t_infomask2 = heaptup->t_data->t_infomask2;
xlhdr.t_infomask = heaptup->t_data->t_infomask;
xlhdr.t_hoff = heaptup->t_data->t_hoff;
Fix problems with loss of tuple commit status bits during WAL redo of VACUUM FULL tuple moves. Store full-width t_infomask in WAL, rather than storing low 8 bits and expecting to be able to reconstruct upper bits. While at it, remove redundant t_oid field from WAL headers (the OID, if present, is now recorded in the data portion of the tuple). WAL version number bumped --- this does not force an initdb, you can instead run pg_resetxlog after a clean shutdown of the old postmaster.
2002-09-27 00:46:29 +02:00
/*
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
* note we mark xlhdr as belonging to buffer; if XLogInsert decides to
* write the whole page to the xlog, we don't need to store
* xl_heap_header in the xlog.
Fix problems with loss of tuple commit status bits during WAL redo of VACUUM FULL tuple moves. Store full-width t_infomask in WAL, rather than storing low 8 bits and expecting to be able to reconstruct upper bits. While at it, remove redundant t_oid field from WAL headers (the OID, if present, is now recorded in the data portion of the tuple). WAL version number bumped --- this does not force an initdb, you can instead run pg_resetxlog after a clean shutdown of the old postmaster.
2002-09-27 00:46:29 +02:00
*/
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
XLogRegisterBufData(0, (char *) &xlhdr, SizeOfHeapHeader);
/* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
XLogRegisterBufData(0,
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
(char *) heaptup->t_data + SizeofHeapTupleHeader,
heaptup->t_len - SizeofHeapTupleHeader);
heap' xlog records
2000-06-02 12:20:27 +02:00
Introduce replication progress tracking infrastructure. When implementing a replication solution ontop of logical decoding, two related problems exist: * How to safely keep track of replication progress * How to change replication behavior, based on the origin of a row; e.g. to avoid loops in bi-directional replication setups The solution to these problems, as implemented here, consist out of three parts: 1) 'replication origins', which identify nodes in a replication setup. 2) 'replication progress tracking', which remembers, for each replication origin, how far replay has progressed in a efficient and crash safe manner. 3) The ability to filter out changes performed on the behest of a replication origin during logical decoding; this allows complex replication topologies. E.g. by filtering all replayed changes out. Most of this could also be implemented in "userspace", e.g. by inserting additional rows contain origin information, but that ends up being much less efficient and more complicated. We don't want to require various replication solutions to reimplement logic for this independently. The infrastructure is intended to be generic enough to be reusable. This infrastructure also replaces the 'nodeid' infrastructure of commit timestamps. It is intended to provide all the former capabilities, except that there's only 2^16 different origins; but now they integrate with logical decoding. Additionally more functionality is accessible via SQL. Since the commit timestamp infrastructure has also been introduced in 9.5 (commit 73c986add) changing the API is not a problem. For now the number of origins for which the replication progress can be tracked simultaneously is determined by the max_replication_slots GUC. That GUC is not a perfect match to configure this, but there doesn't seem to be sufficient reason to introduce a separate new one. Bumps both catversion and wal page magic. Author: Andres Freund, with contributions from Petr Jelinek and Craig Ringer Reviewed-By: Heikki Linnakangas, Petr Jelinek, Robert Haas, Steve Singer Discussion: 20150216002155.GI15326@awork2.anarazel.de, 20140923182422.GA15776@alap3.anarazel.de, 20131114172632.GE7522@alap2.anarazel.de
2015-04-29 19:30:53 +02:00
/* filtering by origin on a row level is much more efficient */
Skip checkpoints, archiving on idle systems. Some background activity (like checkpoints, archive timeout, standby snapshots) is not supposed to happen on an idle system. Unfortunately so far it was not easy to determine when a system is idle, which defeated some of the attempts to avoid redundant activity on an idle system. To make that easier, allow to make individual WAL insertions as not being "important". By checking whether any important activity happened since the last time an activity was performed, it now is easy to check whether some action needs to be repeated. Use the new facility for checkpoints, archive timeout and standby snapshots. The lack of a facility causes some issues in older releases, but in my opinion the consequences (superflous checkpoints / archived segments) aren't grave enough to warrant backpatching. Author: Michael Paquier, editorialized by Andres Freund Reviewed-By: Andres Freund, David Steele, Amit Kapila, Kyotaro HORIGUCHI Bug: #13685 Discussion: https://www.postgresql.org/message-id/20151016203031.3019.72930@wrigleys.postgresql.org https://www.postgresql.org/message-id/CAB7nPqQcPqxEM3S735Bd2RzApNqSNJVietAC=6kfkYv_45dKwA@mail.gmail.com Backpatch: -
2016-12-22 20:31:50 +01:00
XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
Introduce replication progress tracking infrastructure. When implementing a replication solution ontop of logical decoding, two related problems exist: * How to safely keep track of replication progress * How to change replication behavior, based on the origin of a row; e.g. to avoid loops in bi-directional replication setups The solution to these problems, as implemented here, consist out of three parts: 1) 'replication origins', which identify nodes in a replication setup. 2) 'replication progress tracking', which remembers, for each replication origin, how far replay has progressed in a efficient and crash safe manner. 3) The ability to filter out changes performed on the behest of a replication origin during logical decoding; this allows complex replication topologies. E.g. by filtering all replayed changes out. Most of this could also be implemented in "userspace", e.g. by inserting additional rows contain origin information, but that ends up being much less efficient and more complicated. We don't want to require various replication solutions to reimplement logic for this independently. The infrastructure is intended to be generic enough to be reusable. This infrastructure also replaces the 'nodeid' infrastructure of commit timestamps. It is intended to provide all the former capabilities, except that there's only 2^16 different origins; but now they integrate with logical decoding. Additionally more functionality is accessible via SQL. Since the commit timestamp infrastructure has also been introduced in 9.5 (commit 73c986add) changing the API is not a problem. For now the number of origins for which the replication progress can be tracked simultaneously is determined by the max_replication_slots GUC. That GUC is not a perfect match to configure this, but there doesn't seem to be sufficient reason to introduce a separate new one. Bumps both catversion and wal page magic. Author: Andres Freund, with contributions from Petr Jelinek and Craig Ringer Reviewed-By: Heikki Linnakangas, Petr Jelinek, Robert Haas, Steve Singer Discussion: 20150216002155.GI15326@awork2.anarazel.de, 20140923182422.GA15776@alap3.anarazel.de, 20131114172632.GE7522@alap2.anarazel.de
2015-04-29 19:30:53 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
recptr = XLogInsert(RM_HEAP_ID, info);
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
PageSetLSN(page, recptr);
heap' xlog records
2000-06-02 12:20:27 +02:00
}
Restructure local-buffer handling per recent pghackers discussion. The local buffer manager is no longer used for newly-created relations (unless they are TEMP); a new non-TEMP relation goes through the shared bufmgr and thus will participate normally in checkpoints. But TEMP relations use the local buffer manager throughout their lifespan. Also, operations in TEMP relations are not logged in WAL, thus improving performance. Since it's no longer necessary to fsync relations as they move out of the local buffers into shared buffers, quite a lot of smgr.c/md.c/fd.c code is no longer needed and has been removed: there's no concept of a dirty relation anymore in md.c/fd.c, and we never fsync anything but WAL. Still TODO: improve local buffer management algorithms so that it would be reasonable to increase NLocBuffer.
2002-08-06 04:36:35 +02:00
Add more critical-section calls: all code sections that hold spinlocks are now critical sections, so as to ensure die() won't interrupt us while we are munging shared-memory data structures. Avoid insecure intermediate states in some code that proc_exit will call, like palloc/pfree. Rename START/END_CRIT_CODE to START/END_CRIT_SECTION, since that seems to be what people tend to call them anyway, and make them be called with () like a function call, in hopes of not confusing pg_indent. I doubt that this is sufficient to make SIGTERM safe anywhere; there's just too much code that could get invoked during proc_exit().
2001-01-12 22:54:01 +01:00
END_CRIT_SECTION();
heap' xlog records
2000-06-02 12:20:27 +02:00
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
UnlockReleaseBuffer(buffer);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
if (vmbuffer != InvalidBuffer)
ReleaseBuffer(vmbuffer);
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
Correct nasty error in heap_update: it was releasing the buffer refcount before calling RelationInvalidateHeapTuple(), which is bad because the latter needs to look at the tuple data, which is in the shared disk buffer. If another backend manages to recycle the buffer while this is going on, we will compute the wrong hashindex for the tuple or maybe even crash outright. Must hold buffer refcount until afterwards. (This bug is not in 7.0.*; seems to be have introduced during WAL changes.)
2001-01-07 23:14:31 +01:00
/*
Catcaches can now store negative entries as well as positive ones, to speed up repetitive failed searches; per pghackers discussion in late January. inval.c logic substantially simplified, since we can now treat inserts and deletes alike as far as inval events are concerned. Some repair work needed in heap_create_with_catalog, which turns out to have been doing CommandCounterIncrement at a point where the new relation has non-self-consistent catalog entries. With the new inval code, that resulted in assert failures during a relcache entry rebuild.
2002-03-03 18:47:56 +01:00
* If tuple is cachable, mark it for invalidation from the caches in case
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
* we abort. Note it is OK to do this after releasing the buffer, because
* the heaptup data structure is all in local memory, not in the shared
* buffer.
Correct nasty error in heap_update: it was releasing the buffer refcount before calling RelationInvalidateHeapTuple(), which is bad because the latter needs to look at the tuple data, which is in the shared disk buffer. If another backend manages to recycle the buffer while this is going on, we will compute the wrong hashindex for the tuple or maybe even crash outright. Must hold buffer refcount until afterwards. (This bug is not in 7.0.*; seems to be have introduced during WAL changes.)
2001-01-07 23:14:31 +01:00
*/
Revise sinval code to remove no-longer-used tuple TID from inval messages. This requires adjusting the API for syscache callback functions: they now get a hash value, not a TID, to identify the target tuple. Most of them weren't paying any attention to that argument anyway, but plancache did require a small amount of fixing. Also, improve performance a trifle by avoiding sending duplicate inval messages when a heap_update isn't changing the catcache lookup columns.
2011-08-17 01:27:46 +02:00
CacheInvalidateHeapTuple(relation, heaptup, NULL);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
/* Note: speculative insertions are counted too, even if aborted later */
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
pgstat_count_heap_insert(relation, 1);
Revise pgstats stuff to fix the problems with not counting accesses generated by bitmap index scans. Along the way, simplify and speed up the code for counting sequential and index scans; it was both confusing and inefficient to be taking care of that in the per-tuple loops, IMHO. initdb forced because of internal changes in pg_stat view definitions.
2005-10-06 04:29:23 +02:00
Modify tuptoaster's API so that it does not try to modify the passed tuple in-place, but instead passes back an all-new tuple structure if any changes are needed. This is a much cleaner and more robust solution for the bug discovered by Alexey Beschiokov; accordingly, revert the quick hack I installed yesterday. With this change, HeapTupleData.t_datamcxt is no longer needed; will remove it in a separate commit in HEAD only.
2005-11-20 19:38:20 +01:00
/*
* If heaptup is a private copy, release it. Don't forget to copy t_self
* back to the caller's image, too.
*/
if (heaptup != tup)
{
tup->t_self = heaptup->t_self;
heap_freetuple(heaptup);
}
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
}
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
/*
* Subroutine for heap_insert(). Prepares a tuple for insertion. This sets the
Adjust some more comments for WITH OIDS removal. I missed these in 578b229718e8f. Author: Andres Freund
2019-01-21 18:05:51 +01:00
* tuple header fields and toasts the tuple if necessary. Returns a toasted
* version of the tuple if it was toasted, or the original tuple if not. Note
* that in any case, the header fields are also set in the original tuple.
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
*/
static HeapTuple
heap_prepare_insert(Relation relation, HeapTuple tup, TransactionId xid,
CommandId cid, int options)
{
Create an infrastructure for parallel computation in PostgreSQL. This does four basic things. First, it provides convenience routines to coordinate the startup and shutdown of parallel workers. Second, it synchronizes various pieces of state (e.g. GUCs, combo CID mappings, transaction snapshot) from the parallel group leader to the worker processes. Third, it prohibits various operations that would result in unsafe changes to that state while parallelism is active. Finally, it propagates events that would result in an ErrorResponse, NoticeResponse, or NotifyResponse message being sent to the client from the parallel workers back to the master, from which they can then be sent on to the client. Robert Haas, Amit Kapila, Noah Misch, Rushabh Lathia, Jeevan Chalke. Suggestions and review from Andres Freund, Heikki Linnakangas, Noah Misch, Simon Riggs, Euler Taveira, and Jim Nasby.
2015-04-30 21:02:14 +02:00
/*
Fix comments in heapam.c. After commits 85f6b49c2c and 3ba59ccc89, we can allow parallel inserts which was earlier not possible as parallel group members won't conflict for relation extension and page lock. In those commits, we forgot to update comments at few places. Author: Amit Kapila Reviewed-by: Robert Haas and Dilip Kumar Backpatch-through: 13 Discussion: https://postgr.es/m/CAFiTN-tMrQh5FFMPx5aWJ+1gi1H6JxktEhq5mDwCHgnEO5oBkA@mail.gmail.com
2020-09-18 06:10:04 +02:00
* To allow parallel inserts, we need to ensure that they are safe to be
* performed in workers. We have the infrastructure to allow parallel
* inserts in general except for the cases where inserts generate a new
* CommandId (eg. inserts into a table having a foreign key column).
Create an infrastructure for parallel computation in PostgreSQL. This does four basic things. First, it provides convenience routines to coordinate the startup and shutdown of parallel workers. Second, it synchronizes various pieces of state (e.g. GUCs, combo CID mappings, transaction snapshot) from the parallel group leader to the worker processes. Third, it prohibits various operations that would result in unsafe changes to that state while parallelism is active. Finally, it propagates events that would result in an ErrorResponse, NoticeResponse, or NotifyResponse message being sent to the client from the parallel workers back to the master, from which they can then be sent on to the client. Robert Haas, Amit Kapila, Noah Misch, Rushabh Lathia, Jeevan Chalke. Suggestions and review from Andres Freund, Heikki Linnakangas, Noah Misch, Simon Riggs, Euler Taveira, and Jim Nasby.
2015-04-30 21:02:14 +02:00
*/
Allow DML commands that create tables to use parallel query. Haribabu Kommi, reviewed by Dilip Kumar and Rafia Sabih. Various cosmetic changes by me to explain why this appears to be safe but allowing inserts in parallel mode in general wouldn't be. Also, I removed the REFRESH MATERIALIZED VIEW case from Haribabu's patch, since I'm not convinced that case is OK, and hacked on the documentation somewhat. Discussion: http://postgr.es/m/CAJrrPGdo5bak6qnPWe8Kpi8g_jfQEs-G4SYmG9y+OFaw2-dPvA@mail.gmail.com
2017-10-05 17:34:38 +02:00
if (IsParallelWorker())
Create an infrastructure for parallel computation in PostgreSQL. This does four basic things. First, it provides convenience routines to coordinate the startup and shutdown of parallel workers. Second, it synchronizes various pieces of state (e.g. GUCs, combo CID mappings, transaction snapshot) from the parallel group leader to the worker processes. Third, it prohibits various operations that would result in unsafe changes to that state while parallelism is active. Finally, it propagates events that would result in an ErrorResponse, NoticeResponse, or NotifyResponse message being sent to the client from the parallel workers back to the master, from which they can then be sent on to the client. Robert Haas, Amit Kapila, Noah Misch, Rushabh Lathia, Jeevan Chalke. Suggestions and review from Andres Freund, Heikki Linnakangas, Noah Misch, Simon Riggs, Euler Taveira, and Jim Nasby.
2015-04-30 21:02:14 +02:00
ereport(ERROR,
(errcode(ERRCODE_INVALID_TRANSACTION_STATE),
Allow DML commands that create tables to use parallel query. Haribabu Kommi, reviewed by Dilip Kumar and Rafia Sabih. Various cosmetic changes by me to explain why this appears to be safe but allowing inserts in parallel mode in general wouldn't be. Also, I removed the REFRESH MATERIALIZED VIEW case from Haribabu's patch, since I'm not convinced that case is OK, and hacked on the documentation somewhat. Discussion: http://postgr.es/m/CAJrrPGdo5bak6qnPWe8Kpi8g_jfQEs-G4SYmG9y+OFaw2-dPvA@mail.gmail.com
2017-10-05 17:34:38 +02:00
errmsg("cannot insert tuples in a parallel worker")));
Create an infrastructure for parallel computation in PostgreSQL. This does four basic things. First, it provides convenience routines to coordinate the startup and shutdown of parallel workers. Second, it synchronizes various pieces of state (e.g. GUCs, combo CID mappings, transaction snapshot) from the parallel group leader to the worker processes. Third, it prohibits various operations that would result in unsafe changes to that state while parallelism is active. Finally, it propagates events that would result in an ErrorResponse, NoticeResponse, or NotifyResponse message being sent to the client from the parallel workers back to the master, from which they can then be sent on to the client. Robert Haas, Amit Kapila, Noah Misch, Rushabh Lathia, Jeevan Chalke. Suggestions and review from Andres Freund, Heikki Linnakangas, Noah Misch, Simon Riggs, Euler Taveira, and Jim Nasby.
2015-04-30 21:02:14 +02:00
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
tup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
tup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
tup->t_data->t_infomask |= HEAP_XMAX_INVALID;
Change the way we mark tuples as frozen. Instead of changing the tuple xmin to FrozenTransactionId, the combination of HEAP_XMIN_COMMITTED and HEAP_XMIN_INVALID, which were previously never set together, is now defined as HEAP_XMIN_FROZEN. A variety of previous proposals to freeze tuples opportunistically before vacuum_freeze_min_age is reached have foundered on the objection that replacing xmin by FrozenTransactionId might hinder debugging efforts when things in this area go awry; this patch is intended to solve that problem by keeping the XID around (but largely ignoring the value to which it is set). Third-party code that checks for HEAP_XMIN_INVALID on tuples where HEAP_XMIN_COMMITTED might be set will be broken by this change. To fix, use the new accessor macros in htup_details.h rather than consulting the bits directly. HeapTupleHeaderGetXmin has been modified to return FrozenTransactionId when the infomask bits indicate that the tuple is frozen; use HeapTupleHeaderGetRawXmin when you already know that the tuple isn't marked commited or frozen, or want the raw value anyway. We currently do this in routines that display the xmin for user consumption, in tqual.c where it's known to be safe and important for the avoidance of extra cycles, and in the function-caching code for various procedural languages, which shouldn't invalidate the cache just because the tuple gets frozen. Robert Haas and Andres Freund
2013-12-22 21:49:09 +01:00
HeapTupleHeaderSetXmin(tup->t_data, xid);
Reduce scope of changes for COPY FREEZE. Allow support only for freezing tuples by explicit command. Previous coding mistakenly extended slightly beyond what was agreed as correct on -hackers. So essentially a partial revoke of earlier work, leaving just the COPY FREEZE command.
2012-12-02 21:52:52 +01:00
if (options & HEAP_INSERT_FROZEN)
Change the way we mark tuples as frozen. Instead of changing the tuple xmin to FrozenTransactionId, the combination of HEAP_XMIN_COMMITTED and HEAP_XMIN_INVALID, which were previously never set together, is now defined as HEAP_XMIN_FROZEN. A variety of previous proposals to freeze tuples opportunistically before vacuum_freeze_min_age is reached have foundered on the objection that replacing xmin by FrozenTransactionId might hinder debugging efforts when things in this area go awry; this patch is intended to solve that problem by keeping the XID around (but largely ignoring the value to which it is set). Third-party code that checks for HEAP_XMIN_INVALID on tuples where HEAP_XMIN_COMMITTED might be set will be broken by this change. To fix, use the new accessor macros in htup_details.h rather than consulting the bits directly. HeapTupleHeaderGetXmin has been modified to return FrozenTransactionId when the infomask bits indicate that the tuple is frozen; use HeapTupleHeaderGetRawXmin when you already know that the tuple isn't marked commited or frozen, or want the raw value anyway. We currently do this in routines that display the xmin for user consumption, in tqual.c where it's known to be safe and important for the avoidance of extra cycles, and in the function-caching code for various procedural languages, which shouldn't invalidate the cache just because the tuple gets frozen. Robert Haas and Andres Freund
2013-12-22 21:49:09 +01:00
HeapTupleHeaderSetXminFrozen(tup->t_data);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
HeapTupleHeaderSetCmin(tup->t_data, cid);
HeapTupleHeaderSetXmax(tup->t_data, 0); /* for cleanliness */
tup->t_tableOid = RelationGetRelid(relation);
/*
* If the new tuple is too big for storage or contains already toasted
* out-of-line attributes from some other relation, invoke the toaster.
*/
Add a materialized view relations. A materialized view has a rule just like a view and a heap and other physical properties like a table. The rule is only used to populate the table, references in queries refer to the materialized data. This is a minimal implementation, but should still be useful in many cases. Currently data is only populated "on demand" by the CREATE MATERIALIZED VIEW and REFRESH MATERIALIZED VIEW statements. It is expected that future releases will add incremental updates with various timings, and that a more refined concept of defining what is "fresh" data will be developed. At some point it may even be possible to have queries use a materialized in place of references to underlying tables, but that requires the other above-mentioned features to be working first. Much of the documentation work by Robert Haas. Review by Noah Misch, Thom Brown, Robert Haas, Marko Tiikkaja Security review by KaiGai Kohei, with a decision on how best to implement sepgsql still pending.
2013-03-04 01:23:31 +01:00
if (relation->rd_rel->relkind != RELKIND_RELATION &&
relation->rd_rel->relkind != RELKIND_MATVIEW)
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
{
/* toast table entries should never be recursively toasted */
Assert(!HeapTupleHasExternal(tup));
return tup;
}
else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
Rename some toasting functions based on whether they are heap-specific. The old names for the attribute-detoasting functions names included the word "heap," which seems outdated now that the heap is only one of potentially many table access methods. On the other hand, toast_insert_or_update and toast_delete are heap-specific, so rename them by adding "heap_" as a prefix. Not all of the work of making the TOAST system fully accessible to AMs other than the heap is done yet, but there seems to be little harm in getting this renaming out of the way now. Commit 8b94dab06617ef80a0901ab103ebd8754427ef5a already divided up the functions among various files partially according to whether it was intended that they should be heap-specific or AM-agnostic, so this is just clarifying the division contemplated by that commit. Patch by me, reviewed and tested by Prabhat Sabu, Thomas Munro, Andres Freund, and Álvaro Herrera. Discussion: http://postgr.es/m/CA+TgmoZv-=2iWM4jcw5ZhJeL18HF96+W1yJeYrnGMYdkFFnEpQ@mail.gmail.com
2019-10-04 20:24:46 +02:00
return heap_toast_insert_or_update(relation, tup, NULL, options);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
else
return tup;
}
heapam: Pass number of required pages to RelationGetBufferForTuple() A future commit will use this information to determine how aggressively to extend the relation by. In heap_multi_insert() we know accurately how many pages we need once we need to extend the relation, providing an accurate lower bound for how much to extend. Reviewed-by: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://postgr.es/m/20221029025420.eplyow6k7tgu6he3@awork3.anarazel.de
2023-04-06 23:18:24 +02:00
/*
* Helper for heap_multi_insert() that computes the number of entire pages
* that inserting the remaining heaptuples requires. Used to determine how
* much the relation needs to be extended by.
*/
static int
heap_multi_insert_pages(HeapTuple *heaptuples, int done, int ntuples, Size saveFreeSpace)
{
size_t page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
int npages = 1;
for (int i = done; i < ntuples; i++)
{
size_t tup_sz = sizeof(ItemIdData) + MAXALIGN(heaptuples[i]->t_len);
if (page_avail < tup_sz)
{
npages++;
page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
}
page_avail -= tup_sz;
}
return npages;
}
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
/*
Fix typo in heapam.c Author: Hou Zhijie Discussion: https://postgr.es/m/OS0PR01MB571612191738540B27A8DE5894249@OS0PR01MB5716.jpnprd01.prod.outlook.com
2021-05-26 12:53:03 +02:00
* heap_multi_insert - insert multiple tuples into a heap
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
*
* This is like heap_insert(), but inserts multiple tuples in one operation.
* That's faster than calling heap_insert() in a loop, because when multiple
* tuples can be inserted on a single page, we can write just a single WAL
* record covering all of them, and only need to lock/unlock the page once.
*
* Note: this leaks memory into the current memory context. You can create a
* temporary context before calling this, if that's a problem.
*/
void
tableam: Add table_multi_insert() and revamp/speed-up COPY FROM buffering. This adds table_multi_insert(), and converts COPY FROM, the only user of heap_multi_insert, to it. A simple conversion of COPY FROM use slots would have yielded a slowdown when inserting into a partitioned table for some workloads. Different partitions might need different slots (both slot types and their descriptors), and dropping / creating slots when there's constant partition changes is measurable. Thus instead revamp the COPY FROM buffering for partitioned tables to allow to buffer inserts into multiple tables, flushing only when limits are reached across all partition buffers. By only dropping slots when there've been inserts into too many different partitions, the aforementioned overhead is gone. By allowing larger batches, even when there are frequent partition changes, we actuall speed such cases up significantly. By using slots COPY of very narrow rows into unlogged / temporary might slow down very slightly (due to the indirect function calls). Author: David Rowley, Andres Freund, Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190327054923.t3epfuewxfqdt22e@alap3.anarazel.de
2019-04-05 00:47:19 +02:00
heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
CommandId cid, int options, BulkInsertState bistate)
{
TransactionId xid = GetCurrentTransactionId();
HeapTuple *heaptuples;
int i;
int ndone;
Avoid using potentially-under-aligned page buffers. There's a project policy against using plain "char buf[BLCKSZ]" local or static variables as page buffers; preferred style is to palloc or malloc each buffer to ensure it is MAXALIGN'd. However, that policy's been ignored in an increasing number of places. We've apparently got away with it so far, probably because (a) relatively few people use platforms on which misalignment causes core dumps and/or (b) the variables chance to be sufficiently aligned anyway. But this is not something to rely on. Moreover, even if we don't get a core dump, we might be paying a lot of cycles for misaligned accesses. To fix, invent new union types PGAlignedBlock and PGAlignedXLogBlock that the compiler must allocate with sufficient alignment, and use those in place of plain char arrays. I used these types even for variables where there's no risk of a misaligned access, since ensuring proper alignment should make kernel data transfers faster. I also changed some places where we had been palloc'ing short-lived buffers, for coding style uniformity and to save palloc/pfree overhead. Since this seems to be a live portability hazard (despite the lack of field reports), back-patch to all supported versions. Patch by me; thanks to Michael Paquier for review. Discussion: https://postgr.es/m/1535618100.1286.3.camel@credativ.de
2018-09-01 21:27:12 +02:00
PGAlignedBlock scratch;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
Page page;
Set PD_ALL_VISIBLE and visibility map bits in COPY FREEZE Make sure COPY FREEZE marks the pages as PD_ALL_VISIBLE and updates the visibility map. Until now we only marked individual tuples as frozen, but page-level flags were not updated, so the first VACUUM after the COPY FREEZE had to rewrite the whole table. This is a fairly old patch, and multiple people worked on it. The first version was written by Jeff Janes, and then reworked by Pavan Deolasee and Anastasia Lubennikova. Author: Anastasia Lubennikova, Pavan Deolasee, Jeff Janes Reviewed-by: Kuntal Ghosh, Jeff Janes, Tomas Vondra, Masahiko Sawada, Andres Freund, Ibrar Ahmed, Robert Haas, Tatsuro Ishii, Darafei Praliaskouski Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com Discussion: https://postgr.es/m/CAMkU%3D1w3osJJ2FneELhhNRLxfZitDgp9FPHee08NT2FQFmz_pQ%40mail.gmail.com
2021-01-17 22:11:39 +01:00
Buffer vmbuffer = InvalidBuffer;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
bool needwal;
Take fillfactor into account in the new COPY bulk heap insert code. Jeff Janes
2011-11-26 11:11:00 +01:00
Size saveFreeSpace;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
bool need_tuple_data = RelationIsLogicallyLogged(relation);
bool need_cids = RelationIsAccessibleInLogicalDecoding(relation);
heapam: Pass number of required pages to RelationGetBufferForTuple() A future commit will use this information to determine how aggressively to extend the relation by. In heap_multi_insert() we know accurately how many pages we need once we need to extend the relation, providing an accurate lower bound for how much to extend. Reviewed-by: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://postgr.es/m/20221029025420.eplyow6k7tgu6he3@awork3.anarazel.de
2023-04-06 23:18:24 +02:00
bool starting_with_empty_page = false;
int npages = 0;
int npages_used = 0;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
Fix logical decoding error when system table w/ toast is repeatedly rewritten. Repeatedly rewriting a mapped catalog table with VACUUM FULL or CLUSTER could cause logical decoding to fail with: ERROR, "could not map filenode \"%s\" to relation OID" To trigger the problem the rewritten catalog had to have live tuples with toasted columns. The problem was triggered as during catalog table rewrites the heap_insert() check that prevents logical decoding information to be emitted for system catalogs, failed to treat the new heap's toast table as a system catalog (because the new heap is not recognized as a catalog table via RelationIsLogicallyLogged()). The relmapper, in contrast to the normal catalog contents, does not contain historical information. After a single rewrite of a mapped table the new relation is known to the relmapper, but if the table is rewritten twice before logical decoding occurs, the relfilenode cannot be mapped to a relation anymore. Which then leads us to error out. This only happens for toast tables, because the main table contents aren't re-inserted with heap_insert(). The fix is simple, add a new heap_insert() flag that prevents logical decoding information from being emitted, and accept during decoding that there might not be tuple data for toast tables. Unfortunately that does not fix pre-existing logical decoding errors. Doing so would require not throwing an error when a filenode cannot be mapped to a relation during decoding, and that seems too likely to hide bugs. If it's crucial to fix decoding for an existing slot, temporarily changing the ERROR in ReorderBufferCommit() to a WARNING appears to be the best fix. Author: Andres Freund Discussion: https://postgr.es/m/20180914021046.oi7dm4ra3ot2g2kt@alap3.anarazel.de Backpatch: 9.4-, where logical decoding was introduced
2018-10-10 22:53:02 +02:00
/* currently not needed (thus unsupported) for heap_multi_insert() */
Remove AssertArg and AssertState These don't offer anything over plain Assert, and their usage had already been declared obsolescent. Author: Nathan Bossart <nathandbossart@gmail.com> Reviewed-by: Michael Paquier <michael@paquier.xyz> Discussion: https://www.postgresql.org/message-id/20221009210148.GA900071@nathanxps13
2022-10-28 09:19:06 +02:00
Assert(!(options & HEAP_INSERT_NO_LOGICAL));
Fix logical decoding error when system table w/ toast is repeatedly rewritten. Repeatedly rewriting a mapped catalog table with VACUUM FULL or CLUSTER could cause logical decoding to fail with: ERROR, "could not map filenode \"%s\" to relation OID" To trigger the problem the rewritten catalog had to have live tuples with toasted columns. The problem was triggered as during catalog table rewrites the heap_insert() check that prevents logical decoding information to be emitted for system catalogs, failed to treat the new heap's toast table as a system catalog (because the new heap is not recognized as a catalog table via RelationIsLogicallyLogged()). The relmapper, in contrast to the normal catalog contents, does not contain historical information. After a single rewrite of a mapped table the new relation is known to the relmapper, but if the table is rewritten twice before logical decoding occurs, the relfilenode cannot be mapped to a relation anymore. Which then leads us to error out. This only happens for toast tables, because the main table contents aren't re-inserted with heap_insert(). The fix is simple, add a new heap_insert() flag that prevents logical decoding information from being emitted, and accept during decoding that there might not be tuple data for toast tables. Unfortunately that does not fix pre-existing logical decoding errors. Doing so would require not throwing an error when a filenode cannot be mapped to a relation during decoding, and that seems too likely to hide bugs. If it's crucial to fix decoding for an existing slot, temporarily changing the ERROR in ReorderBufferCommit() to a WARNING appears to be the best fix. Author: Andres Freund Discussion: https://postgr.es/m/20180914021046.oi7dm4ra3ot2g2kt@alap3.anarazel.de Backpatch: 9.4-, where logical decoding was introduced
2018-10-10 22:53:02 +02:00
Skip WAL for new relfilenodes, under wal_level=minimal. Until now, only selected bulk operations (e.g. COPY) did this. If a given relfilenode received both a WAL-skipping COPY and a WAL-logged operation (e.g. INSERT), recovery could lose tuples from the COPY. See src/backend/access/transam/README section "Skipping WAL for New RelFileNode" for the new coding rules. Maintainers of table access methods should examine that section. To maintain data durability, just before commit, we choose between an fsync of the relfilenode and copying its contents to WAL. A new GUC, wal_skip_threshold, guides that choice. If this change slows a workload that creates small, permanent relfilenodes under wal_level=minimal, try adjusting wal_skip_threshold. Users setting a timeout on COMMIT may need to adjust that timeout, and log_min_duration_statement analysis will reflect time consumption moving to COMMIT from commands like COPY. Internally, this requires a reliable determination of whether RollbackAndReleaseCurrentSubTransaction() would unlink a relation's current relfilenode. Introduce rd_firstRelfilenodeSubid. Amend the specification of rd_createSubid such that the field is zero when a new rel has an old rd_node. Make relcache.c retain entries for certain dropped relations until end of transaction. Bump XLOG_PAGE_MAGIC, since this introduces XLOG_GIST_ASSIGN_LSN. Future servers accept older WAL, so this bump is discretionary. Kyotaro Horiguchi, reviewed (in earlier, similar versions) by Robert Haas. Heikki Linnakangas and Michael Paquier implemented earlier designs that materially clarified the problem. Reviewed, in earlier designs, by Andrew Dunstan, Andres Freund, Alvaro Herrera, Tom Lane, Fujii Masao, and Simon Riggs. Reported by Martijn van Oosterhout. Discussion: https://postgr.es/m/20150702220524.GA9392@svana.org
2020-04-04 21:25:34 +02:00
needwal = RelationNeedsWAL(relation);
Take fillfactor into account in the new COPY bulk heap insert code. Jeff Janes
2011-11-26 11:11:00 +01:00
saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
HEAP_DEFAULT_FILLFACTOR);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
tableam: Add table_multi_insert() and revamp/speed-up COPY FROM buffering. This adds table_multi_insert(), and converts COPY FROM, the only user of heap_multi_insert, to it. A simple conversion of COPY FROM use slots would have yielded a slowdown when inserting into a partitioned table for some workloads. Different partitions might need different slots (both slot types and their descriptors), and dropping / creating slots when there's constant partition changes is measurable. Thus instead revamp the COPY FROM buffering for partitioned tables to allow to buffer inserts into multiple tables, flushing only when limits are reached across all partition buffers. By only dropping slots when there've been inserts into too many different partitions, the aforementioned overhead is gone. By allowing larger batches, even when there are frequent partition changes, we actuall speed such cases up significantly. By using slots COPY of very narrow rows into unlogged / temporary might slow down very slightly (due to the indirect function calls). Author: David Rowley, Andres Freund, Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190327054923.t3epfuewxfqdt22e@alap3.anarazel.de
2019-04-05 00:47:19 +02:00
/* Toast and set header data in all the slots */
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
heaptuples = palloc(ntuples * sizeof(HeapTuple));
for (i = 0; i < ntuples; i++)
tableam: Add table_multi_insert() and revamp/speed-up COPY FROM buffering. This adds table_multi_insert(), and converts COPY FROM, the only user of heap_multi_insert, to it. A simple conversion of COPY FROM use slots would have yielded a slowdown when inserting into a partitioned table for some workloads. Different partitions might need different slots (both slot types and their descriptors), and dropping / creating slots when there's constant partition changes is measurable. Thus instead revamp the COPY FROM buffering for partitioned tables to allow to buffer inserts into multiple tables, flushing only when limits are reached across all partition buffers. By only dropping slots when there've been inserts into too many different partitions, the aforementioned overhead is gone. By allowing larger batches, even when there are frequent partition changes, we actuall speed such cases up significantly. By using slots COPY of very narrow rows into unlogged / temporary might slow down very slightly (due to the indirect function calls). Author: David Rowley, Andres Freund, Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190327054923.t3epfuewxfqdt22e@alap3.anarazel.de
2019-04-05 00:47:19 +02:00
{
HeapTuple tuple;
tuple = ExecFetchSlotHeapTuple(slots[i], true, NULL);
slots[i]->tts_tableOid = RelationGetRelid(relation);
tuple->t_tableOid = slots[i]->tts_tableOid;
heaptuples[i] = heap_prepare_insert(relation, tuple, xid, cid,
options);
}
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
/*
* We're about to do the actual inserts -- but check for conflict first,
Fix serialization anomalies due to race conditions on INSERT. On insert the CheckForSerializableConflictIn() test was performed before the page(s) which were going to be modified had been locked (with an exclusive buffer content lock). If another process acquired a relation SIReadLock on the heap and scanned to a page on which an insert was going to occur before the page was so locked, a rw-conflict would be missed, which could allow a serialization anomaly to be missed. The window between the check and the page lock was small, so the bug was generally not noticed unless there was high concurrency with multiple processes inserting into the same table. This was reported by Peter Bailis as bug #11732, by Sean Chittenden as bug #13667, and by others. The race condition was eliminated in heap_insert() by moving the check down below the acquisition of the buffer lock, which had been the very next statement. Because of the loop locking and unlocking multiple buffers in heap_multi_insert() a check was added after all inserts were completed. The check before the start of the inserts was left because it might avoid a large amount of work to detect a serialization anomaly before performing the all of the inserts and the related WAL logging. While investigating this bug, other SSI bugs which were even harder to hit in practice were noticed and fixed, an unnecessary check (covered by another check, so redundant) was removed from heap_update(), and comments were improved. Back-patch to all supported branches. Kevin Grittner and Thomas Munro
2015-10-31 20:43:34 +01:00
* to minimize the possibility of having to roll back work we've just
* done.
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
*
Fix serialization anomalies due to race conditions on INSERT. On insert the CheckForSerializableConflictIn() test was performed before the page(s) which were going to be modified had been locked (with an exclusive buffer content lock). If another process acquired a relation SIReadLock on the heap and scanned to a page on which an insert was going to occur before the page was so locked, a rw-conflict would be missed, which could allow a serialization anomaly to be missed. The window between the check and the page lock was small, so the bug was generally not noticed unless there was high concurrency with multiple processes inserting into the same table. This was reported by Peter Bailis as bug #11732, by Sean Chittenden as bug #13667, and by others. The race condition was eliminated in heap_insert() by moving the check down below the acquisition of the buffer lock, which had been the very next statement. Because of the loop locking and unlocking multiple buffers in heap_multi_insert() a check was added after all inserts were completed. The check before the start of the inserts was left because it might avoid a large amount of work to detect a serialization anomaly before performing the all of the inserts and the related WAL logging. While investigating this bug, other SSI bugs which were even harder to hit in practice were noticed and fixed, an unnecessary check (covered by another check, so redundant) was removed from heap_update(), and comments were improved. Back-patch to all supported branches. Kevin Grittner and Thomas Munro
2015-10-31 20:43:34 +01:00
* A check here does not definitively prevent a serialization anomaly;
* that check MUST be done at least past the point of acquiring an
* exclusive buffer content lock on every buffer that will be affected,
* and MAY be done after all inserts are reflected in the buffers and
Fix comment in heapam.c Improvement per suggestion from Tom Lane. Author: Daniel Gustafsson Discussion: https://postgr.es/m/FED18699-4270-4778-8DA8-10F119A5ECF3@yesql.se
2020-01-13 09:57:38 +01:00
* those locks are released; otherwise there is a race condition. Since
Fix serialization anomalies due to race conditions on INSERT. On insert the CheckForSerializableConflictIn() test was performed before the page(s) which were going to be modified had been locked (with an exclusive buffer content lock). If another process acquired a relation SIReadLock on the heap and scanned to a page on which an insert was going to occur before the page was so locked, a rw-conflict would be missed, which could allow a serialization anomaly to be missed. The window between the check and the page lock was small, so the bug was generally not noticed unless there was high concurrency with multiple processes inserting into the same table. This was reported by Peter Bailis as bug #11732, by Sean Chittenden as bug #13667, and by others. The race condition was eliminated in heap_insert() by moving the check down below the acquisition of the buffer lock, which had been the very next statement. Because of the loop locking and unlocking multiple buffers in heap_multi_insert() a check was added after all inserts were completed. The check before the start of the inserts was left because it might avoid a large amount of work to detect a serialization anomaly before performing the all of the inserts and the related WAL logging. While investigating this bug, other SSI bugs which were even harder to hit in practice were noticed and fixed, an unnecessary check (covered by another check, so redundant) was removed from heap_update(), and comments were improved. Back-patch to all supported branches. Kevin Grittner and Thomas Munro
2015-10-31 20:43:34 +01:00
* multiple buffers can be locked and unlocked in the loop below, and it
* would not be feasible to identify and lock all of those buffers before
* the loop, we must do a final check at the end.
*
* The check here could be omitted with no loss of correctness; it is
* present strictly as an optimization.
*
* For heap inserts, we only need to check for table-level SSI locks. Our
* new tuples can't possibly conflict with existing tuple locks, and heap
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
* page locks are only consolidated versions of tuple locks; they do not
Fix serialization anomalies due to race conditions on INSERT. On insert the CheckForSerializableConflictIn() test was performed before the page(s) which were going to be modified had been locked (with an exclusive buffer content lock). If another process acquired a relation SIReadLock on the heap and scanned to a page on which an insert was going to occur before the page was so locked, a rw-conflict would be missed, which could allow a serialization anomaly to be missed. The window between the check and the page lock was small, so the bug was generally not noticed unless there was high concurrency with multiple processes inserting into the same table. This was reported by Peter Bailis as bug #11732, by Sean Chittenden as bug #13667, and by others. The race condition was eliminated in heap_insert() by moving the check down below the acquisition of the buffer lock, which had been the very next statement. Because of the loop locking and unlocking multiple buffers in heap_multi_insert() a check was added after all inserts were completed. The check before the start of the inserts was left because it might avoid a large amount of work to detect a serialization anomaly before performing the all of the inserts and the related WAL logging. While investigating this bug, other SSI bugs which were even harder to hit in practice were noticed and fixed, an unnecessary check (covered by another check, so redundant) was removed from heap_update(), and comments were improved. Back-patch to all supported branches. Kevin Grittner and Thomas Munro
2015-10-31 20:43:34 +01:00
* lock "gaps" as index page locks do. So we don't need to specify a
* buffer when making the call, which makes for a faster check.
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
*/
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
ndone = 0;
while (ndone < ntuples)
{
Fix bugs in the COPY heap-insert batching patch. Forgot to call RestoreBkpBlocks() in the redo-function, as pointed out by Simon Riggs. In redo of a regular heap insert, it's taken care of in heap_redo(), but this new record type uses the heap2 RM, and heap2_redo() does not take care of that for you. Also, failed to reset the vmbuffer and all_visibile_cleared local variables after switching to a new buffer.
2011-11-09 20:24:26 +01:00
Buffer buffer;
bool all_visible_cleared = false;
Set PD_ALL_VISIBLE and visibility map bits in COPY FREEZE Make sure COPY FREEZE marks the pages as PD_ALL_VISIBLE and updates the visibility map. Until now we only marked individual tuples as frozen, but page-level flags were not updated, so the first VACUUM after the COPY FREEZE had to rewrite the whole table. This is a fairly old patch, and multiple people worked on it. The first version was written by Jeff Janes, and then reworked by Pavan Deolasee and Anastasia Lubennikova. Author: Anastasia Lubennikova, Pavan Deolasee, Jeff Janes Reviewed-by: Kuntal Ghosh, Jeff Janes, Tomas Vondra, Masahiko Sawada, Andres Freund, Ibrar Ahmed, Robert Haas, Tatsuro Ishii, Darafei Praliaskouski Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com Discussion: https://postgr.es/m/CAMkU%3D1w3osJJ2FneELhhNRLxfZitDgp9FPHee08NT2FQFmz_pQ%40mail.gmail.com
2021-01-17 22:11:39 +01:00
bool all_frozen_set = false;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
int nthispage;
Check for interrupts during tuple-insertion loops. Normally, this won't matter too much; but if I/O is really slow, for example because the system is overloaded, we might write many pages before checking for interrupts. A single toast insertion might write up to 1GB of data, and a multi-insert could write hundreds of tuples (and their corresponding TOAST data).
2014-06-24 03:45:21 +02:00
CHECK_FOR_INTERRUPTS();
heapam: Pass number of required pages to RelationGetBufferForTuple() A future commit will use this information to determine how aggressively to extend the relation by. In heap_multi_insert() we know accurately how many pages we need once we need to extend the relation, providing an accurate lower bound for how much to extend. Reviewed-by: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://postgr.es/m/20221029025420.eplyow6k7tgu6he3@awork3.anarazel.de
2023-04-06 23:18:24 +02:00
/*
* Compute number of pages needed to fit the to-be-inserted tuples in
* the worst case. This will be used to determine how much to extend
* the relation by in RelationGetBufferForTuple(), if needed. If we
* filled a prior page from scratch, we can just update our last
* computation, but if we started with a partially filled page,
* recompute from scratch, the number of potentially required pages
* can vary due to tuples needing to fit onto the page, page headers
* etc.
*/
if (ndone == 0 || !starting_with_empty_page)
{
npages = heap_multi_insert_pages(heaptuples, ndone, ntuples,
saveFreeSpace);
npages_used = 0;
}
else
npages_used++;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
/*
* Find buffer where at least the next tuple will fit. If the page is
* all-visible, this will also pin the requisite visibility map page.
Set PD_ALL_VISIBLE and visibility map bits in COPY FREEZE Make sure COPY FREEZE marks the pages as PD_ALL_VISIBLE and updates the visibility map. Until now we only marked individual tuples as frozen, but page-level flags were not updated, so the first VACUUM after the COPY FREEZE had to rewrite the whole table. This is a fairly old patch, and multiple people worked on it. The first version was written by Jeff Janes, and then reworked by Pavan Deolasee and Anastasia Lubennikova. Author: Anastasia Lubennikova, Pavan Deolasee, Jeff Janes Reviewed-by: Kuntal Ghosh, Jeff Janes, Tomas Vondra, Masahiko Sawada, Andres Freund, Ibrar Ahmed, Robert Haas, Tatsuro Ishii, Darafei Praliaskouski Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com Discussion: https://postgr.es/m/CAMkU%3D1w3osJJ2FneELhhNRLxfZitDgp9FPHee08NT2FQFmz_pQ%40mail.gmail.com
2021-01-17 22:11:39 +01:00
*
* Also pin visibility map page if COPY FREEZE inserts tuples into an
* empty page. See all_frozen_set below.
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
*/
buffer = RelationGetBufferForTuple(relation, heaptuples[ndone]->t_len,
InvalidBuffer, options, bistate,
heapam: Pass number of required pages to RelationGetBufferForTuple() A future commit will use this information to determine how aggressively to extend the relation by. In heap_multi_insert() we know accurately how many pages we need once we need to extend the relation, providing an accurate lower bound for how much to extend. Reviewed-by: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://postgr.es/m/20221029025420.eplyow6k7tgu6he3@awork3.anarazel.de
2023-04-06 23:18:24 +02:00
&vmbuffer, NULL,
npages - npages_used);
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(buffer);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
Set PD_ALL_VISIBLE and visibility map bits in COPY FREEZE Make sure COPY FREEZE marks the pages as PD_ALL_VISIBLE and updates the visibility map. Until now we only marked individual tuples as frozen, but page-level flags were not updated, so the first VACUUM after the COPY FREEZE had to rewrite the whole table. This is a fairly old patch, and multiple people worked on it. The first version was written by Jeff Janes, and then reworked by Pavan Deolasee and Anastasia Lubennikova. Author: Anastasia Lubennikova, Pavan Deolasee, Jeff Janes Reviewed-by: Kuntal Ghosh, Jeff Janes, Tomas Vondra, Masahiko Sawada, Andres Freund, Ibrar Ahmed, Robert Haas, Tatsuro Ishii, Darafei Praliaskouski Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com Discussion: https://postgr.es/m/CAMkU%3D1w3osJJ2FneELhhNRLxfZitDgp9FPHee08NT2FQFmz_pQ%40mail.gmail.com
2021-01-17 22:11:39 +01:00
starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
if (starting_with_empty_page && (options & HEAP_INSERT_FROZEN))
all_frozen_set = true;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
/* NO EREPORT(ERROR) from here till changes are logged */
START_CRIT_SECTION();
In multi-insert, don't go into infinite loop on a huge tuple and fillfactor. If a tuple is larger than page size minus space reserved for fillfactor, heap_multi_insert would never find a page that it fits in and repeatedly ask for a new page from RelationGetBufferForTuple. If a tuple is too large to fit on any page, taking fillfactor into account, RelationGetBufferForTuple will always expand the relation. In a normal insert, heap_insert will accept that and put the tuple on the new page. heap_multi_insert, however, does a fillfactor check of its own, and doesn't accept the newly-extended page RelationGetBufferForTuple returns, even though there is no other choice to make the tuple fit. Fix that by making the logic in heap_multi_insert more like the heap_insert logic. The first tuple is always put on the page RelationGetBufferForTuple gives us, and the fillfactor check is only applied to the subsequent tuples. Report from David Gould, although I didn't use his patch.
2012-12-12 12:34:03 +01:00
/*
* RelationGetBufferForTuple has ensured that the first tuple fits.
* Put that on the page, and then as many other tuples as fit.
*/
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
RelationPutHeapTuple(relation, buffer, heaptuples[ndone], false);
Issue properly WAL record for CID of first catalog tuple in multi-insert Multi-insert for heap is not yet used actively for catalogs, but the code to support this case is in place for logical decoding. The existing code forgot to issue a XLOG_HEAP2_NEW_CID record for the first tuple inserted, leading to failures when attempting to use multiple inserts for catalogs at decoding time. This commit fixes the problem by WAL-logging the needed CID. This is not an active bug, so no back-patch is done. Author: Daniel Gustafsson Discussion: https://postgr.es/m/E0D4CC67-A1CF-4DF4-991D-B3AC2EB5FAE9@yesql.se
2020-02-24 23:55:22 +01:00
/*
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
* For logical decoding we need combo CIDs to properly decode the
Use multi-inserts for pg_attribute and pg_shdepend For pg_attribute, this allows to insert at once a full set of attributes for a relation (roughly 15% of WAL reduction in extreme cases). For pg_shdepend, this reduces the work done when creating new shared dependencies from a database template. The number of slots used for the insertion is capped at 64kB of data inserted for both, depending on the number of items to insert and the length of the rows involved. More can be done for other catalogs, like pg_depend. This part requires a different approach as the number of slots to use depends also on the number of entries discarded as pinned dependencies. This is also related to the rework or dependency handling for ALTER TABLE and CREATE TABLE, mainly. Author: Daniel Gustafsson Reviewed-by: Andres Freund, Michael Paquier Discussion: https://postgr.es/m/20190213182737.mxn6hkdxwrzgxk35@alap3.anarazel.de
2020-07-31 03:54:26 +02:00
* catalog.
Issue properly WAL record for CID of first catalog tuple in multi-insert Multi-insert for heap is not yet used actively for catalogs, but the code to support this case is in place for logical decoding. The existing code forgot to issue a XLOG_HEAP2_NEW_CID record for the first tuple inserted, leading to failures when attempting to use multiple inserts for catalogs at decoding time. This commit fixes the problem by WAL-logging the needed CID. This is not an active bug, so no back-patch is done. Author: Daniel Gustafsson Discussion: https://postgr.es/m/E0D4CC67-A1CF-4DF4-991D-B3AC2EB5FAE9@yesql.se
2020-02-24 23:55:22 +01:00
*/
if (needwal && need_cids)
log_heap_new_cid(relation, heaptuples[ndone]);
In multi-insert, don't go into infinite loop on a huge tuple and fillfactor. If a tuple is larger than page size minus space reserved for fillfactor, heap_multi_insert would never find a page that it fits in and repeatedly ask for a new page from RelationGetBufferForTuple. If a tuple is too large to fit on any page, taking fillfactor into account, RelationGetBufferForTuple will always expand the relation. In a normal insert, heap_insert will accept that and put the tuple on the new page. heap_multi_insert, however, does a fillfactor check of its own, and doesn't accept the newly-extended page RelationGetBufferForTuple returns, even though there is no other choice to make the tuple fit. Fix that by making the logic in heap_multi_insert more like the heap_insert logic. The first tuple is always put on the page RelationGetBufferForTuple gives us, and the fillfactor check is only applied to the subsequent tuples. Report from David Gould, although I didn't use his patch.
2012-12-12 12:34:03 +01:00
for (nthispage = 1; ndone + nthispage < ntuples; nthispage++)
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
{
HeapTuple heaptup = heaptuples[ndone + nthispage];
Fix bug in freespace calculation in heap_multi_insert(). If the amount of freespace on page was less than the amount reserved by fillfactor, the calculation would underflow. This fixes bug #6643 reported by Tomonari Katsumata.
2012-05-16 13:10:29 +02:00
if (PageGetHeapFreeSpace(page) < MAXALIGN(heaptup->t_len) + saveFreeSpace)
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
break;
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
RelationPutHeapTuple(relation, buffer, heaptup, false);
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
/*
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
* For logical decoding we need combo CIDs to properly decode the
Use multi-inserts for pg_attribute and pg_shdepend For pg_attribute, this allows to insert at once a full set of attributes for a relation (roughly 15% of WAL reduction in extreme cases). For pg_shdepend, this reduces the work done when creating new shared dependencies from a database template. The number of slots used for the insertion is capped at 64kB of data inserted for both, depending on the number of items to insert and the length of the rows involved. More can be done for other catalogs, like pg_depend. This part requires a different approach as the number of slots to use depends also on the number of entries discarded as pinned dependencies. This is also related to the rework or dependency handling for ALTER TABLE and CREATE TABLE, mainly. Author: Daniel Gustafsson Reviewed-by: Andres Freund, Michael Paquier Discussion: https://postgr.es/m/20190213182737.mxn6hkdxwrzgxk35@alap3.anarazel.de
2020-07-31 03:54:26 +02:00
* catalog.
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
*/
if (needwal && need_cids)
log_heap_new_cid(relation, heaptup);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
}
Set PD_ALL_VISIBLE and visibility map bits in COPY FREEZE Make sure COPY FREEZE marks the pages as PD_ALL_VISIBLE and updates the visibility map. Until now we only marked individual tuples as frozen, but page-level flags were not updated, so the first VACUUM after the COPY FREEZE had to rewrite the whole table. This is a fairly old patch, and multiple people worked on it. The first version was written by Jeff Janes, and then reworked by Pavan Deolasee and Anastasia Lubennikova. Author: Anastasia Lubennikova, Pavan Deolasee, Jeff Janes Reviewed-by: Kuntal Ghosh, Jeff Janes, Tomas Vondra, Masahiko Sawada, Andres Freund, Ibrar Ahmed, Robert Haas, Tatsuro Ishii, Darafei Praliaskouski Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com Discussion: https://postgr.es/m/CAMkU%3D1w3osJJ2FneELhhNRLxfZitDgp9FPHee08NT2FQFmz_pQ%40mail.gmail.com
2021-01-17 22:11:39 +01:00
/*
* If the page is all visible, need to clear that, unless we're only
* going to add further frozen rows to it.
*
* If we're only adding already frozen rows to a previously empty
* page, mark it as all-visible.
*/
if (PageIsAllVisible(page) && !(options & HEAP_INSERT_FROZEN))
Fix more crash-safe visibility map bugs, and improve comments. In lazy_scan_heap, we could issue bogus warnings about incorrect information in the visibility map, because we checked the visibility map bit before locking the heap page, creating a race condition. Fix by rechecking the visibility map bit before we complain. Rejigger some related logic so that we rely on the possibly-outdated all_visible_according_to_vm value as little as possible. In heap_multi_insert, it's not safe to clear the visibility map bit before beginning the critical section. The visibility map is not crash-safe unless we treat clearing the bit as a critical operation. Specifically, if the transaction were to error out after we set the bit and before entering the critical section, we could end up writing the heap page to disk (with the bit cleared) and crashing before the visibility map page made it to disk. That would be bad. heap_insert has this correct, but somehow the order of operations got rearranged when heap_multi_insert was added. Also, add some more comments to visibilitymap_test, lazy_scan_heap, and IndexOnlyNext, expounding on concurrency issues. Per extensive code review by Andres Freund, and further review by Tom Lane, who also made the original report about the bogus warnings.
2012-06-07 18:25:41 +02:00
{
all_visible_cleared = true;
PageClearAllVisible(page);
visibilitymap_clear(relation,
BufferGetBlockNumber(buffer),
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
vmbuffer, VISIBILITYMAP_VALID_BITS);
Fix more crash-safe visibility map bugs, and improve comments. In lazy_scan_heap, we could issue bogus warnings about incorrect information in the visibility map, because we checked the visibility map bit before locking the heap page, creating a race condition. Fix by rechecking the visibility map bit before we complain. Rejigger some related logic so that we rely on the possibly-outdated all_visible_according_to_vm value as little as possible. In heap_multi_insert, it's not safe to clear the visibility map bit before beginning the critical section. The visibility map is not crash-safe unless we treat clearing the bit as a critical operation. Specifically, if the transaction were to error out after we set the bit and before entering the critical section, we could end up writing the heap page to disk (with the bit cleared) and crashing before the visibility map page made it to disk. That would be bad. heap_insert has this correct, but somehow the order of operations got rearranged when heap_multi_insert was added. Also, add some more comments to visibilitymap_test, lazy_scan_heap, and IndexOnlyNext, expounding on concurrency issues. Per extensive code review by Andres Freund, and further review by Tom Lane, who also made the original report about the bogus warnings.
2012-06-07 18:25:41 +02:00
}
Set PD_ALL_VISIBLE and visibility map bits in COPY FREEZE Make sure COPY FREEZE marks the pages as PD_ALL_VISIBLE and updates the visibility map. Until now we only marked individual tuples as frozen, but page-level flags were not updated, so the first VACUUM after the COPY FREEZE had to rewrite the whole table. This is a fairly old patch, and multiple people worked on it. The first version was written by Jeff Janes, and then reworked by Pavan Deolasee and Anastasia Lubennikova. Author: Anastasia Lubennikova, Pavan Deolasee, Jeff Janes Reviewed-by: Kuntal Ghosh, Jeff Janes, Tomas Vondra, Masahiko Sawada, Andres Freund, Ibrar Ahmed, Robert Haas, Tatsuro Ishii, Darafei Praliaskouski Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com Discussion: https://postgr.es/m/CAMkU%3D1w3osJJ2FneELhhNRLxfZitDgp9FPHee08NT2FQFmz_pQ%40mail.gmail.com
2021-01-17 22:11:39 +01:00
else if (all_frozen_set)
PageSetAllVisible(page);
Fix more crash-safe visibility map bugs, and improve comments. In lazy_scan_heap, we could issue bogus warnings about incorrect information in the visibility map, because we checked the visibility map bit before locking the heap page, creating a race condition. Fix by rechecking the visibility map bit before we complain. Rejigger some related logic so that we rely on the possibly-outdated all_visible_according_to_vm value as little as possible. In heap_multi_insert, it's not safe to clear the visibility map bit before beginning the critical section. The visibility map is not crash-safe unless we treat clearing the bit as a critical operation. Specifically, if the transaction were to error out after we set the bit and before entering the critical section, we could end up writing the heap page to disk (with the bit cleared) and crashing before the visibility map page made it to disk. That would be bad. heap_insert has this correct, but somehow the order of operations got rearranged when heap_multi_insert was added. Also, add some more comments to visibilitymap_test, lazy_scan_heap, and IndexOnlyNext, expounding on concurrency issues. Per extensive code review by Andres Freund, and further review by Tom Lane, who also made the original report about the bogus warnings.
2012-06-07 18:25:41 +02:00
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
/*
* XXX Should we set PageSetPrunable on this page ? See heap_insert()
*/
MarkBufferDirty(buffer);
/* XLOG stuff */
if (needwal)
{
XLogRecPtr recptr;
xl_heap_multi_insert *xlrec;
uint8 info = XLOG_HEAP2_MULTI_INSERT;
char *tupledata;
int totaldatalen;
Avoid using potentially-under-aligned page buffers. There's a project policy against using plain "char buf[BLCKSZ]" local or static variables as page buffers; preferred style is to palloc or malloc each buffer to ensure it is MAXALIGN'd. However, that policy's been ignored in an increasing number of places. We've apparently got away with it so far, probably because (a) relatively few people use platforms on which misalignment causes core dumps and/or (b) the variables chance to be sufficiently aligned anyway. But this is not something to rely on. Moreover, even if we don't get a core dump, we might be paying a lot of cycles for misaligned accesses. To fix, invent new union types PGAlignedBlock and PGAlignedXLogBlock that the compiler must allocate with sufficient alignment, and use those in place of plain char arrays. I used these types even for variables where there's no risk of a misaligned access, since ensuring proper alignment should make kernel data transfers faster. I also changed some places where we had been palloc'ing short-lived buffers, for coding style uniformity and to save palloc/pfree overhead. Since this seems to be a live portability hazard (despite the lack of field reports), back-patch to all supported versions. Patch by me; thanks to Michael Paquier for review. Discussion: https://postgr.es/m/1535618100.1286.3.camel@credativ.de
2018-09-01 21:27:12 +02:00
char *scratchptr = scratch.data;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
bool init;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
int bufflags = 0;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
/*
* If the page was previously empty, we can reinit the page
* instead of restoring the whole thing.
*/
Set PD_ALL_VISIBLE and visibility map bits in COPY FREEZE Make sure COPY FREEZE marks the pages as PD_ALL_VISIBLE and updates the visibility map. Until now we only marked individual tuples as frozen, but page-level flags were not updated, so the first VACUUM after the COPY FREEZE had to rewrite the whole table. This is a fairly old patch, and multiple people worked on it. The first version was written by Jeff Janes, and then reworked by Pavan Deolasee and Anastasia Lubennikova. Author: Anastasia Lubennikova, Pavan Deolasee, Jeff Janes Reviewed-by: Kuntal Ghosh, Jeff Janes, Tomas Vondra, Masahiko Sawada, Andres Freund, Ibrar Ahmed, Robert Haas, Tatsuro Ishii, Darafei Praliaskouski Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com Discussion: https://postgr.es/m/CAMkU%3D1w3osJJ2FneELhhNRLxfZitDgp9FPHee08NT2FQFmz_pQ%40mail.gmail.com
2021-01-17 22:11:39 +01:00
init = starting_with_empty_page;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
/* allocate xl_heap_multi_insert struct from the scratch area */
xlrec = (xl_heap_multi_insert *) scratchptr;
scratchptr += SizeOfHeapMultiInsert;
/*
* Allocate offsets array. Unless we're reinitializing the page,
* in that case the tuples are stored in order starting at
* FirstOffsetNumber and we don't need to store the offsets
* explicitly.
*/
if (!init)
scratchptr += nthispage * sizeof(OffsetNumber);
/* the rest of the scratch space is used for tuple data */
tupledata = scratchptr;
Set PD_ALL_VISIBLE and visibility map bits in COPY FREEZE Make sure COPY FREEZE marks the pages as PD_ALL_VISIBLE and updates the visibility map. Until now we only marked individual tuples as frozen, but page-level flags were not updated, so the first VACUUM after the COPY FREEZE had to rewrite the whole table. This is a fairly old patch, and multiple people worked on it. The first version was written by Jeff Janes, and then reworked by Pavan Deolasee and Anastasia Lubennikova. Author: Anastasia Lubennikova, Pavan Deolasee, Jeff Janes Reviewed-by: Kuntal Ghosh, Jeff Janes, Tomas Vondra, Masahiko Sawada, Andres Freund, Ibrar Ahmed, Robert Haas, Tatsuro Ishii, Darafei Praliaskouski Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com Discussion: https://postgr.es/m/CAMkU%3D1w3osJJ2FneELhhNRLxfZitDgp9FPHee08NT2FQFmz_pQ%40mail.gmail.com
2021-01-17 22:11:39 +01:00
/* check that the mutually exclusive flags are not both set */
Assert(!(all_visible_cleared && all_frozen_set));
xlrec->flags = 0;
if (all_visible_cleared)
xlrec->flags = XLH_INSERT_ALL_VISIBLE_CLEARED;
if (all_frozen_set)
xlrec->flags = XLH_INSERT_ALL_FROZEN_SET;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
xlrec->ntuples = nthispage;
/*
* Write out an xl_multi_insert_tuple and the tuple data itself
* for each tuple.
*/
for (i = 0; i < nthispage; i++)
{
HeapTuple heaptup = heaptuples[ndone + i];
xl_multi_insert_tuple *tuphdr;
int datalen;
if (!init)
xlrec->offsets[i] = ItemPointerGetOffsetNumber(&heaptup->t_self);
/* xl_multi_insert_tuple needs two-byte alignment. */
tuphdr = (xl_multi_insert_tuple *) SHORTALIGN(scratchptr);
scratchptr = ((char *) tuphdr) + SizeOfMultiInsertTuple;
tuphdr->t_infomask2 = heaptup->t_data->t_infomask2;
tuphdr->t_infomask = heaptup->t_data->t_infomask;
tuphdr->t_hoff = heaptup->t_data->t_hoff;
/* write bitmap [+ padding] [+ oid] + data */
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
datalen = heaptup->t_len - SizeofHeapTupleHeader;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
memcpy(scratchptr,
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
(char *) heaptup->t_data + SizeofHeapTupleHeader,
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
datalen);
tuphdr->datalen = datalen;
scratchptr += datalen;
}
totaldatalen = scratchptr - tupledata;
Avoid using potentially-under-aligned page buffers. There's a project policy against using plain "char buf[BLCKSZ]" local or static variables as page buffers; preferred style is to palloc or malloc each buffer to ensure it is MAXALIGN'd. However, that policy's been ignored in an increasing number of places. We've apparently got away with it so far, probably because (a) relatively few people use platforms on which misalignment causes core dumps and/or (b) the variables chance to be sufficiently aligned anyway. But this is not something to rely on. Moreover, even if we don't get a core dump, we might be paying a lot of cycles for misaligned accesses. To fix, invent new union types PGAlignedBlock and PGAlignedXLogBlock that the compiler must allocate with sufficient alignment, and use those in place of plain char arrays. I used these types even for variables where there's no risk of a misaligned access, since ensuring proper alignment should make kernel data transfers faster. I also changed some places where we had been palloc'ing short-lived buffers, for coding style uniformity and to save palloc/pfree overhead. Since this seems to be a live portability hazard (despite the lack of field reports), back-patch to all supported versions. Patch by me; thanks to Michael Paquier for review. Discussion: https://postgr.es/m/1535618100.1286.3.camel@credativ.de
2018-09-01 21:27:12 +02:00
Assert((scratchptr - scratch.data) < BLCKSZ);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
if (need_tuple_data)
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
xlrec->flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
/*
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
* Signal that this is the last xl_heap_multi_insert record
* emitted by this call to heap_multi_insert(). Needed for logical
* decoding so it knows when to cleanup temporary data.
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
*/
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (ndone + nthispage == ntuples)
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
xlrec->flags |= XLH_INSERT_LAST_IN_MULTI;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
if (init)
{
info |= XLOG_HEAP_INIT_PAGE;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
bufflags |= REGBUF_WILL_INIT;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
}
Fix decoding of consecutive MULTI_INSERTs emitted by one heap_multi_insert(). Commit 1b86c81d2d fixed the decoding of toasted columns for the rows contained in one xl_heap_multi_insert record. But that's not actually enough, because heap_multi_insert() will actually first toast all passed in rows and then emit several *_multi_insert records; one for each page it fills with tuples. Add a XLOG_HEAP_LAST_MULTI_INSERT flag which is set in xl_heap_multi_insert->flag denoting that this multi_insert record is the last emitted by one heap_multi_insert() call. Then use that flag in decode.c to only set clear_toast_afterwards in the right situation. Expand the number of rows inserted via COPY in the corresponding regression test to make sure that more than one heap page is filled with tuples by one heap_multi_insert() call. Backpatch to 9.4 like the previous commit.
2014-07-12 14:28:19 +02:00
/*
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
* If we're doing logical decoding, include the new tuple data
* even if we take a full-page image of the page.
Fix decoding of consecutive MULTI_INSERTs emitted by one heap_multi_insert(). Commit 1b86c81d2d fixed the decoding of toasted columns for the rows contained in one xl_heap_multi_insert record. But that's not actually enough, because heap_multi_insert() will actually first toast all passed in rows and then emit several *_multi_insert records; one for each page it fills with tuples. Add a XLOG_HEAP_LAST_MULTI_INSERT flag which is set in xl_heap_multi_insert->flag denoting that this multi_insert record is the last emitted by one heap_multi_insert() call. Then use that flag in decode.c to only set clear_toast_afterwards in the right situation. Expand the number of rows inserted via COPY in the corresponding regression test to make sure that more than one heap page is filled with tuples by one heap_multi_insert() call. Backpatch to 9.4 like the previous commit.
2014-07-12 14:28:19 +02:00
*/
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (need_tuple_data)
bufflags |= REGBUF_KEEP_DATA;
XLogBeginInsert();
Avoid using potentially-under-aligned page buffers. There's a project policy against using plain "char buf[BLCKSZ]" local or static variables as page buffers; preferred style is to palloc or malloc each buffer to ensure it is MAXALIGN'd. However, that policy's been ignored in an increasing number of places. We've apparently got away with it so far, probably because (a) relatively few people use platforms on which misalignment causes core dumps and/or (b) the variables chance to be sufficiently aligned anyway. But this is not something to rely on. Moreover, even if we don't get a core dump, we might be paying a lot of cycles for misaligned accesses. To fix, invent new union types PGAlignedBlock and PGAlignedXLogBlock that the compiler must allocate with sufficient alignment, and use those in place of plain char arrays. I used these types even for variables where there's no risk of a misaligned access, since ensuring proper alignment should make kernel data transfers faster. I also changed some places where we had been palloc'ing short-lived buffers, for coding style uniformity and to save palloc/pfree overhead. Since this seems to be a live portability hazard (despite the lack of field reports), back-patch to all supported versions. Patch by me; thanks to Michael Paquier for review. Discussion: https://postgr.es/m/1535618100.1286.3.camel@credativ.de
2018-09-01 21:27:12 +02:00
XLogRegisterData((char *) xlrec, tupledata - scratch.data);
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
Fix decoding of consecutive MULTI_INSERTs emitted by one heap_multi_insert(). Commit 1b86c81d2d fixed the decoding of toasted columns for the rows contained in one xl_heap_multi_insert record. But that's not actually enough, because heap_multi_insert() will actually first toast all passed in rows and then emit several *_multi_insert records; one for each page it fills with tuples. Add a XLOG_HEAP_LAST_MULTI_INSERT flag which is set in xl_heap_multi_insert->flag denoting that this multi_insert record is the last emitted by one heap_multi_insert() call. Then use that flag in decode.c to only set clear_toast_afterwards in the right situation. Expand the number of rows inserted via COPY in the corresponding regression test to make sure that more than one heap page is filled with tuples by one heap_multi_insert() call. Backpatch to 9.4 like the previous commit.
2014-07-12 14:28:19 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterBufData(0, tupledata, totaldatalen);
Introduce replication progress tracking infrastructure. When implementing a replication solution ontop of logical decoding, two related problems exist: * How to safely keep track of replication progress * How to change replication behavior, based on the origin of a row; e.g. to avoid loops in bi-directional replication setups The solution to these problems, as implemented here, consist out of three parts: 1) 'replication origins', which identify nodes in a replication setup. 2) 'replication progress tracking', which remembers, for each replication origin, how far replay has progressed in a efficient and crash safe manner. 3) The ability to filter out changes performed on the behest of a replication origin during logical decoding; this allows complex replication topologies. E.g. by filtering all replayed changes out. Most of this could also be implemented in "userspace", e.g. by inserting additional rows contain origin information, but that ends up being much less efficient and more complicated. We don't want to require various replication solutions to reimplement logic for this independently. The infrastructure is intended to be generic enough to be reusable. This infrastructure also replaces the 'nodeid' infrastructure of commit timestamps. It is intended to provide all the former capabilities, except that there's only 2^16 different origins; but now they integrate with logical decoding. Additionally more functionality is accessible via SQL. Since the commit timestamp infrastructure has also been introduced in 9.5 (commit 73c986add) changing the API is not a problem. For now the number of origins for which the replication progress can be tracked simultaneously is determined by the max_replication_slots GUC. That GUC is not a perfect match to configure this, but there doesn't seem to be sufficient reason to introduce a separate new one. Bumps both catversion and wal page magic. Author: Andres Freund, with contributions from Petr Jelinek and Craig Ringer Reviewed-By: Heikki Linnakangas, Petr Jelinek, Robert Haas, Steve Singer Discussion: 20150216002155.GI15326@awork2.anarazel.de, 20140923182422.GA15776@alap3.anarazel.de, 20131114172632.GE7522@alap2.anarazel.de
2015-04-29 19:30:53 +02:00
/* filtering by origin on a row level is much more efficient */
Skip checkpoints, archiving on idle systems. Some background activity (like checkpoints, archive timeout, standby snapshots) is not supposed to happen on an idle system. Unfortunately so far it was not easy to determine when a system is idle, which defeated some of the attempts to avoid redundant activity on an idle system. To make that easier, allow to make individual WAL insertions as not being "important". By checking whether any important activity happened since the last time an activity was performed, it now is easy to check whether some action needs to be repeated. Use the new facility for checkpoints, archive timeout and standby snapshots. The lack of a facility causes some issues in older releases, but in my opinion the consequences (superflous checkpoints / archived segments) aren't grave enough to warrant backpatching. Author: Michael Paquier, editorialized by Andres Freund Reviewed-By: Andres Freund, David Steele, Amit Kapila, Kyotaro HORIGUCHI Bug: #13685 Discussion: https://www.postgresql.org/message-id/20151016203031.3019.72930@wrigleys.postgresql.org https://www.postgresql.org/message-id/CAB7nPqQcPqxEM3S735Bd2RzApNqSNJVietAC=6kfkYv_45dKwA@mail.gmail.com Backpatch: -
2016-12-22 20:31:50 +01:00
XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
Introduce replication progress tracking infrastructure. When implementing a replication solution ontop of logical decoding, two related problems exist: * How to safely keep track of replication progress * How to change replication behavior, based on the origin of a row; e.g. to avoid loops in bi-directional replication setups The solution to these problems, as implemented here, consist out of three parts: 1) 'replication origins', which identify nodes in a replication setup. 2) 'replication progress tracking', which remembers, for each replication origin, how far replay has progressed in a efficient and crash safe manner. 3) The ability to filter out changes performed on the behest of a replication origin during logical decoding; this allows complex replication topologies. E.g. by filtering all replayed changes out. Most of this could also be implemented in "userspace", e.g. by inserting additional rows contain origin information, but that ends up being much less efficient and more complicated. We don't want to require various replication solutions to reimplement logic for this independently. The infrastructure is intended to be generic enough to be reusable. This infrastructure also replaces the 'nodeid' infrastructure of commit timestamps. It is intended to provide all the former capabilities, except that there's only 2^16 different origins; but now they integrate with logical decoding. Additionally more functionality is accessible via SQL. Since the commit timestamp infrastructure has also been introduced in 9.5 (commit 73c986add) changing the API is not a problem. For now the number of origins for which the replication progress can be tracked simultaneously is determined by the max_replication_slots GUC. That GUC is not a perfect match to configure this, but there doesn't seem to be sufficient reason to introduce a separate new one. Bumps both catversion and wal page magic. Author: Andres Freund, with contributions from Petr Jelinek and Craig Ringer Reviewed-By: Heikki Linnakangas, Petr Jelinek, Robert Haas, Steve Singer Discussion: 20150216002155.GI15326@awork2.anarazel.de, 20140923182422.GA15776@alap3.anarazel.de, 20131114172632.GE7522@alap2.anarazel.de
2015-04-29 19:30:53 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
recptr = XLogInsert(RM_HEAP2_ID, info);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
PageSetLSN(page, recptr);
}
END_CRIT_SECTION();
Set PD_ALL_VISIBLE and visibility map bits in COPY FREEZE Make sure COPY FREEZE marks the pages as PD_ALL_VISIBLE and updates the visibility map. Until now we only marked individual tuples as frozen, but page-level flags were not updated, so the first VACUUM after the COPY FREEZE had to rewrite the whole table. This is a fairly old patch, and multiple people worked on it. The first version was written by Jeff Janes, and then reworked by Pavan Deolasee and Anastasia Lubennikova. Author: Anastasia Lubennikova, Pavan Deolasee, Jeff Janes Reviewed-by: Kuntal Ghosh, Jeff Janes, Tomas Vondra, Masahiko Sawada, Andres Freund, Ibrar Ahmed, Robert Haas, Tatsuro Ishii, Darafei Praliaskouski Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com Discussion: https://postgr.es/m/CAMkU%3D1w3osJJ2FneELhhNRLxfZitDgp9FPHee08NT2FQFmz_pQ%40mail.gmail.com
2021-01-17 22:11:39 +01:00
/*
* If we've frozen everything on the page, update the visibilitymap.
* We're already holding pin on the vmbuffer.
*/
if (all_frozen_set)
{
Assert(PageIsAllVisible(page));
Assert(visibilitymap_pin_ok(BufferGetBlockNumber(buffer), vmbuffer));
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
Set PD_ALL_VISIBLE and visibility map bits in COPY FREEZE Make sure COPY FREEZE marks the pages as PD_ALL_VISIBLE and updates the visibility map. Until now we only marked individual tuples as frozen, but page-level flags were not updated, so the first VACUUM after the COPY FREEZE had to rewrite the whole table. This is a fairly old patch, and multiple people worked on it. The first version was written by Jeff Janes, and then reworked by Pavan Deolasee and Anastasia Lubennikova. Author: Anastasia Lubennikova, Pavan Deolasee, Jeff Janes Reviewed-by: Kuntal Ghosh, Jeff Janes, Tomas Vondra, Masahiko Sawada, Andres Freund, Ibrar Ahmed, Robert Haas, Tatsuro Ishii, Darafei Praliaskouski Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com Discussion: https://postgr.es/m/CAMkU%3D1w3osJJ2FneELhhNRLxfZitDgp9FPHee08NT2FQFmz_pQ%40mail.gmail.com
2021-01-17 22:11:39 +01:00
/*
* It's fine to use InvalidTransactionId here - this is only used
* when HEAP_INSERT_FROZEN is specified, which intentionally
* violates visibility rules.
*/
visibilitymap_set(relation, BufferGetBlockNumber(buffer), buffer,
InvalidXLogRecPtr, vmbuffer,
InvalidTransactionId,
VISIBILITYMAP_ALL_VISIBLE | VISIBILITYMAP_ALL_FROZEN);
}
UnlockReleaseBuffer(buffer);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
ndone += nthispage;
Set PD_ALL_VISIBLE and visibility map bits in COPY FREEZE Make sure COPY FREEZE marks the pages as PD_ALL_VISIBLE and updates the visibility map. Until now we only marked individual tuples as frozen, but page-level flags were not updated, so the first VACUUM after the COPY FREEZE had to rewrite the whole table. This is a fairly old patch, and multiple people worked on it. The first version was written by Jeff Janes, and then reworked by Pavan Deolasee and Anastasia Lubennikova. Author: Anastasia Lubennikova, Pavan Deolasee, Jeff Janes Reviewed-by: Kuntal Ghosh, Jeff Janes, Tomas Vondra, Masahiko Sawada, Andres Freund, Ibrar Ahmed, Robert Haas, Tatsuro Ishii, Darafei Praliaskouski Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com Discussion: https://postgr.es/m/CAMkU%3D1w3osJJ2FneELhhNRLxfZitDgp9FPHee08NT2FQFmz_pQ%40mail.gmail.com
2021-01-17 22:11:39 +01:00
/*
* NB: Only release vmbuffer after inserting all tuples - it's fairly
* likely that we'll insert into subsequent heap pages that are likely
* to use the same vm page.
*/
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
}
Set PD_ALL_VISIBLE and visibility map bits in COPY FREEZE Make sure COPY FREEZE marks the pages as PD_ALL_VISIBLE and updates the visibility map. Until now we only marked individual tuples as frozen, but page-level flags were not updated, so the first VACUUM after the COPY FREEZE had to rewrite the whole table. This is a fairly old patch, and multiple people worked on it. The first version was written by Jeff Janes, and then reworked by Pavan Deolasee and Anastasia Lubennikova. Author: Anastasia Lubennikova, Pavan Deolasee, Jeff Janes Reviewed-by: Kuntal Ghosh, Jeff Janes, Tomas Vondra, Masahiko Sawada, Andres Freund, Ibrar Ahmed, Robert Haas, Tatsuro Ishii, Darafei Praliaskouski Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com Discussion: https://postgr.es/m/CAMkU%3D1w3osJJ2FneELhhNRLxfZitDgp9FPHee08NT2FQFmz_pQ%40mail.gmail.com
2021-01-17 22:11:39 +01:00
/* We're done with inserting all tuples, so release the last vmbuffer. */
if (vmbuffer != InvalidBuffer)
ReleaseBuffer(vmbuffer);
Fix serialization anomalies due to race conditions on INSERT. On insert the CheckForSerializableConflictIn() test was performed before the page(s) which were going to be modified had been locked (with an exclusive buffer content lock). If another process acquired a relation SIReadLock on the heap and scanned to a page on which an insert was going to occur before the page was so locked, a rw-conflict would be missed, which could allow a serialization anomaly to be missed. The window between the check and the page lock was small, so the bug was generally not noticed unless there was high concurrency with multiple processes inserting into the same table. This was reported by Peter Bailis as bug #11732, by Sean Chittenden as bug #13667, and by others. The race condition was eliminated in heap_insert() by moving the check down below the acquisition of the buffer lock, which had been the very next statement. Because of the loop locking and unlocking multiple buffers in heap_multi_insert() a check was added after all inserts were completed. The check before the start of the inserts was left because it might avoid a large amount of work to detect a serialization anomaly before performing the all of the inserts and the related WAL logging. While investigating this bug, other SSI bugs which were even harder to hit in practice were noticed and fixed, an unnecessary check (covered by another check, so redundant) was removed from heap_update(), and comments were improved. Back-patch to all supported branches. Kevin Grittner and Thomas Munro
2015-10-31 20:43:34 +01:00
/*
* We're done with the actual inserts. Check for conflicts again, to
* ensure that all rw-conflicts in to these inserts are detected. Without
* this final check, a sequential scan of the heap may have locked the
* table after the "before" check, missing one opportunity to detect the
* conflict, and then scanned the table before the new tuples were there,
* missing the other chance to detect the conflict.
*
* For heap inserts, we only need to check for table-level SSI locks. Our
* new tuples can't possibly conflict with existing tuple locks, and heap
* page locks are only consolidated versions of tuple locks; they do not
* lock "gaps" as index page locks do. So we don't need to specify a
* buffer when making the call.
*/
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
Fix serialization anomalies due to race conditions on INSERT. On insert the CheckForSerializableConflictIn() test was performed before the page(s) which were going to be modified had been locked (with an exclusive buffer content lock). If another process acquired a relation SIReadLock on the heap and scanned to a page on which an insert was going to occur before the page was so locked, a rw-conflict would be missed, which could allow a serialization anomaly to be missed. The window between the check and the page lock was small, so the bug was generally not noticed unless there was high concurrency with multiple processes inserting into the same table. This was reported by Peter Bailis as bug #11732, by Sean Chittenden as bug #13667, and by others. The race condition was eliminated in heap_insert() by moving the check down below the acquisition of the buffer lock, which had been the very next statement. Because of the loop locking and unlocking multiple buffers in heap_multi_insert() a check was added after all inserts were completed. The check before the start of the inserts was left because it might avoid a large amount of work to detect a serialization anomaly before performing the all of the inserts and the related WAL logging. While investigating this bug, other SSI bugs which were even harder to hit in practice were noticed and fixed, an unnecessary check (covered by another check, so redundant) was removed from heap_update(), and comments were improved. Back-patch to all supported branches. Kevin Grittner and Thomas Munro
2015-10-31 20:43:34 +01:00
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
/*
* If tuples are cachable, mark them for invalidation from the caches in
* case we abort. Note it is OK to do this after releasing the buffer,
* because the heaptuples data structure is all in local memory, not in
* the shared buffer.
*/
Refine our definition of what constitutes a system relation. Although user-defined relations can't be directly created in pg_catalog, it's possible for them to end up there, because you can create them in some other schema and then use ALTER TABLE .. SET SCHEMA to move them there. Previously, such relations couldn't afterwards be manipulated, because IsSystemRelation()/IsSystemClass() rejected all attempts to modify objects in the pg_catalog schema, regardless of their origin. With this patch, they now reject only those objects in pg_catalog which were created at initdb-time, allowing most operations on user-created tables in pg_catalog to proceed normally. This patch also adds new functions IsCatalogRelation() and IsCatalogClass(), which is similar to IsSystemRelation() and IsSystemClass() but with a slightly narrower definition: only TOAST tables of system catalogs are included, rather than *all* TOAST tables. This is currently used only for making decisions about when invalidation messages need to be sent, but upcoming logical decoding patches will find other uses for this information. Andres Freund, with some modifications by me.
2013-11-29 02:57:20 +01:00
if (IsCatalogRelation(relation))
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
{
for (i = 0; i < ntuples; i++)
CacheInvalidateHeapTuple(relation, heaptuples[i], NULL);
}
tableam: Add table_multi_insert() and revamp/speed-up COPY FROM buffering. This adds table_multi_insert(), and converts COPY FROM, the only user of heap_multi_insert, to it. A simple conversion of COPY FROM use slots would have yielded a slowdown when inserting into a partitioned table for some workloads. Different partitions might need different slots (both slot types and their descriptors), and dropping / creating slots when there's constant partition changes is measurable. Thus instead revamp the COPY FROM buffering for partitioned tables to allow to buffer inserts into multiple tables, flushing only when limits are reached across all partition buffers. By only dropping slots when there've been inserts into too many different partitions, the aforementioned overhead is gone. By allowing larger batches, even when there are frequent partition changes, we actuall speed such cases up significantly. By using slots COPY of very narrow rows into unlogged / temporary might slow down very slightly (due to the indirect function calls). Author: David Rowley, Andres Freund, Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190327054923.t3epfuewxfqdt22e@alap3.anarazel.de
2019-04-05 00:47:19 +02:00
/* copy t_self fields back to the caller's slots */
Fix heap_multi_insert to set t_self field in the caller's tuples. If tuples were toasted, heap_multi_insert didn't update the ctid on the original tuples. This caused a failure if there was an after trigger (including a foreign key), on the table, and a tuple got toasted. Per off-list report and test case from Ted Phelps
2012-02-13 09:14:49 +01:00
for (i = 0; i < ntuples; i++)
tableam: Add table_multi_insert() and revamp/speed-up COPY FROM buffering. This adds table_multi_insert(), and converts COPY FROM, the only user of heap_multi_insert, to it. A simple conversion of COPY FROM use slots would have yielded a slowdown when inserting into a partitioned table for some workloads. Different partitions might need different slots (both slot types and their descriptors), and dropping / creating slots when there's constant partition changes is measurable. Thus instead revamp the COPY FROM buffering for partitioned tables to allow to buffer inserts into multiple tables, flushing only when limits are reached across all partition buffers. By only dropping slots when there've been inserts into too many different partitions, the aforementioned overhead is gone. By allowing larger batches, even when there are frequent partition changes, we actuall speed such cases up significantly. By using slots COPY of very narrow rows into unlogged / temporary might slow down very slightly (due to the indirect function calls). Author: David Rowley, Andres Freund, Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190327054923.t3epfuewxfqdt22e@alap3.anarazel.de
2019-04-05 00:47:19 +02:00
slots[i]->tts_tid = heaptuples[i]->t_self;
Fix heap_multi_insert to set t_self field in the caller's tuples. If tuples were toasted, heap_multi_insert didn't update the ctid on the original tuples. This caused a failure if there was an after trigger (including a foreign key), on the table, and a tuple got toasted. Per off-list report and test case from Ted Phelps
2012-02-13 09:14:49 +01:00
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
pgstat_count_heap_insert(relation, ntuples);
}
Remove global variable scanCommandId in favor of storing a command ID in snapshots, per my proposal of a few days ago. Also, tweak heapam.c routines (heap_insert, heap_update, heap_delete, heap_mark4update) to be passed the command ID to use, instead of doing GetCurrentCommandID. For catalog updates they'll still get passed current command ID, but for updates generated from the main executor they'll get passed the command ID saved in the snapshot the query is using. This should fix some corner cases associated with functions and triggers that advance current command ID while an outer query is still in progress.
2002-05-22 00:05:55 +02:00
/*
* simple_heap_insert - insert a tuple
*
* Currently, this routine differs from heap_insert only in supplying
Teach CLUSTER to skip writing WAL if not needed (ie, not using archiving) --- Simon. Also, code review and cleanup for the previous COPY-no-WAL patches --- Tom.
2007-03-29 02:15:39 +02:00
* a default command ID and not allowing access to the speedup options.
*
* This should be used rather than using heap_insert directly in most places
* where we are modifying system catalogs.
Remove global variable scanCommandId in favor of storing a command ID in snapshots, per my proposal of a few days ago. Also, tweak heapam.c routines (heap_insert, heap_update, heap_delete, heap_mark4update) to be passed the command ID to use, instead of doing GetCurrentCommandID. For catalog updates they'll still get passed current command ID, but for updates generated from the main executor they'll get passed the command ID saved in the snapshot the query is using. This should fix some corner cases associated with functions and triggers that advance current command ID while an outer query is still in progress.
2002-05-22 00:05:55 +02:00
*/
Remove WITH OIDS support, change oid catalog column visibility. Previously tables declared WITH OIDS, including a significant fraction of the catalog tables, stored the oid column not as a normal column, but as part of the tuple header. This special column was not shown by default, which was somewhat odd, as it's often (consider e.g. pg_class.oid) one of the more important parts of a row. Neither pg_dump nor COPY included the contents of the oid column by default. The fact that the oid column was not an ordinary column necessitated a significant amount of special case code to support oid columns. That already was painful for the existing, but upcoming work aiming to make table storage pluggable, would have required expanding and duplicating that "specialness" significantly. WITH OIDS has been deprecated since 2005 (commit ff02d0a05280e0). Remove it. Removing includes: - CREATE TABLE and ALTER TABLE syntax for declaring the table to be WITH OIDS has been removed (WITH (oids[ = true]) will error out) - pg_dump does not support dumping tables declared WITH OIDS and will issue a warning when dumping one (and ignore the oid column). - restoring an pg_dump archive with pg_restore will warn when restoring a table with oid contents (and ignore the oid column) - COPY will refuse to load binary dump that includes oids. - pg_upgrade will error out when encountering tables declared WITH OIDS, they have to be altered to remove the oid column first. - Functionality to access the oid of the last inserted row (like plpgsql's RESULT_OID, spi's SPI_lastoid, ...) has been removed. The syntax for declaring a table WITHOUT OIDS (or WITH (oids = false) for CREATE TABLE) is still supported. While that requires a bit of support code, it seems unnecessary to break applications / dumps that do not use oids, and are explicit about not using them. The biggest user of WITH OID columns was postgres' catalog. This commit changes all 'magic' oid columns to be columns that are normally declared and stored. To reduce unnecessary query breakage all the newly added columns are still named 'oid', even if a table's column naming scheme would indicate 'reloid' or such. This obviously requires adapting a lot code, mostly replacing oid access via HeapTupleGetOid() with access to the underlying Form_pg_*->oid column. The bootstrap process now assigns oids for all oid columns in genbki.pl that do not have an explicit value (starting at the largest oid previously used), only oids assigned later by oids will be above FirstBootstrapObjectId. As the oid column now is a normal column the special bootstrap syntax for oids has been removed. Oids are not automatically assigned during insertion anymore, all backend code explicitly assigns oids with GetNewOidWithIndex(). For the rare case that insertions into the catalog via SQL are called for the new pg_nextoid() function can be used (which only works on catalog tables). The fact that oid columns on system tables are now normal columns means that they will be included in the set of columns expanded by * (i.e. SELECT * FROM pg_class will now include the table's oid, previously it did not). It'd not technically be hard to hide oid column by default, but that'd mean confusing behavior would either have to be carried forward forever, or it'd cause breakage down the line. While it's not unlikely that further adjustments are needed, the scope/invasiveness of the patch makes it worthwhile to get merge this now. It's painful to maintain externally, too complicated to commit after the code code freeze, and a dependency of a number of other patches. Catversion bump, for obvious reasons. Author: Andres Freund, with contributions by John Naylor Discussion: https://postgr.es/m/20180930034810.ywp2c7awz7opzcfr@alap3.anarazel.de
2018-11-21 00:36:57 +01:00
void
Remove global variable scanCommandId in favor of storing a command ID in snapshots, per my proposal of a few days ago. Also, tweak heapam.c routines (heap_insert, heap_update, heap_delete, heap_mark4update) to be passed the command ID to use, instead of doing GetCurrentCommandID. For catalog updates they'll still get passed current command ID, but for updates generated from the main executor they'll get passed the command ID saved in the snapshot the query is using. This should fix some corner cases associated with functions and triggers that advance current command ID while an outer query is still in progress.
2002-05-22 00:05:55 +02:00
simple_heap_insert(Relation relation, HeapTuple tup)
{
Remove WITH OIDS support, change oid catalog column visibility. Previously tables declared WITH OIDS, including a significant fraction of the catalog tables, stored the oid column not as a normal column, but as part of the tuple header. This special column was not shown by default, which was somewhat odd, as it's often (consider e.g. pg_class.oid) one of the more important parts of a row. Neither pg_dump nor COPY included the contents of the oid column by default. The fact that the oid column was not an ordinary column necessitated a significant amount of special case code to support oid columns. That already was painful for the existing, but upcoming work aiming to make table storage pluggable, would have required expanding and duplicating that "specialness" significantly. WITH OIDS has been deprecated since 2005 (commit ff02d0a05280e0). Remove it. Removing includes: - CREATE TABLE and ALTER TABLE syntax for declaring the table to be WITH OIDS has been removed (WITH (oids[ = true]) will error out) - pg_dump does not support dumping tables declared WITH OIDS and will issue a warning when dumping one (and ignore the oid column). - restoring an pg_dump archive with pg_restore will warn when restoring a table with oid contents (and ignore the oid column) - COPY will refuse to load binary dump that includes oids. - pg_upgrade will error out when encountering tables declared WITH OIDS, they have to be altered to remove the oid column first. - Functionality to access the oid of the last inserted row (like plpgsql's RESULT_OID, spi's SPI_lastoid, ...) has been removed. The syntax for declaring a table WITHOUT OIDS (or WITH (oids = false) for CREATE TABLE) is still supported. While that requires a bit of support code, it seems unnecessary to break applications / dumps that do not use oids, and are explicit about not using them. The biggest user of WITH OID columns was postgres' catalog. This commit changes all 'magic' oid columns to be columns that are normally declared and stored. To reduce unnecessary query breakage all the newly added columns are still named 'oid', even if a table's column naming scheme would indicate 'reloid' or such. This obviously requires adapting a lot code, mostly replacing oid access via HeapTupleGetOid() with access to the underlying Form_pg_*->oid column. The bootstrap process now assigns oids for all oid columns in genbki.pl that do not have an explicit value (starting at the largest oid previously used), only oids assigned later by oids will be above FirstBootstrapObjectId. As the oid column now is a normal column the special bootstrap syntax for oids has been removed. Oids are not automatically assigned during insertion anymore, all backend code explicitly assigns oids with GetNewOidWithIndex(). For the rare case that insertions into the catalog via SQL are called for the new pg_nextoid() function can be used (which only works on catalog tables). The fact that oid columns on system tables are now normal columns means that they will be included in the set of columns expanded by * (i.e. SELECT * FROM pg_class will now include the table's oid, previously it did not). It'd not technically be hard to hide oid column by default, but that'd mean confusing behavior would either have to be carried forward forever, or it'd cause breakage down the line. While it's not unlikely that further adjustments are needed, the scope/invasiveness of the patch makes it worthwhile to get merge this now. It's painful to maintain externally, too complicated to commit after the code code freeze, and a dependency of a number of other patches. Catversion bump, for obvious reasons. Author: Andres Freund, with contributions by John Naylor Discussion: https://postgr.es/m/20180930034810.ywp2c7awz7opzcfr@alap3.anarazel.de
2018-11-21 00:36:57 +01:00
heap_insert(relation, tup, GetCurrentCommandId(true), 0, NULL);
Remove global variable scanCommandId in favor of storing a command ID in snapshots, per my proposal of a few days ago. Also, tweak heapam.c routines (heap_insert, heap_update, heap_delete, heap_mark4update) to be passed the command ID to use, instead of doing GetCurrentCommandID. For catalog updates they'll still get passed current command ID, but for updates generated from the main executor they'll get passed the command ID saved in the snapshot the query is using. This should fix some corner cases associated with functions and triggers that advance current command ID while an outer query is still in progress.
2002-05-22 00:05:55 +02:00
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* Given infomask/infomask2, compute the bits that must be saved in the
* "infobits" field of xl_heap_delete, xl_heap_update, xl_heap_lock,
* xl_heap_lock_updated WAL records.
*
* See fix_infomask_from_infobits.
*/
static uint8
compute_infobits(uint16 infomask, uint16 infomask2)
{
return
((infomask & HEAP_XMAX_IS_MULTI) != 0 ? XLHL_XMAX_IS_MULTI : 0) |
((infomask & HEAP_XMAX_LOCK_ONLY) != 0 ? XLHL_XMAX_LOCK_ONLY : 0) |
((infomask & HEAP_XMAX_EXCL_LOCK) != 0 ? XLHL_XMAX_EXCL_LOCK : 0) |
/* note we ignore HEAP_XMAX_SHR_LOCK here */
((infomask & HEAP_XMAX_KEYSHR_LOCK) != 0 ? XLHL_XMAX_KEYSHR_LOCK : 0) |
((infomask2 & HEAP_KEYS_UPDATED) != 0 ?
XLHL_KEYS_UPDATED : 0);
}
Fix race when updating a tuple concurrently locked by another process If a tuple is locked, and this lock is later upgraded either to an update or to a stronger lock, and in the meantime some other process tries to lock, update or delete the same tuple, it (the tuple) could end up being updated twice, or having conflicting locks held. The reason for this is that the second updater checks for a change in Xmax value, or in the HEAP_XMAX_IS_MULTI infomask bit, after noticing the first lock; and if there's a change, it restarts and re-evaluates its ability to update the tuple. But it neglected to check for changes in lock strength or in lock-vs-update status when those two properties stayed the same. This would lead it to take the wrong decision and continue with its own update, when in reality it shouldn't do so but instead restart from the top. This could lead to either an assertion failure much later (when a multixact containing multiple updates is detected), or duplicate copies of tuples. To fix, make sure to compare the other relevant infomask bits alongside the Xmax value and HEAP_XMAX_IS_MULTI bit, and restart from the top if necessary. Also, in the belt-and-suspenders spirit, add a check to MultiXactCreateFromMembers that a multixact being created does not have two or more members that are claimed to be updates. This should protect against other bugs that might cause similar bogus situations. Backpatch to 9.3, where the possibility of multixacts containing updates was introduced. (In prior versions it was possible to have the tuple lock upgraded from shared to exclusive, and an update would not restart from the top; yet we're protected against a bug there because there's always a sleep to wait for the locking transaction to complete before continuing to do anything. Really, the fact that tuple locks always conflicted with concurrent updates is what protected against bugs here.) Per report from Andrew Dunstan and Josh Berkus in thread at http://www.postgresql.org/message-id/534C8B33.9050807@pgexperts.com Bug analysis by Andres Freund.
2014-04-24 20:41:55 +02:00
/*
* Given two versions of the same t_infomask for a tuple, compare them and
* return whether the relevant status for a tuple Xmax has changed. This is
* used after a buffer lock has been released and reacquired: we want to ensure
* that the tuple state continues to be the same it was when we previously
* examined it.
*
* Note the Xmax field itself must be compared separately.
*/
static inline bool
xmax_infomask_changed(uint16 new_infomask, uint16 old_infomask)
{
const uint16 interesting =
Pre-beta mechanical code beautification. Run pgindent, pgperltidy, and reformat-dat-files. This set of diffs is a bit larger than typical. We've updated to pg_bsd_indent 2.1.2, which properly indents variable declarations that have multi-line initialization expressions (the continuation lines are now indented one tab stop). We've also updated to perltidy version 20230309 and changed some of its settings, which reduces its desire to add whitespace to lines to make assignments etc. line up. Going forward, that should make for fewer random-seeming changes to existing code. Discussion: https://postgr.es/m/20230428092545.qfb3y5wcu4cm75ur@alvherre.pgsql
2023-05-19 23:24:48 +02:00
HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY | HEAP_LOCK_MASK;
Fix race when updating a tuple concurrently locked by another process If a tuple is locked, and this lock is later upgraded either to an update or to a stronger lock, and in the meantime some other process tries to lock, update or delete the same tuple, it (the tuple) could end up being updated twice, or having conflicting locks held. The reason for this is that the second updater checks for a change in Xmax value, or in the HEAP_XMAX_IS_MULTI infomask bit, after noticing the first lock; and if there's a change, it restarts and re-evaluates its ability to update the tuple. But it neglected to check for changes in lock strength or in lock-vs-update status when those two properties stayed the same. This would lead it to take the wrong decision and continue with its own update, when in reality it shouldn't do so but instead restart from the top. This could lead to either an assertion failure much later (when a multixact containing multiple updates is detected), or duplicate copies of tuples. To fix, make sure to compare the other relevant infomask bits alongside the Xmax value and HEAP_XMAX_IS_MULTI bit, and restart from the top if necessary. Also, in the belt-and-suspenders spirit, add a check to MultiXactCreateFromMembers that a multixact being created does not have two or more members that are claimed to be updates. This should protect against other bugs that might cause similar bogus situations. Backpatch to 9.3, where the possibility of multixacts containing updates was introduced. (In prior versions it was possible to have the tuple lock upgraded from shared to exclusive, and an update would not restart from the top; yet we're protected against a bug there because there's always a sleep to wait for the locking transaction to complete before continuing to do anything. Really, the fact that tuple locks always conflicted with concurrent updates is what protected against bugs here.) Per report from Andrew Dunstan and Josh Berkus in thread at http://www.postgresql.org/message-id/534C8B33.9050807@pgexperts.com Bug analysis by Andres Freund.
2014-04-24 20:41:55 +02:00
if ((new_infomask & interesting) != (old_infomask & interesting))
return true;
return false;
}
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
/*
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
* heap_delete - delete a tuple
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
*
tableam: Rename wrapper functions to match callback names. Some of the wrapper functions didn't match the callback names. Many of them due to staying "consistent" with historic naming of the wrapped functionality. We decided that for most cases it's more important to be for tableam to be consistent going forward, than with the past. The one exception is beginscan/endscan/... because it'd have looked odd to have systable_beginscan/endscan/... with a different naming scheme, and changing the systable_* APIs would have caused way too much churn (including breaking a lot of external users). Author: Ashwin Agrawal, with some small additions by Andres Freund Reviewed-By: Andres Freund Discussion: https://postgr.es/m/CALfoeiugyrXZfX7n0ORCa4L-m834dzmaE8eFdbNR6PMpetU4Ww@mail.gmail.com
2019-05-24 01:25:48 +02:00
* See table_tuple_delete() for an explanation of the parameters, except that
* this routine directly takes a tuple rather than a slot.
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
*
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
* In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
* t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
* only for TM_SelfModified, since we cannot obtain cmax from a combo CID
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
* generated by another transaction).
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
TM_Result
Remove global variable scanCommandId in favor of storing a command ID in snapshots, per my proposal of a few days ago. Also, tweak heapam.c routines (heap_insert, heap_update, heap_delete, heap_mark4update) to be passed the command ID to use, instead of doing GetCurrentCommandID. For catalog updates they'll still get passed current command ID, but for updates generated from the main executor they'll get passed the command ID saved in the snapshot the query is using. This should fix some corner cases associated with functions and triggers that advance current command ID while an outer query is still in progress.
2002-05-22 00:05:55 +02:00
heap_delete(Relation relation, ItemPointer tid,
Throw error if expiring tuple is again updated or deleted. This prevents surprising behavior when a FOR EACH ROW trigger BEFORE UPDATE or BEFORE DELETE directly or indirectly updates or deletes the the old row. Prior to this patch the requested action on the row could be silently ignored while all triggered actions based on the occurence of the requested action could be committed. One example of how this could happen is if the BEFORE DELETE trigger for a "parent" row deleted "children" which had trigger functions to update summary or status data on the parent. This also prevents similar surprising problems if the query has a volatile function which updates a target row while it is already being updated. There are related issues present in FOR UPDATE cursors and READ COMMITTED queries which are not handled by this patch. These issues need further evalution to determine what change, if any, is needed. Where the new error messages are generated, in most cases the best fix will be to move code from the BEFORE trigger to an AFTER trigger. Where this is not feasible, the trigger can avoid the error by re-issuing the triggering statement and returning NULL. Documentation changes will be submitted in a separate patch. Kevin Grittner and Tom Lane with input from Florian Pflug and Robert Haas, based on problems encountered during conversion of Wisconsin Circuit Court trigger logic to plpgsql triggers.
2012-10-26 21:55:36 +02:00
CommandId cid, Snapshot crosscheck, bool wait,
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
TM_FailureData *tmfd, bool changingPart)
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
TM_Result result;
Restructure subtransaction handling to reduce resource consumption, as per recent discussions. Invent SubTransactionIds that are managed like CommandIds (ie, counter is reset at start of each top transaction), and use these instead of TransactionIds to keep track of subtransaction status in those modules that need it. This means that a subtransaction does not need an XID unless it actually inserts/modifies rows in the database. Accordingly, don't assign it an XID nor take a lock on the XID until it tries to do that. This saves a lot of overhead for subtransactions that are only used for error recovery (eg plpgsql exceptions). Also, arrange to release a subtransaction's XID lock as soon as the subtransaction exits, in both the commit and abort cases. This avoids holding many unique locks after a long series of subtransactions. The price is some additional overhead in XactLockTableWait, but that seems acceptable. Finally, restructure the state machine in xact.c to have a more orthogonal set of states for subtransactions.
2004-09-16 18:58:44 +02:00
TransactionId xid = GetCurrentTransactionId();
New HeapTuple structure/interface.
1998-11-27 20:52:36 +01:00
ItemId lp;
HeapTupleData tp;
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
Page page;
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
BlockNumber block;
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
Buffer buffer;
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Buffer vmbuffer = InvalidBuffer;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
TransactionId new_xmax;
uint16 new_infomask,
new_infomask2;
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
bool have_tuple_lock = false;
Combine cmin and cmax fields of HeapTupleHeaders into a single field, by keeping private state in each backend that has inserted and deleted the same tuple during its current top-level transaction. This is sufficient since there is no need to be able to determine the cmin/cmax from any other transaction. This gets us back down to 23-byte headers, removing a penalty paid in 8.0 to support subtransactions. Patch by Heikki Linnakangas, with minor revisions by moi, following a design hashed out awhile back on the pghackers list.
2007-02-09 04:35:35 +01:00
bool iscombo;
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
bool all_visible_cleared = false;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
HeapTuple old_key_tuple = NULL; /* replica identity of the tuple */
bool old_key_copied = false;
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
Assert(ItemPointerIsValid(tid));
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Create an infrastructure for parallel computation in PostgreSQL. This does four basic things. First, it provides convenience routines to coordinate the startup and shutdown of parallel workers. Second, it synchronizes various pieces of state (e.g. GUCs, combo CID mappings, transaction snapshot) from the parallel group leader to the worker processes. Third, it prohibits various operations that would result in unsafe changes to that state while parallelism is active. Finally, it propagates events that would result in an ErrorResponse, NoticeResponse, or NotifyResponse message being sent to the client from the parallel workers back to the master, from which they can then be sent on to the client. Robert Haas, Amit Kapila, Noah Misch, Rushabh Lathia, Jeevan Chalke. Suggestions and review from Andres Freund, Heikki Linnakangas, Noah Misch, Simon Riggs, Euler Taveira, and Jim Nasby.
2015-04-30 21:02:14 +02:00
/*
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
* Forbid this during a parallel operation, lest it allocate a combo CID.
* Other workers might need that combo CID for visibility checks, and we
Create an infrastructure for parallel computation in PostgreSQL. This does four basic things. First, it provides convenience routines to coordinate the startup and shutdown of parallel workers. Second, it synchronizes various pieces of state (e.g. GUCs, combo CID mappings, transaction snapshot) from the parallel group leader to the worker processes. Third, it prohibits various operations that would result in unsafe changes to that state while parallelism is active. Finally, it propagates events that would result in an ErrorResponse, NoticeResponse, or NotifyResponse message being sent to the client from the parallel workers back to the master, from which they can then be sent on to the client. Robert Haas, Amit Kapila, Noah Misch, Rushabh Lathia, Jeevan Chalke. Suggestions and review from Andres Freund, Heikki Linnakangas, Noah Misch, Simon Riggs, Euler Taveira, and Jim Nasby.
2015-04-30 21:02:14 +02:00
* have no provision for broadcasting it to them.
*/
if (IsInParallelMode())
ereport(ERROR,
(errcode(ERRCODE_INVALID_TRANSACTION_STATE),
errmsg("cannot delete tuples during a parallel operation")));
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
block = ItemPointerGetBlockNumber(tid);
buffer = ReadBuffer(relation, block);
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(buffer);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
/*
* Before locking the buffer, pin the visibility map page if it appears to
* be necessary. Since we haven't got the lock yet, someone else might be
* in the middle of changing this, so we'll need to recheck after we have
* the lock.
*/
if (PageIsAllVisible(page))
visibilitymap_pin(relation, block, &vmbuffer);
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Fix race condition where heap_delete() fails to pin VM page. Similar to 5f12bc94dc, the code must re-check PageIsAllVisible() after buffer lock is re-acquired. Backpatching to the same version, 12. Discussion: https://postgr.es/m/CAEP4nAw9jYQDKd_5Y+-s2E4YiUJq1vqiikFjYGpLShtp-K3gag@mail.gmail.com Reported-by: Robins Tharakan Reviewed-by: Robins Tharakan Backpatch-through: 12
2022-09-22 19:58:49 +02:00
lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
Assert(ItemIdIsNormal(lp));
tp.t_tableOid = RelationGetRelid(relation);
tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
tp.t_len = ItemIdGetLength(lp);
tp.t_self = *tid;
l1:
Run pgindent on heapam.c An upcoming patch by Melanie Plageman does some refactoring work in this area. Run pgindent on that file now before making any changes so that it's easier to maintain/evolve each of the individual patches doing the refactor work. Additionally, add a few new required typedefs to the list to make it easier to do future pgindent runs on this file during the refactor work. Discussion: https://postgr.es/m/CAAKRu_YSOnhKsDyFcqJsKtBSrd32DP-jjXmv7hL0BPD-z0TGXQ@mail.gmail.com
2023-01-23 11:08:38 +01:00
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
/*
* If we didn't pin the visibility map page and the page has become all
* visible while we were busy locking the buffer, we'll have to unlock and
* re-lock, to avoid holding the buffer lock across an I/O. That's a bit
* unfortunate, but hopefully shouldn't happen often.
*/
if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
{
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
visibilitymap_pin(relation, block, &vmbuffer);
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
}
Adjust HeapTupleSatisfies* routines to take a HeapTuple. Previously, these functions took a HeapTupleHeader, but upcoming patches for logical replication will introduce new a new snapshot type under which the tuple's TID will be used to lookup (CMIN, CMAX) for visibility determination purposes. This makes that information available. Code churn is minimal since HeapTupleSatisfiesVisibility took the HeapTuple anyway, and deferenced it before calling the satisfies function. Independently of logical replication, this allows t_tableOid and t_self to be cross-checked via assertions in tqual.c. This seems like a useful way to make sure that all callers are setting these values properly, which has been previously put forward as desirable. Andres Freund, reviewed by Álvaro Herrera
2013-07-22 19:26:33 +02:00
result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
pgindent run over code.
1999-05-25 18:15:34 +02:00
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (result == TM_Invisible)
Catch non-functional delete attempts.
1997-08-06 04:08:39 +02:00
{
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
UnlockReleaseBuffer(buffer);
Fix a number of places that produced XX000 errors in the regression tests. It's against project policy to use elog() for user-facing errors, or to omit an errcode() selection for errors that aren't supposed to be "can't happen" cases. Fix all the violations of this policy that result in ERRCODE_INTERNAL_ERROR log entries during the standard regression tests, as errors that can reliably be triggered from SQL surely should be considered user-facing. I also looked through all the files touched by this commit and fixed other nearby problems of the same ilk. I do not claim to have fixed all violations of the policy, just the ones in these files. In a few places I also changed existing ERRCODE choices that didn't seem particularly appropriate; mainly replacing ERRCODE_SYNTAX_ERROR by something more specific. Back-patch to 9.5, but no further; changing ERRCODE assignments in stable branches doesn't seem like a good idea.
2015-08-03 05:49:19 +02:00
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("attempted to delete invisible tuple")));
Catch non-functional delete attempts.
1997-08-06 04:08:39 +02:00
}
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
else if (result == TM_BeingModified && wait)
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
{
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
TransactionId xwait;
uint16 infomask;
Ye-old pgindent run. Same 4-space tabs.
2000-04-12 19:17:23 +02:00
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
/* must copy state data before unlocking buffer */
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
infomask = tp.t_data->t_infomask;
/*
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
* Sleep until concurrent transaction ends -- except when there's a
* single locker and it's our own transaction. Note we don't care
* which lock mode the locker has, because we need the strongest one.
*
* Before sleeping, we need to acquire tuple lock to establish our
* priority for the tuple (see heap_lock_tuple). LockTuple will
* release us when we are next-in-line for the tuple.
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
*
* If we are forced to "start over" below, we keep the tuple lock;
* this arranges that we stay at the head of the line while rechecking
* tuple state.
*/
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
if (infomask & HEAP_XMAX_IS_MULTI)
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
{
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
bool current_is_member = false;
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
LockTupleExclusive, &current_is_member))
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
{
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
/*
* Acquire the lock, if necessary (but skip it when we're
* requesting a lock and already have one; avoids deadlock).
*/
if (!current_is_member)
heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
LockWaitBlock, &have_tuple_lock);
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
/* wait for multixact */
MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
relation, &(tp.t_self), XLTW_Delete,
NULL);
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
/*
* If xwait had just locked the tuple then some other xact
* could update this tuple before we get to this point. Check
* for xmax change, and start over if so.
Fix race condition where heap_delete() fails to pin VM page. Similar to 5f12bc94dc, the code must re-check PageIsAllVisible() after buffer lock is re-acquired. Backpatching to the same version, 12. Discussion: https://postgr.es/m/CAEP4nAw9jYQDKd_5Y+-s2E4YiUJq1vqiikFjYGpLShtp-K3gag@mail.gmail.com Reported-by: Robins Tharakan Reviewed-by: Robins Tharakan Backpatch-through: 12
2022-09-22 19:58:49 +02:00
*
* We also must start over if we didn't pin the VM page, and
* the page has become all visible.
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
*/
Fix race condition where heap_delete() fails to pin VM page. Similar to 5f12bc94dc, the code must re-check PageIsAllVisible() after buffer lock is re-acquired. Backpatching to the same version, 12. Discussion: https://postgr.es/m/CAEP4nAw9jYQDKd_5Y+-s2E4YiUJq1vqiikFjYGpLShtp-K3gag@mail.gmail.com Reported-by: Robins Tharakan Reviewed-by: Robins Tharakan Backpatch-through: 12
2022-09-22 19:58:49 +02:00
if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
!TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
xwait))
goto l1;
}
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
/*
* You might think the multixact is necessarily done here, but not
* so: it could have surviving members, namely our own xact or
* other subxacts of this backend. It is legal for us to delete
* the tuple in either case, however (the latter case is
* essentially a situation of upgrading our former shared lock to
* exclusive). We don't bother changing the on-disk hint bits
* since we are about to overwrite the xmax altogether.
*/
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
}
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
else if (!TransactionIdIsCurrentTransactionId(xwait))
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
{
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
/*
* Wait for regular transaction to end; but first, acquire tuple
* lock.
*/
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
LockWaitBlock, &have_tuple_lock);
Fix reference-after-free when waiting for another xact due to constraint. If an insertion or update had to wait for another transaction to finish, because there was another insertion with conflicting key in progress, we would pass a just-free'd item pointer to XactLockTableWait(). All calls to XactLockTableWait() and MultiXactIdWait() had similar issues. Some passed a pointer to a buffer in the buffer cache, after already releasing the lock. The call in EvalPlanQualFetch had already released the pin too. All but the call in execUtils.c would merely lead to reporting a bogus ctid, however (or an assertion failure, if enabled). All the callers that passed HeapTuple->t_data->t_ctid were slightly bogus anyway: if the tuple was updated (again) in the same transaction, its ctid field would point to the next tuple in the chain, not the tuple itself. Backpatch to 9.4, where the 'ctid' argument to XactLockTableWait was added (in commit f88d4cfc)
2015-02-04 15:00:34 +01:00
XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
/*
* xwait is done, but if xwait had just locked the tuple then some
* other xact could update this tuple before we get to this point.
* Check for xmax change, and start over if so.
Fix race condition where heap_delete() fails to pin VM page. Similar to 5f12bc94dc, the code must re-check PageIsAllVisible() after buffer lock is re-acquired. Backpatching to the same version, 12. Discussion: https://postgr.es/m/CAEP4nAw9jYQDKd_5Y+-s2E4YiUJq1vqiikFjYGpLShtp-K3gag@mail.gmail.com Reported-by: Robins Tharakan Reviewed-by: Robins Tharakan Backpatch-through: 12
2022-09-22 19:58:49 +02:00
*
* We also must start over if we didn't pin the VM page, and the
* page has become all visible.
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
*/
Fix race condition where heap_delete() fails to pin VM page. Similar to 5f12bc94dc, the code must re-check PageIsAllVisible() after buffer lock is re-acquired. Backpatching to the same version, 12. Discussion: https://postgr.es/m/CAEP4nAw9jYQDKd_5Y+-s2E4YiUJq1vqiikFjYGpLShtp-K3gag@mail.gmail.com Reported-by: Robins Tharakan Reviewed-by: Robins Tharakan Backpatch-through: 12
2022-09-22 19:58:49 +02:00
if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
!TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
xwait))
goto l1;
Fix oversight in async-commit patch: there were some places in heapam.c that still thought they could set HEAP_XMAX_COMMITTED immediately after seeing the other transaction commit. Make them use the same logic as tqual.c does to determine if the hint bit can be set yet.
2007-08-14 19:35:18 +02:00
/* Otherwise check if it committed or aborted */
UpdateXmaxHintBits(tp.t_data, buffer, xwait);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
}
/*
* We may overwrite if previous xmax aborted, or if it committed but
* only locked the tuple without updating it.
*/
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
HEAP_XMAX_IS_LOCKED_ONLY(tp.t_data->t_infomask) ||
HeapTupleHeaderIsOnlyLocked(tp.t_data))
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Ok;
Remove pointless HeapTupleHeaderIndicatesMovedPartitions calls Pavan Deolasee recently noted that a few of the HeapTupleHeaderIndicatesMovedPartitions calls added by commit 5db6df0c0117 are useless, since they are done after comparing t_self with t_ctid. But because t_self can never be set to the magical values that indicate that the tuple moved partition, this can never succeed: if the first test fails (so we know t_self equals t_ctid), necessarily the second test will also fail. So these checks can be removed and no harm is done. There's no bug here, just a code legibility issue. Reported-by: Pavan Deolasee <pavan.deolasee@gmail.com> Discussion: https://postgr.es/m/20200929164411.GA15497@alvherre.pgsql
2021-02-22 20:51:44 +01:00
else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Updated;
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
else
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Deleted;
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
}
Repair RI trigger visibility problems (this time for sure ;-)) per recent discussion on pgsql-hackers: in READ COMMITTED mode we just have to force a QuerySnapshot update in the trigger, but in SERIALIZABLE mode we have to run the scan under a current snapshot and then complain if any rows would be updated/deleted that are not visible in the transaction snapshot.
2003-10-01 23:30:53 +02:00
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (crosscheck != InvalidSnapshot && result == TM_Ok)
Repair RI trigger visibility problems (this time for sure ;-)) per recent discussion on pgsql-hackers: in READ COMMITTED mode we just have to force a QuerySnapshot update in the trigger, but in SERIALIZABLE mode we have to run the scan under a current snapshot and then complain if any rows would be updated/deleted that are not visible in the transaction snapshot.
2003-10-01 23:30:53 +02:00
{
SERIALIZABLE transactions are actually implemented beneath the covers with transaction snapshots, i.e. a snapshot registered at the beginning of a transaction. Change variable naming and comments to reflect this reality in preparation for a future, truly serializable mode, e.g. Serializable Snapshot Isolation (SSI). For the moment transaction snapshots are still used to implement SERIALIZABLE, but hopefully not for too much longer. Patch by Kevin Grittner and Dan Ports with review and some minor wording changes by me.
2010-09-11 20:38:58 +02:00
/* Perform additional check for transaction-snapshot mode RI updates */
Clean up the representation of special snapshots by including a "method pointer" in every Snapshot struct. This allows removal of the case-by-case tests in HeapTupleSatisfiesVisibility, which should make it a bit faster (I didn't try any performance tests though). More importantly, we are no longer violating portable C practices by assuming that small integers are distinct from all pointer values, and HeapTupleSatisfiesDirty no longer has a non-reentrant API involving side-effects on a global variable. There were a couple of places calling HeapTupleSatisfiesXXX routines directly rather than through the HeapTupleSatisfiesVisibility macro. Since these places had to be changed anyway, I chose to make them go through the macro for uniformity. Along the way I renamed HeapTupleSatisfiesSnapshot to HeapTupleSatisfiesMVCC to emphasize that it's only used with MVCC-type snapshots. I was sorely tempted to rename HeapTupleSatisfiesVisibility to HeapTupleSatisfiesSnapshot, but forebore for the moment to avoid confusion and reduce the likelihood that this patch breaks some of the pending patches. Might want to reconsider doing that later.
2007-03-25 21:45:14 +02:00
if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Updated;
Repair RI trigger visibility problems (this time for sure ;-)) per recent discussion on pgsql-hackers: in READ COMMITTED mode we just have to force a QuerySnapshot update in the trigger, but in SERIALIZABLE mode we have to run the scan under a current snapshot and then complain if any rows would be updated/deleted that are not visible in the transaction snapshot.
2003-10-01 23:30:53 +02:00
}
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (result != TM_Ok)
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
Assert(result == TM_SelfModified ||
result == TM_Updated ||
result == TM_Deleted ||
result == TM_BeingModified);
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
Assert(!(tp.t_data->t_infomask & HEAP_XMAX_INVALID));
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
Assert(result != TM_Updated ||
!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
tmfd->ctid = tp.t_data->t_ctid;
tmfd->xmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
if (result == TM_SelfModified)
tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
Throw error if expiring tuple is again updated or deleted. This prevents surprising behavior when a FOR EACH ROW trigger BEFORE UPDATE or BEFORE DELETE directly or indirectly updates or deletes the the old row. Prior to this patch the requested action on the row could be silently ignored while all triggered actions based on the occurence of the requested action could be committed. One example of how this could happen is if the BEFORE DELETE trigger for a "parent" row deleted "children" which had trigger functions to update summary or status data on the parent. This also prevents similar surprising problems if the query has a volatile function which updates a target row while it is already being updated. There are related issues present in FOR UPDATE cursors and READ COMMITTED queries which are not handled by this patch. These issues need further evalution to determine what change, if any, is needed. Where the new error messages are generated, in most cases the best fix will be to move code from the BEFORE trigger to an AFTER trigger. Where this is not feasible, the trigger can avoid the error by re-issuing the triggering statement and returning NULL. Documentation changes will be submitted in a separate patch. Kevin Grittner and Tom Lane with input from Florian Pflug and Robert Haas, based on problems encountered during conversion of Wisconsin Circuit Court trigger logic to plpgsql triggers.
2012-10-26 21:55:36 +02:00
else
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
tmfd->cmax = InvalidCommandId;
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
UnlockReleaseBuffer(buffer);
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
if (have_tuple_lock)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
if (vmbuffer != InvalidBuffer)
ReleaseBuffer(vmbuffer);
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
return result;
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
}
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
/*
* We're about to do the actual delete -- check for conflict first, to
* avoid possibly having to roll back work we've just done.
Fix serialization anomalies due to race conditions on INSERT. On insert the CheckForSerializableConflictIn() test was performed before the page(s) which were going to be modified had been locked (with an exclusive buffer content lock). If another process acquired a relation SIReadLock on the heap and scanned to a page on which an insert was going to occur before the page was so locked, a rw-conflict would be missed, which could allow a serialization anomaly to be missed. The window between the check and the page lock was small, so the bug was generally not noticed unless there was high concurrency with multiple processes inserting into the same table. This was reported by Peter Bailis as bug #11732, by Sean Chittenden as bug #13667, and by others. The race condition was eliminated in heap_insert() by moving the check down below the acquisition of the buffer lock, which had been the very next statement. Because of the loop locking and unlocking multiple buffers in heap_multi_insert() a check was added after all inserts were completed. The check before the start of the inserts was left because it might avoid a large amount of work to detect a serialization anomaly before performing the all of the inserts and the related WAL logging. While investigating this bug, other SSI bugs which were even harder to hit in practice were noticed and fixed, an unnecessary check (covered by another check, so redundant) was removed from heap_update(), and comments were improved. Back-patch to all supported branches. Kevin Grittner and Thomas Munro
2015-10-31 20:43:34 +01:00
*
* This is safe without a recheck as long as there is no possibility of
* another process scanning the page between this check and the delete
* being visible to the scan (i.e., an exclusive buffer content lock is
* continuously held from this point until the tuple delete is visible).
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
*/
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
CheckForSerializableConflictIn(relation, tid, BufferGetBlockNumber(buffer));
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
/* replace cid with a combo CID if necessary */
Combine cmin and cmax fields of HeapTupleHeaders into a single field, by keeping private state in each backend that has inserted and deleted the same tuple during its current top-level transaction. This is sufficient since there is no need to be able to determine the cmin/cmax from any other transaction. This gets us back down to 23-byte headers, removing a penalty paid in 8.0 to support subtransactions. Patch by Heikki Linnakangas, with minor revisions by moi, following a design hashed out awhile back on the pghackers list.
2007-02-09 04:35:35 +01:00
HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
/*
* Compute replica identity tuple before entering the critical section so
* we don't PANIC upon a memory allocation failure.
*/
old_key_tuple = ExtractReplicaIdentity(relation, &tp, true, &old_key_copied);
Move multixid allocation out of critical section. It can fail if you run out of memory. This call was added in 9.3, so backpatch to 9.3 only.
2014-04-04 17:20:22 +02:00
/*
* If this is the first possibly-multixact-able operation in the current
* transaction, set my per-backend OldestMemberMXactId setting. We can be
* certain that the transaction will never become a member of any older
* MultiXactIds than that. (We have to do this even if we end up just
* using our own TransactionId below, since some other backend could
* incorporate our XID into a MultiXact immediately afterwards.)
*/
MultiXactIdSetOldestMember();
compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(tp.t_data),
tp.t_data->t_infomask, tp.t_data->t_infomask2,
xid, LockTupleExclusive, true,
&new_xmax, &new_infomask, &new_infomask2);
Add more critical-section calls: all code sections that hold spinlocks are now critical sections, so as to ensure die() won't interrupt us while we are munging shared-memory data structures. Avoid insecure intermediate states in some code that proc_exit will call, like palloc/pfree. Rename START/END_CRIT_CODE to START/END_CRIT_SECTION, since that seems to be what people tend to call them anyway, and make them be called with () like a function call, in hopes of not confusing pg_indent. I doubt that this is sufficient to make SIGTERM safe anywhere; there's just too much code that could get invoked during proc_exit().
2001-01-12 22:54:01 +01:00
START_CRIT_SECTION();
Restructure local-buffer handling per recent pghackers discussion. The local buffer manager is no longer used for newly-created relations (unless they are TEMP); a new non-TEMP relation goes through the shared bufmgr and thus will participate normally in checkpoints. But TEMP relations use the local buffer manager throughout their lifespan. Also, operations in TEMP relations are not logged in WAL, thus improving performance. Since it's no longer necessary to fsync relations as they move out of the local buffers into shared buffers, quite a lot of smgr.c/md.c/fd.c code is no longer needed and has been removed: there's no concept of a dirty relation anymore in md.c/fd.c, and we never fsync anything but WAL. Still TODO: improve local buffer management algorithms so that it would be reasonable to increase NLocBuffer.
2002-08-06 04:36:35 +02:00
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
/*
* If this transaction commits, the tuple will become DEAD sooner or
Improve handling of prune/no-prune decisions by storing a page's oldest unpruned XMAX in its header. At the cost of 4 bytes per page, this keeps us from performing heap_page_prune when there's no chance of pruning anything. Seems to be necessary per Heikki's preliminary performance testing.
2007-09-21 23:25:42 +02:00
* later. Set flag that this page is a candidate for pruning once our xid
* falls below the OldestXmin horizon. If the transaction finally aborts,
* the subsequent page pruning will be a no-op and the hint will be
* cleared.
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
*/
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
PageSetPrunable(page, xid);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
if (PageIsAllVisible(page))
{
all_visible_cleared = true;
PageClearAllVisible(page);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
vmbuffer, VISIBILITYMAP_VALID_BITS);
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
}
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
/* store transaction information of xact deleting the tuple */
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
tp.t_data->t_infomask |= new_infomask;
tp.t_data->t_infomask2 |= new_infomask2;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
HeapTupleHeaderClearHotUpdated(tp.t_data);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
Combine cmin and cmax fields of HeapTupleHeaders into a single field, by keeping private state in each backend that has inserted and deleted the same tuple during its current top-level transaction. This is sufficient since there is no need to be able to determine the cmin/cmax from any other transaction. This gets us back down to 23-byte headers, removing a penalty paid in 8.0 to support subtransactions. Patch by Heikki Linnakangas, with minor revisions by moi, following a design hashed out awhile back on the pghackers list.
2007-02-09 04:35:35 +01:00
HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
Make sure that t_ctid is reset to equal t_self in heap_delete and heap_mark4update; this avoids situations where a deleted tuple might look like it is chained to something else. Also, cause all the WAL redo routines to set t_ctid to equal t_self, rather than leaving it undefined as before. Make heap_xlog_clean set the page's LSN and SUI correctly. All per past discussions in pghackers, ranging back to last December.
2002-08-13 22:11:03 +02:00
/* Make sure there is no forward chain link in t_ctid */
tp.t_data->t_ctid = tp.t_self;
Restructure local-buffer handling per recent pghackers discussion. The local buffer manager is no longer used for newly-created relations (unless they are TEMP); a new non-TEMP relation goes through the shared bufmgr and thus will participate normally in checkpoints. But TEMP relations use the local buffer manager throughout their lifespan. Also, operations in TEMP relations are not logged in WAL, thus improving performance. Since it's no longer necessary to fsync relations as they move out of the local buffers into shared buffers, quite a lot of smgr.c/md.c/fd.c code is no longer needed and has been removed: there's no concept of a dirty relation anymore in md.c/fd.c, and we never fsync anything but WAL. Still TODO: improve local buffer management algorithms so that it would be reasonable to increase NLocBuffer.
2002-08-06 04:36:35 +02:00
Raise error when affecting tuple moved into different partition. When an update moves a row between partitions (supported since 2f178441044b), our normal logic for following update chains in READ COMMITTED mode doesn't work anymore. Cross partition updates are modeled as an delete from the old and insert into the new partition. No ctid chain exists across partitions, and there's no convenient space to introduce that link. Not throwing an error in a partitioned context when one would have been thrown without partitioning is obviously problematic. This commit introduces infrastructure to detect when a tuple has been moved, not just plainly deleted. That allows to throw an error when encountering a deletion that's actually a move, while attempting to following a ctid chain. The row deleted as part of a cross partition update is marked by pointing it's t_ctid to an invalid block, instead of self as a normal update would. That was deemed to be the least invasive and most future proof way to represent the knowledge, given how few infomask bits are there to be recycled (there's also some locking issues with using infomask bits). External code following ctid chains should be updated to check for moved tuples. The most likely consequence of not doing so is a missed error. Author: Amul Sul, editorialized by me Reviewed-By: Amit Kapila, Pavan Deolasee, Andres Freund, Robert Haas Discussion: http://postgr.es/m/CAAJ_b95PkwojoYfz0bzXU8OokcTVGzN6vYGCNVUukeUDrnF3dw@mail.gmail.com
2018-04-07 22:24:10 +02:00
/* Signal that this is actually a move into another partition */
if (changingPart)
HeapTupleHeaderSetMovedPartitions(tp.t_data);
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
MarkBufferDirty(buffer);
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
/*
* XLOG stuff
*
* NB: heap_abort_speculative() uses the same xlog record and replay
* routines.
*/
Generalize concept of temporary relations to "relation persistence". This commit replaces pg_class.relistemp with pg_class.relpersistence; and also modifies the RangeVar node type to carry relpersistence rather than istemp. It also removes removes rd_istemp from RelationData and instead performs the correct computation based on relpersistence. For clarity, we add three new macros: RelationNeedsWAL(), RelationUsesLocalBuffers(), and RelationUsesTempNamespace(), so that we can clarify the purpose of each check that previous depended on rd_istemp. This is intended as infrastructure for the upcoming unlogged tables patch, as well as for future possible work on global temporary tables.
2010-12-13 18:34:26 +01:00
if (RelationNeedsWAL(relation))
heap' xlog records
2000-06-02 12:20:27 +02:00
{
xl_heap_delete xlrec;
Fix use of dangling pointer in heap_delete() when logging replica identity When logging the replica identity of a deleted tuple, XLOG_HEAP_DELETE records include references of the old tuple. Its data is stored in an intermediate variable used to register this information for the WAL record, but this variable gets away from the stack when the record gets actually inserted. Spotted by clang's AddressSanitizer. Author: Stas Kelvish Discussion: https://postgr.es/m/085C8825-AD86-4E93-AF80-E26CDF03D1EA@postgrespro.ru Backpatch-through: 9.4
2019-02-01 02:35:16 +01:00
xl_heap_header xlhdr;
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
XLogRecPtr recptr;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
/*
* For logical decode we need combo CIDs to properly decode the
* catalog
*/
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
if (RelationIsAccessibleInLogicalDecoding(relation))
log_heap_new_cid(relation, &tp);
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
Raise error when affecting tuple moved into different partition. When an update moves a row between partitions (supported since 2f178441044b), our normal logic for following update chains in READ COMMITTED mode doesn't work anymore. Cross partition updates are modeled as an delete from the old and insert into the new partition. No ctid chain exists across partitions, and there's no convenient space to introduce that link. Not throwing an error in a partitioned context when one would have been thrown without partitioning is obviously problematic. This commit introduces infrastructure to detect when a tuple has been moved, not just plainly deleted. That allows to throw an error when encountering a deletion that's actually a move, while attempting to following a ctid chain. The row deleted as part of a cross partition update is marked by pointing it's t_ctid to an invalid block, instead of self as a normal update would. That was deemed to be the least invasive and most future proof way to represent the knowledge, given how few infomask bits are there to be recycled (there's also some locking issues with using infomask bits). External code following ctid chains should be updated to check for moved tuples. The most likely consequence of not doing so is a missed error. Author: Amul Sul, editorialized by me Reviewed-By: Amit Kapila, Pavan Deolasee, Andres Freund, Robert Haas Discussion: http://postgr.es/m/CAAJ_b95PkwojoYfz0bzXU8OokcTVGzN6vYGCNVUukeUDrnF3dw@mail.gmail.com
2018-04-07 22:24:10 +02:00
xlrec.flags = 0;
if (all_visible_cleared)
xlrec.flags |= XLH_DELETE_ALL_VISIBLE_CLEARED;
if (changingPart)
xlrec.flags |= XLH_DELETE_IS_PARTITION_MOVE;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
tp.t_data->t_infomask2);
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
xlrec.xmax = new_xmax;
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (old_key_tuple != NULL)
{
if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
xlrec.flags |= XLH_DELETE_CONTAINS_OLD_TUPLE;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
else
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
xlrec.flags |= XLH_DELETE_CONTAINS_OLD_KEY;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
}
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHeapDelete);
XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
/*
* Log replica identity of the deleted tuple if there is one
*/
if (old_key_tuple != NULL)
{
xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterData((char *) &xlhdr, SizeOfHeapHeader);
XLogRegisterData((char *) old_key_tuple->t_data
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
+ SizeofHeapTupleHeader,
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
old_key_tuple->t_len
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
- SizeofHeapTupleHeader);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
}
Introduce replication progress tracking infrastructure. When implementing a replication solution ontop of logical decoding, two related problems exist: * How to safely keep track of replication progress * How to change replication behavior, based on the origin of a row; e.g. to avoid loops in bi-directional replication setups The solution to these problems, as implemented here, consist out of three parts: 1) 'replication origins', which identify nodes in a replication setup. 2) 'replication progress tracking', which remembers, for each replication origin, how far replay has progressed in a efficient and crash safe manner. 3) The ability to filter out changes performed on the behest of a replication origin during logical decoding; this allows complex replication topologies. E.g. by filtering all replayed changes out. Most of this could also be implemented in "userspace", e.g. by inserting additional rows contain origin information, but that ends up being much less efficient and more complicated. We don't want to require various replication solutions to reimplement logic for this independently. The infrastructure is intended to be generic enough to be reusable. This infrastructure also replaces the 'nodeid' infrastructure of commit timestamps. It is intended to provide all the former capabilities, except that there's only 2^16 different origins; but now they integrate with logical decoding. Additionally more functionality is accessible via SQL. Since the commit timestamp infrastructure has also been introduced in 9.5 (commit 73c986add) changing the API is not a problem. For now the number of origins for which the replication progress can be tracked simultaneously is determined by the max_replication_slots GUC. That GUC is not a perfect match to configure this, but there doesn't seem to be sufficient reason to introduce a separate new one. Bumps both catversion and wal page magic. Author: Andres Freund, with contributions from Petr Jelinek and Craig Ringer Reviewed-By: Heikki Linnakangas, Petr Jelinek, Robert Haas, Steve Singer Discussion: 20150216002155.GI15326@awork2.anarazel.de, 20140923182422.GA15776@alap3.anarazel.de, 20131114172632.GE7522@alap2.anarazel.de
2015-04-29 19:30:53 +02:00
/* filtering by origin on a row level is much more efficient */
Skip checkpoints, archiving on idle systems. Some background activity (like checkpoints, archive timeout, standby snapshots) is not supposed to happen on an idle system. Unfortunately so far it was not easy to determine when a system is idle, which defeated some of the attempts to avoid redundant activity on an idle system. To make that easier, allow to make individual WAL insertions as not being "important". By checking whether any important activity happened since the last time an activity was performed, it now is easy to check whether some action needs to be repeated. Use the new facility for checkpoints, archive timeout and standby snapshots. The lack of a facility causes some issues in older releases, but in my opinion the consequences (superflous checkpoints / archived segments) aren't grave enough to warrant backpatching. Author: Michael Paquier, editorialized by Andres Freund Reviewed-By: Andres Freund, David Steele, Amit Kapila, Kyotaro HORIGUCHI Bug: #13685 Discussion: https://www.postgresql.org/message-id/20151016203031.3019.72930@wrigleys.postgresql.org https://www.postgresql.org/message-id/CAB7nPqQcPqxEM3S735Bd2RzApNqSNJVietAC=6kfkYv_45dKwA@mail.gmail.com Backpatch: -
2016-12-22 20:31:50 +01:00
XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
Introduce replication progress tracking infrastructure. When implementing a replication solution ontop of logical decoding, two related problems exist: * How to safely keep track of replication progress * How to change replication behavior, based on the origin of a row; e.g. to avoid loops in bi-directional replication setups The solution to these problems, as implemented here, consist out of three parts: 1) 'replication origins', which identify nodes in a replication setup. 2) 'replication progress tracking', which remembers, for each replication origin, how far replay has progressed in a efficient and crash safe manner. 3) The ability to filter out changes performed on the behest of a replication origin during logical decoding; this allows complex replication topologies. E.g. by filtering all replayed changes out. Most of this could also be implemented in "userspace", e.g. by inserting additional rows contain origin information, but that ends up being much less efficient and more complicated. We don't want to require various replication solutions to reimplement logic for this independently. The infrastructure is intended to be generic enough to be reusable. This infrastructure also replaces the 'nodeid' infrastructure of commit timestamps. It is intended to provide all the former capabilities, except that there's only 2^16 different origins; but now they integrate with logical decoding. Additionally more functionality is accessible via SQL. Since the commit timestamp infrastructure has also been introduced in 9.5 (commit 73c986add) changing the API is not a problem. For now the number of origins for which the replication progress can be tracked simultaneously is determined by the max_replication_slots GUC. That GUC is not a perfect match to configure this, but there doesn't seem to be sufficient reason to introduce a separate new one. Bumps both catversion and wal page magic. Author: Andres Freund, with contributions from Petr Jelinek and Craig Ringer Reviewed-By: Heikki Linnakangas, Petr Jelinek, Robert Haas, Steve Singer Discussion: 20150216002155.GI15326@awork2.anarazel.de, 20140923182422.GA15776@alap3.anarazel.de, 20131114172632.GE7522@alap2.anarazel.de
2015-04-29 19:30:53 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
heap' xlog records
2000-06-02 12:20:27 +02:00
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
PageSetLSN(page, recptr);
heap' xlog records
2000-06-02 12:20:27 +02:00
}
Restructure local-buffer handling per recent pghackers discussion. The local buffer manager is no longer used for newly-created relations (unless they are TEMP); a new non-TEMP relation goes through the shared bufmgr and thus will participate normally in checkpoints. But TEMP relations use the local buffer manager throughout their lifespan. Also, operations in TEMP relations are not logged in WAL, thus improving performance. Since it's no longer necessary to fsync relations as they move out of the local buffers into shared buffers, quite a lot of smgr.c/md.c/fd.c code is no longer needed and has been removed: there's no concept of a dirty relation anymore in md.c/fd.c, and we never fsync anything but WAL. Still TODO: improve local buffer management algorithms so that it would be reasonable to increase NLocBuffer.
2002-08-06 04:36:35 +02:00
Add more critical-section calls: all code sections that hold spinlocks are now critical sections, so as to ensure die() won't interrupt us while we are munging shared-memory data structures. Avoid insecure intermediate states in some code that proc_exit will call, like palloc/pfree. Rename START/END_CRIT_CODE to START/END_CRIT_SECTION, since that seems to be what people tend to call them anyway, and make them be called with () like a function call, in hopes of not confusing pg_indent. I doubt that this is sufficient to make SIGTERM safe anywhere; there's just too much code that could get invoked during proc_exit().
2001-01-12 22:54:01 +01:00
END_CRIT_SECTION();
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Tweak heap_update/delete so that we do not hold the buffer context lock on the old tuple's page while we are doing TOAST pushups.
2001-01-15 06:29:19 +01:00
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
if (vmbuffer != InvalidBuffer)
ReleaseBuffer(vmbuffer);
TOAST WARNING: This is actually broken - we have self-deadlocks due to concurrent changes in buffer management. Vadim and me are working on it. Jan
2000-07-04 01:10:14 +02:00
/*
Tighten short-circuit tests for deciding whether we need to invoke tuptoaster.c --- fields that are compressed in-line are not a reason to invoke the toaster. Along the way, add a couple more htup.h macros to eliminate confusing negated tests, and get rid of the already vestigial TUPLE_TOASTER_ACTIVE symbol.
2004-01-16 21:51:30 +01:00
* If the tuple has toasted out-of-line attributes, we need to delete
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
* those items too. We have to do this before releasing the buffer
* because we need to look at the contents of the tuple, but it's OK to
* release the content lock on the buffer first.
TOAST WARNING: This is actually broken - we have self-deadlocks due to concurrent changes in buffer management. Vadim and me are working on it. Jan
2000-07-04 01:10:14 +02:00
*/
Add a materialized view relations. A materialized view has a rule just like a view and a heap and other physical properties like a table. The rule is only used to populate the table, references in queries refer to the materialized data. This is a minimal implementation, but should still be useful in many cases. Currently data is only populated "on demand" by the CREATE MATERIALIZED VIEW and REFRESH MATERIALIZED VIEW statements. It is expected that future releases will add incremental updates with various timings, and that a more refined concept of defining what is "fresh" data will be developed. At some point it may even be possible to have queries use a materialized in place of references to underlying tables, but that requires the other above-mentioned features to be working first. Much of the documentation work by Robert Haas. Review by Noah Misch, Thom Brown, Robert Haas, Marko Tiikkaja Security review by KaiGai Kohei, with a decision on how best to implement sepgsql still pending.
2013-03-04 01:23:31 +01:00
if (relation->rd_rel->relkind != RELKIND_RELATION &&
relation->rd_rel->relkind != RELKIND_MATVIEW)
Decouple the values of TOAST_TUPLE_THRESHOLD and TOAST_MAX_CHUNK_SIZE. Add the latter to the values checked in pg_control, since it can't be changed without invalidating toast table content. This commit in itself shouldn't change any behavior, but it lays some necessary groundwork for experimentation with these toast-control numbers. Note: while TOAST_TUPLE_THRESHOLD can now be changed without initdb, some thought still needs to be given to needs_toast_table() in toasting.c before unleashing random changes.
2007-04-03 06:14:26 +02:00
{
/* toast table entries should never be recursively toasted */
Assert(!HeapTupleHasExternal(&tp));
}
else if (HeapTupleHasExternal(&tp))
Rename some toasting functions based on whether they are heap-specific. The old names for the attribute-detoasting functions names included the word "heap," which seems outdated now that the heap is only one of potentially many table access methods. On the other hand, toast_insert_or_update and toast_delete are heap-specific, so rename them by adding "heap_" as a prefix. Not all of the work of making the TOAST system fully accessible to AMs other than the heap is done yet, but there seems to be little harm in getting this renaming out of the way now. Commit 8b94dab06617ef80a0901ab103ebd8754427ef5a already divided up the functions among various files partially according to whether it was intended that they should be heap-specific or AM-agnostic, so this is just clarifying the division contemplated by that commit. Patch by me, reviewed and tested by Prabhat Sabu, Thomas Munro, Andres Freund, and Álvaro Herrera. Discussion: http://postgr.es/m/CA+TgmoZv-=2iWM4jcw5ZhJeL18HF96+W1yJeYrnGMYdkFFnEpQ@mail.gmail.com
2019-10-04 20:24:46 +02:00
heap_toast_delete(relation, &tp, false);
TOAST WARNING: This is actually broken - we have self-deadlocks due to concurrent changes in buffer management. Vadim and me are working on it. Jan
2000-07-04 01:10:14 +02:00
Correct nasty error in heap_update: it was releasing the buffer refcount before calling RelationInvalidateHeapTuple(), which is bad because the latter needs to look at the tuple data, which is in the shared disk buffer. If another backend manages to recycle the buffer while this is going on, we will compute the wrong hashindex for the tuple or maybe even crash outright. Must hold buffer refcount until afterwards. (This bug is not in 7.0.*; seems to be have introduced during WAL changes.)
2001-01-07 23:14:31 +01:00
/*
* Mark tuple for invalidation from system caches at next command
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
* boundary. We have to do this before releasing the buffer because we
* need to look at the contents of the tuple.
Correct nasty error in heap_update: it was releasing the buffer refcount before calling RelationInvalidateHeapTuple(), which is bad because the latter needs to look at the tuple data, which is in the shared disk buffer. If another backend manages to recycle the buffer while this is going on, we will compute the wrong hashindex for the tuple or maybe even crash outright. Must hold buffer refcount until afterwards. (This bug is not in 7.0.*; seems to be have introduced during WAL changes.)
2001-01-07 23:14:31 +01:00
*/
Revise sinval code to remove no-longer-used tuple TID from inval messages. This requires adjusting the API for syscache callback functions: they now get a hash value, not a TID, to identify the target tuple. Most of them weren't paying any attention to that argument anyway, but plancache did require a small amount of fixing. Also, improve performance a trifle by avoiding sending duplicate inval messages when a heap_update isn't changing the catcache lookup columns.
2011-08-17 01:27:46 +02:00
CacheInvalidateHeapTuple(relation, &tp, NULL);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
/* Now we can release the buffer */
ReleaseBuffer(buffer);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
/*
* Release the lmgr tuple lock, if we had it.
*/
if (have_tuple_lock)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
Fix up pgstats counting of live and dead tuples to recognize that committed and aborted transactions have different effects; also teach it not to assume that prepared transactions are always committed. Along the way, simplify the pgstats API by tying counting directly to Relations; I cannot detect any redeeming social value in having stats pointers in HeapScanDesc and IndexScanDesc structures. And fix a few corner cases in which counts might be missed because the relation's pgstat_info pointer hadn't been set.
2007-05-27 05:50:39 +02:00
pgstat_count_heap_delete(relation);
Revise pgstats stuff to fix the problems with not counting accesses generated by bitmap index scans. Along the way, simplify and speed up the code for counting sequential and index scans; it was both confusing and inefficient to be taking care of that in the per-tuple loops, IMHO. initdb forced because of internal changes in pg_stat view definitions.
2005-10-06 04:29:23 +02:00
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
if (old_key_tuple != NULL && old_key_copied)
heap_freetuple(old_key_tuple);
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
return TM_Ok;
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
}
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
/*
* simple_heap_delete - delete a tuple
*
* This routine may be used to delete a tuple when concurrent updates of
* the target tuple are not expected (for example, because we have a lock
* on the relation associated with the tuple). Any failure is reported
Error message editing in backend/access.
2003-07-21 22:29:40 +02:00
* via ereport().
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
*/
void
simple_heap_delete(Relation relation, ItemPointer tid)
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
TM_Result result;
TM_FailureData tmfd;
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
Fix LISTEN/NOTIFY race condition reported by Gavin Sherry. While a really general fix might be difficult, I believe the only case where AtCommit_Notify could see an uncommitted tuple is where the other guy has just unlistened and not yet committed. The best solution seems to be to just skip updating that tuple, on the assumption that the other guy does not want to hear about the notification anyway. This is not perfect --- if the other guy rolls back his unlisten instead of committing, then he really should have gotten this notify. But to do that, we'd have to wait to see if he commits or not, or make UNLISTEN hold exclusive lock on pg_listener until commit. Either of these answers is deadlock-prone, not to mention horrible for interactive performance. Do it this way for now. (What happened to that project to do LISTEN/NOTIFY in memory with no table, anyway?)
2003-09-16 01:33:43 +02:00
result = heap_delete(relation, tid,
Avoid incrementing the CommandCounter when CommandCounterIncrement is called but no database changes have been made since the last CommandCounterIncrement. This should result in a significant improvement in the number of "commands" that can typically be performed within a transaction before hitting the 2^32 CommandId size limit. In particular this buys back (and more) the possible adverse consequences of my previous patch to fix plan caching behavior. The implementation requires tracking whether the current CommandCounter value has been "used" to mark any tuples. CommandCounter values stored into snapshots are presumed not to be used for this purpose. This requires some small executor changes, since the executor used to conflate the curcid of the snapshot it was using with the command ID to mark output tuples with. Separating these concepts allows some small simplifications in executor APIs. Something for the TODO list: look into having CommandCounterIncrement not do AcceptInvalidationMessages. It seems fairly bogus to be doing it there, but exactly where to do it instead isn't clear, and I'm disinclined to mess with asynchronous behavior during late beta.
2007-11-30 22:22:54 +01:00
GetCurrentCommandId(true), InvalidSnapshot,
Throw error if expiring tuple is again updated or deleted. This prevents surprising behavior when a FOR EACH ROW trigger BEFORE UPDATE or BEFORE DELETE directly or indirectly updates or deletes the the old row. Prior to this patch the requested action on the row could be silently ignored while all triggered actions based on the occurence of the requested action could be committed. One example of how this could happen is if the BEFORE DELETE trigger for a "parent" row deleted "children" which had trigger functions to update summary or status data on the parent. This also prevents similar surprising problems if the query has a volatile function which updates a target row while it is already being updated. There are related issues present in FOR UPDATE cursors and READ COMMITTED queries which are not handled by this patch. These issues need further evalution to determine what change, if any, is needed. Where the new error messages are generated, in most cases the best fix will be to move code from the BEFORE trigger to an AFTER trigger. Where this is not feasible, the trigger can avoid the error by re-issuing the triggering statement and returning NULL. Documentation changes will be submitted in a separate patch. Kevin Grittner and Tom Lane with input from Florian Pflug and Robert Haas, based on problems encountered during conversion of Wisconsin Circuit Court trigger logic to plpgsql triggers.
2012-10-26 21:55:36 +02:00
true /* wait for commit */ ,
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
&tmfd, false /* changingPart */ );
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
switch (result)
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
case TM_SelfModified:
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
/* Tuple was already updated in current command? */
Error message editing in backend/access.
2003-07-21 22:29:40 +02:00
elog(ERROR, "tuple already updated by self");
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
break;
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
case TM_Ok:
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
/* done successfully */
break;
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
case TM_Updated:
Error message editing in backend/access.
2003-07-21 22:29:40 +02:00
elog(ERROR, "tuple concurrently updated");
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
break;
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
case TM_Deleted:
elog(ERROR, "tuple concurrently deleted");
break;
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
default:
Error message editing in backend/access.
2003-07-21 22:29:40 +02:00
elog(ERROR, "unrecognized heap_delete status: %u", result);
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
break;
}
}
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
/*
Rename heap_replace to heap_update.
1999-11-24 01:44:37 +01:00
* heap_update - replace a tuple
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
*
tableam: Rename wrapper functions to match callback names. Some of the wrapper functions didn't match the callback names. Many of them due to staying "consistent" with historic naming of the wrapped functionality. We decided that for most cases it's more important to be for tableam to be consistent going forward, than with the past. The one exception is beginscan/endscan/... because it'd have looked odd to have systable_beginscan/endscan/... with a different naming scheme, and changing the systable_* APIs would have caused way too much churn (including breaking a lot of external users). Author: Ashwin Agrawal, with some small additions by Andres Freund Reviewed-By: Andres Freund Discussion: https://postgr.es/m/CALfoeiugyrXZfX7n0ORCa4L-m834dzmaE8eFdbNR6PMpetU4Ww@mail.gmail.com
2019-05-24 01:25:48 +02:00
* See table_tuple_update() for an explanation of the parameters, except that
* this routine directly takes a tuple rather than a slot.
Fix LISTEN/NOTIFY race condition reported by Gavin Sherry. While a really general fix might be difficult, I believe the only case where AtCommit_Notify could see an uncommitted tuple is where the other guy has just unlistened and not yet committed. The best solution seems to be to just skip updating that tuple, on the assumption that the other guy does not want to hear about the notification anyway. This is not perfect --- if the other guy rolls back his unlisten instead of committing, then he really should have gotten this notify. But to do that, we'd have to wait to see if he commits or not, or make UNLISTEN hold exclusive lock on pg_listener until commit. Either of these answers is deadlock-prone, not to mention horrible for interactive performance. Do it this way for now. (What happened to that project to do LISTEN/NOTIFY in memory with no table, anyway?)
2003-09-16 01:33:43 +02:00
*
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
* In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
* t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
* only for TM_SelfModified, since we cannot obtain cmax from a combo CID
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
* generated by another transaction).
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
TM_Result
Rename heap_replace to heap_update.
1999-11-24 01:44:37 +01:00
heap_update(Relation relation, ItemPointer otid, HeapTuple newtup,
Throw error if expiring tuple is again updated or deleted. This prevents surprising behavior when a FOR EACH ROW trigger BEFORE UPDATE or BEFORE DELETE directly or indirectly updates or deletes the the old row. Prior to this patch the requested action on the row could be silently ignored while all triggered actions based on the occurence of the requested action could be committed. One example of how this could happen is if the BEFORE DELETE trigger for a "parent" row deleted "children" which had trigger functions to update summary or status data on the parent. This also prevents similar surprising problems if the query has a volatile function which updates a target row while it is already being updated. There are related issues present in FOR UPDATE cursors and READ COMMITTED queries which are not handled by this patch. These issues need further evalution to determine what change, if any, is needed. Where the new error messages are generated, in most cases the best fix will be to move code from the BEFORE trigger to an AFTER trigger. Where this is not feasible, the trigger can avoid the error by re-issuing the triggering statement and returning NULL. Documentation changes will be submitted in a separate patch. Kevin Grittner and Tom Lane with input from Florian Pflug and Robert Haas, based on problems encountered during conversion of Wisconsin Circuit Court trigger logic to plpgsql triggers.
2012-10-26 21:55:36 +02:00
CommandId cid, Snapshot crosscheck, bool wait,
Ignore BRIN indexes when checking for HOT updates When determining whether an index update may be skipped by using HOT, we can ignore attributes indexed by block summarizing indexes without references to individual tuples that need to be cleaned up. A new type TU_UpdateIndexes provides a signal to the executor to determine which indexes to update - no indexes, all indexes, or only the summarizing indexes. This also removes rd_indexattr list, and replaces it with rd_attrsvalid flag. The list was not used anywhere, and a simple flag is sufficient. This was originally committed as 5753d4ee32, but then got reverted by e3fcca0d0d because of correctness issues. Original patch by Josef Simanek, various fixes and improvements by Tomas Vondra and me. Authors: Matthias van de Meent, Josef Simanek, Tomas Vondra Reviewed-by: Tomas Vondra, Alvaro Herrera Discussion: https://postgr.es/m/05ebcb44-f383-86e3-4f31-0a97a55634cf@enterprisedb.com Discussion: https://postgr.es/m/CAFp7QwpMRGcDAQumN7onN9HjrJ3u4X3ZRXdGFT0K5G2JWvnbWg%40mail.gmail.com
2023-03-20 10:34:07 +01:00
TM_FailureData *tmfd, LockTupleMode *lockmode,
TU_UpdateIndexes *update_indexes)
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
TM_Result result;
Restructure subtransaction handling to reduce resource consumption, as per recent discussions. Invent SubTransactionIds that are managed like CommandIds (ie, counter is reset at start of each top transaction), and use these instead of TransactionIds to keep track of subtransaction status in those modules that need it. This means that a subtransaction does not need an XID unless it actually inserts/modifies rows in the database. Accordingly, don't assign it an XID nor take a lock on the XID until it tries to do that. This saves a lot of overhead for subtransactions that are only used for error recovery (eg plpgsql exceptions). Also, arrange to release a subtransaction's XID lock as soon as the subtransaction exits, in both the commit and abort cases. This avoids holding many unique locks after a long series of subtransactions. The price is some additional overhead in XactLockTableWait, but that seems acceptable. Finally, restructure the state machine in xact.c to have a more orthogonal set of states for subtransactions.
2004-09-16 18:58:44 +02:00
TransactionId xid = GetCurrentTransactionId();
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
Bitmapset *hot_attrs;
Ignore BRIN indexes when checking for HOT updates When determining whether an index update may be skipped by using HOT, we can ignore attributes indexed by block summarizing indexes without references to individual tuples that need to be cleaned up. A new type TU_UpdateIndexes provides a signal to the executor to determine which indexes to update - no indexes, all indexes, or only the summarizing indexes. This also removes rd_indexattr list, and replaces it with rd_attrsvalid flag. The list was not used anywhere, and a simple flag is sufficient. This was originally committed as 5753d4ee32, but then got reverted by e3fcca0d0d because of correctness issues. Original patch by Josef Simanek, various fixes and improvements by Tomas Vondra and me. Authors: Matthias van de Meent, Josef Simanek, Tomas Vondra Reviewed-by: Tomas Vondra, Alvaro Herrera Discussion: https://postgr.es/m/05ebcb44-f383-86e3-4f31-0a97a55634cf@enterprisedb.com Discussion: https://postgr.es/m/CAFp7QwpMRGcDAQumN7onN9HjrJ3u4X3ZRXdGFT0K5G2JWvnbWg%40mail.gmail.com
2023-03-20 10:34:07 +01:00
Bitmapset *sum_attrs;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Bitmapset *key_attrs;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
Bitmapset *id_attrs;
Simplify check of modified attributes in heap_update The old coding was getting more complicated as new things were added, and it would be barely tolerable with upcoming WARM updates and other future features such as indirect indexes. The new coding incurs a small performance cost in synthetic benchmark cases, and is barely measurable in normal cases. A much larger benefit is expected from WARM, which could actually bolt its needs on top of the existing coding, but it is much uglier and bug-prone than doing it on this new code. Additional optimization can be applied on top of this, if need be. Reviewed-by: Pavan Deolasee, Amit Kapila, Mithun CY Discussion: https://postgr.es/m/20161228232018.4hc66ndrzpz4g4wn@alvherre.pgsql https://postgr.es/m/CABOikdMJfz69dBNRTOZcB6s5A0tf8OMCyQVYQyR-WFFdoEwKMQ@mail.gmail.com
2017-03-29 17:18:48 +02:00
Bitmapset *interesting_attrs;
Bitmapset *modified_attrs;
New HeapTuple structure/interface.
1998-11-27 20:52:36 +01:00
ItemId lp;
HeapTupleData oldtup;
Modify tuptoaster's API so that it does not try to modify the passed tuple in-place, but instead passes back an all-new tuple structure if any changes are needed. This is a much cleaner and more robust solution for the bug discovered by Alexey Beschiokov; accordingly, revert the quick hack I installed yesterday. With this change, HeapTupleData.t_datamcxt is no longer needed; will remove it in a separate commit in HEAD only.
2005-11-20 19:38:20 +01:00
HeapTuple heaptup;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
HeapTuple old_key_tuple = NULL;
bool old_key_copied = false;
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
Page page;
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
BlockNumber block;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
MultiXactStatus mxact_status;
Tweak heap_update/delete so that we do not hold the buffer context lock on the old tuple's page while we are doing TOAST pushups.
2001-01-15 06:29:19 +01:00
Buffer buffer,
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
newbuf,
vmbuffer = InvalidBuffer,
vmbuffer_new = InvalidBuffer;
Fix torn-page, unlogged xid and further risks from heap_update(). When heap_update needs to look for a page for the new tuple version, because the current one doesn't have sufficient free space, or when columns have to be processed by the tuple toaster, it has to release the lock on the old page during that. Otherwise there'd be lock ordering and lock nesting issues. To avoid concurrent sessions from trying to update / delete / lock the tuple while the page's content lock is released, the tuple's xmax is set to the current session's xid. That unfortunately was done without any WAL logging, thereby violating the rule that no XIDs may appear on disk, without an according WAL record. If the database were to crash / fail over when the page level lock is released, and some activity lead to the page being written out to disk, the xid could end up being reused; potentially leading to the row becoming invisible. There might be additional risks by not having t_ctid point at the tuple itself, without having set the appropriate lock infomask fields. To fix, compute the appropriate xmax/infomask combination for locking the tuple, and perform WAL logging using the existing XLOG_HEAP_LOCK record. That allows the fix to be backpatched. This issue has existed for a long time. There appears to have been partial attempts at preventing dangers, but these never have fully been implemented, and were removed a long time ago, in 11919160 (cf. HEAP_XMAX_UNLOGGED). In master / 9.6, there's an additional issue, namely that the visibilitymap's freeze bit isn't reset at that point yet. Since that's a new issue, introduced only in a892234f830, that'll be fixed in a separate commit. Author: Masahiko Sawada and Andres Freund Reported-By: Different aspects by Thomas Munro, Noah Misch, and others Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: 9.1/all supported versions
2016-07-16 02:49:48 +02:00
bool need_toast;
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
Size newtupsize,
pagefree;
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
bool have_tuple_lock = false;
Combine cmin and cmax fields of HeapTupleHeaders into a single field, by keeping private state in each backend that has inserted and deleted the same tuple during its current top-level transaction. This is sufficient since there is no need to be able to determine the cmin/cmax from any other transaction. This gets us back down to 23-byte headers, removing a penalty paid in 8.0 to support subtransactions. Patch by Heikki Linnakangas, with minor revisions by moi, following a design hashed out awhile back on the pghackers list.
2007-02-09 04:35:35 +01:00
bool iscombo;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
bool use_hot_update = false;
Ignore BRIN indexes when checking for HOT updates When determining whether an index update may be skipped by using HOT, we can ignore attributes indexed by block summarizing indexes without references to individual tuples that need to be cleaned up. A new type TU_UpdateIndexes provides a signal to the executor to determine which indexes to update - no indexes, all indexes, or only the summarizing indexes. This also removes rd_indexattr list, and replaces it with rd_attrsvalid flag. The list was not used anywhere, and a simple flag is sufficient. This was originally committed as 5753d4ee32, but then got reverted by e3fcca0d0d because of correctness issues. Original patch by Josef Simanek, various fixes and improvements by Tomas Vondra and me. Authors: Matthias van de Meent, Josef Simanek, Tomas Vondra Reviewed-by: Tomas Vondra, Alvaro Herrera Discussion: https://postgr.es/m/05ebcb44-f383-86e3-4f31-0a97a55634cf@enterprisedb.com Discussion: https://postgr.es/m/CAFp7QwpMRGcDAQumN7onN9HjrJ3u4X3ZRXdGFT0K5G2JWvnbWg%40mail.gmail.com
2023-03-20 10:34:07 +01:00
bool summarized_update = false;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
bool key_intact;
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
bool all_visible_cleared = false;
bool all_visible_cleared_new = false;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
bool checked_lockers;
bool locker_remains;
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
bool id_has_external = false;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
TransactionId xmax_new_tuple,
xmax_old_tuple;
uint16 infomask_old_tuple,
infomask2_old_tuple,
infomask_new_tuple,
infomask2_new_tuple;
pgindent run before 6.3 release, with Thomas' requested changes.
1998-02-26 05:46:47 +01:00
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
Assert(ItemPointerIsValid(otid));
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Fix mishandling of resjunk columns in ON CONFLICT ... UPDATE tlists. It's unusual to have any resjunk columns in an ON CONFLICT ... UPDATE list, but it can happen when MULTIEXPR_SUBLINK SubPlans are present. If it happens, the ON CONFLICT UPDATE code path would end up storing tuples that include the values of the extra resjunk columns. That's fairly harmless in the short run, but if new columns are added to the table then the values would become accessible, possibly leading to malfunctions if they don't match the datatypes of the new columns. This had escaped notice through a confluence of missing sanity checks, including * There's no cross-check that a tuple presented to heap_insert or heap_update matches the table rowtype. While it's difficult to check that fully at reasonable cost, we can easily add assertions that there aren't too many columns. * The output-column-assignment cases in execExprInterp.c lacked any sanity checks on the output column numbers, which seems like an oversight considering there are plenty of assertion checks on input column numbers. Add assertions there too. * We failed to apply nodeModifyTable's ExecCheckPlanOutput() to the ON CONFLICT UPDATE tlist. That wouldn't have caught this specific error, since that function is chartered to ignore resjunk columns; but it sure seems like a bad omission now that we've seen this bug. In HEAD, the right way to fix this is to make the processing of ON CONFLICT UPDATE tlists work the same as regular UPDATE tlists now do, that is don't add "SET x = x" entries, and use ExecBuildUpdateProjection to evaluate the tlist and combine it with old values of the not-set columns. This adds a little complication to ExecBuildUpdateProjection, but allows removal of a comparable amount of now-dead code from the planner. In the back branches, the most expedient solution seems to be to (a) use an output slot for the ON CONFLICT UPDATE projection that actually matches the target table, and then (b) invent a variant of ExecBuildProjectionInfo that can be told to not store values resulting from resjunk columns, so it doesn't try to store into nonexistent columns of the output slot. (We can't simply ignore the resjunk columns altogether; they have to be evaluated for MULTIEXPR_SUBLINK to work.) This works back to v10. In 9.6, projections work much differently and we can't cheaply give them such an option. The 9.6 version of this patch works by inserting a JunkFilter when it's necessary to get rid of resjunk columns. In addition, v11 and up have the reverse problem when trying to perform ON CONFLICT UPDATE on a partitioned table. Through a further oversight, adjust_partition_tlist() discarded resjunk columns when re-ordering the ON CONFLICT UPDATE tlist to match a partition. This accidentally prevented the storing-bogus-tuples problem, but at the cost that MULTIEXPR_SUBLINK cases didn't work, typically crashing if more than one row has to be updated. Fix by preserving resjunk columns in that routine. (I failed to resist the temptation to add more assertions there too, and to do some minor code beautification.) Per report from Andres Freund. Back-patch to all supported branches. Security: CVE-2021-32028
2021-05-10 17:02:29 +02:00
/* Cheap, simplistic check that the tuple matches the rel's rowtype. */
Assert(HeapTupleHeaderGetNatts(newtup->t_data) <=
RelationGetNumberOfAttributes(relation));
Create an infrastructure for parallel computation in PostgreSQL. This does four basic things. First, it provides convenience routines to coordinate the startup and shutdown of parallel workers. Second, it synchronizes various pieces of state (e.g. GUCs, combo CID mappings, transaction snapshot) from the parallel group leader to the worker processes. Third, it prohibits various operations that would result in unsafe changes to that state while parallelism is active. Finally, it propagates events that would result in an ErrorResponse, NoticeResponse, or NotifyResponse message being sent to the client from the parallel workers back to the master, from which they can then be sent on to the client. Robert Haas, Amit Kapila, Noah Misch, Rushabh Lathia, Jeevan Chalke. Suggestions and review from Andres Freund, Heikki Linnakangas, Noah Misch, Simon Riggs, Euler Taveira, and Jim Nasby.
2015-04-30 21:02:14 +02:00
/*
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
* Forbid this during a parallel operation, lest it allocate a combo CID.
* Other workers might need that combo CID for visibility checks, and we
Create an infrastructure for parallel computation in PostgreSQL. This does four basic things. First, it provides convenience routines to coordinate the startup and shutdown of parallel workers. Second, it synchronizes various pieces of state (e.g. GUCs, combo CID mappings, transaction snapshot) from the parallel group leader to the worker processes. Third, it prohibits various operations that would result in unsafe changes to that state while parallelism is active. Finally, it propagates events that would result in an ErrorResponse, NoticeResponse, or NotifyResponse message being sent to the client from the parallel workers back to the master, from which they can then be sent on to the client. Robert Haas, Amit Kapila, Noah Misch, Rushabh Lathia, Jeevan Chalke. Suggestions and review from Andres Freund, Heikki Linnakangas, Noah Misch, Simon Riggs, Euler Taveira, and Jim Nasby.
2015-04-30 21:02:14 +02:00
* have no provision for broadcasting it to them.
*/
if (IsInParallelMode())
ereport(ERROR,
(errcode(ERRCODE_INVALID_TRANSACTION_STATE),
errmsg("cannot update tuples during a parallel operation")));
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
/*
Simplify check of modified attributes in heap_update The old coding was getting more complicated as new things were added, and it would be barely tolerable with upcoming WARM updates and other future features such as indirect indexes. The new coding incurs a small performance cost in synthetic benchmark cases, and is barely measurable in normal cases. A much larger benefit is expected from WARM, which could actually bolt its needs on top of the existing coding, but it is much uglier and bug-prone than doing it on this new code. Additional optimization can be applied on top of this, if need be. Reviewed-by: Pavan Deolasee, Amit Kapila, Mithun CY Discussion: https://postgr.es/m/20161228232018.4hc66ndrzpz4g4wn@alvherre.pgsql https://postgr.es/m/CABOikdMJfz69dBNRTOZcB6s5A0tf8OMCyQVYQyR-WFFdoEwKMQ@mail.gmail.com
2017-03-29 17:18:48 +02:00
* Fetch the list of attributes to be checked for various operations.
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
*
Simplify check of modified attributes in heap_update The old coding was getting more complicated as new things were added, and it would be barely tolerable with upcoming WARM updates and other future features such as indirect indexes. The new coding incurs a small performance cost in synthetic benchmark cases, and is barely measurable in normal cases. A much larger benefit is expected from WARM, which could actually bolt its needs on top of the existing coding, but it is much uglier and bug-prone than doing it on this new code. Additional optimization can be applied on top of this, if need be. Reviewed-by: Pavan Deolasee, Amit Kapila, Mithun CY Discussion: https://postgr.es/m/20161228232018.4hc66ndrzpz4g4wn@alvherre.pgsql https://postgr.es/m/CABOikdMJfz69dBNRTOZcB6s5A0tf8OMCyQVYQyR-WFFdoEwKMQ@mail.gmail.com
2017-03-29 17:18:48 +02:00
* For HOT considerations, this is wasted effort if we fail to update or
* have to put the new tuple on a different page. But we must compute the
* list before obtaining buffer lock --- in the worst case, if we are
* doing an update on one of the relevant system catalogs, we could
* deadlock if we try to fetch the list later. In any case, the relcache
* caches the data so this is usually pretty cheap.
*
* We also need columns used by the replica identity and columns that are
* considered the "key" of rows in the table.
*
* Note that we get copies of each bitmap, so we need not worry about
* relcache flush happening midway through.
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
*/
Ignore BRIN indexes when checking for HOT updates When determining whether an index update may be skipped by using HOT, we can ignore attributes indexed by block summarizing indexes without references to individual tuples that need to be cleaned up. A new type TU_UpdateIndexes provides a signal to the executor to determine which indexes to update - no indexes, all indexes, or only the summarizing indexes. This also removes rd_indexattr list, and replaces it with rd_attrsvalid flag. The list was not used anywhere, and a simple flag is sufficient. This was originally committed as 5753d4ee32, but then got reverted by e3fcca0d0d because of correctness issues. Original patch by Josef Simanek, various fixes and improvements by Tomas Vondra and me. Authors: Matthias van de Meent, Josef Simanek, Tomas Vondra Reviewed-by: Tomas Vondra, Alvaro Herrera Discussion: https://postgr.es/m/05ebcb44-f383-86e3-4f31-0a97a55634cf@enterprisedb.com Discussion: https://postgr.es/m/CAFp7QwpMRGcDAQumN7onN9HjrJ3u4X3ZRXdGFT0K5G2JWvnbWg%40mail.gmail.com
2023-03-20 10:34:07 +01:00
hot_attrs = RelationGetIndexAttrBitmap(relation,
INDEX_ATTR_BITMAP_HOT_BLOCKING);
sum_attrs = RelationGetIndexAttrBitmap(relation,
INDEX_ATTR_BITMAP_SUMMARIZED);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
key_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_KEY);
id_attrs = RelationGetIndexAttrBitmap(relation,
INDEX_ATTR_BITMAP_IDENTITY_KEY);
Go back to considering HOT on pages marked full. Commit 2fd8685e7f simplified the checking of modified attributes that takes place within heap_update(). This included a micro-optimization affecting pages marked PD_PAGE_FULL: don't even try to use HOT to save a few cycles on determining HOT safety. The assumption was that it won't work out this time around, since it can't have worked out last time around. Remove the micro-optimization. It could only ever save cycles that are consumed by the vast majority of heap_update() calls, which hardly seems worth the added complexity. It also seems quite possible that there are workloads that will do worse over time by repeated application of the micro-optimization, despite saving some cycles on average, in the short term. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Álvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/CAH2-WznU1L3+DMPr1F7o2eJBT7=3bAJoY6ZkWABAxNt+-afyTA@mail.gmail.com
2021-11-26 19:58:38 +01:00
interesting_attrs = NULL;
interesting_attrs = bms_add_members(interesting_attrs, hot_attrs);
Ignore BRIN indexes when checking for HOT updates When determining whether an index update may be skipped by using HOT, we can ignore attributes indexed by block summarizing indexes without references to individual tuples that need to be cleaned up. A new type TU_UpdateIndexes provides a signal to the executor to determine which indexes to update - no indexes, all indexes, or only the summarizing indexes. This also removes rd_indexattr list, and replaces it with rd_attrsvalid flag. The list was not used anywhere, and a simple flag is sufficient. This was originally committed as 5753d4ee32, but then got reverted by e3fcca0d0d because of correctness issues. Original patch by Josef Simanek, various fixes and improvements by Tomas Vondra and me. Authors: Matthias van de Meent, Josef Simanek, Tomas Vondra Reviewed-by: Tomas Vondra, Alvaro Herrera Discussion: https://postgr.es/m/05ebcb44-f383-86e3-4f31-0a97a55634cf@enterprisedb.com Discussion: https://postgr.es/m/CAFp7QwpMRGcDAQumN7onN9HjrJ3u4X3ZRXdGFT0K5G2JWvnbWg%40mail.gmail.com
2023-03-20 10:34:07 +01:00
interesting_attrs = bms_add_members(interesting_attrs, sum_attrs);
Go back to considering HOT on pages marked full. Commit 2fd8685e7f simplified the checking of modified attributes that takes place within heap_update(). This included a micro-optimization affecting pages marked PD_PAGE_FULL: don't even try to use HOT to save a few cycles on determining HOT safety. The assumption was that it won't work out this time around, since it can't have worked out last time around. Remove the micro-optimization. It could only ever save cycles that are consumed by the vast majority of heap_update() calls, which hardly seems worth the added complexity. It also seems quite possible that there are workloads that will do worse over time by repeated application of the micro-optimization, despite saving some cycles on average, in the short term. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Álvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/CAH2-WznU1L3+DMPr1F7o2eJBT7=3bAJoY6ZkWABAxNt+-afyTA@mail.gmail.com
2021-11-26 19:58:38 +01:00
interesting_attrs = bms_add_members(interesting_attrs, key_attrs);
interesting_attrs = bms_add_members(interesting_attrs, id_attrs);
Finish reverting "recheck_on_update" patch. This reverts commit c203d6cf8 and some follow-on fixes, completing the task begun in commit 5d28c9bd7. If that feature is ever resurrected, the code will look quite a bit different from this, so it seems best to start from a clean slate. The v11 branch is not touched; in that branch, the recheck_on_update storage option remains present, but nonfunctional and undocumented. Discussion: https://postgr.es/m/20190114223409.3tcvejfhlvbucrv5@alap3.anarazel.de
2019-01-15 18:07:10 +01:00
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
block = ItemPointerGetBlockNumber(otid);
buffer = ReadBuffer(relation, block);
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(buffer);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
/*
* Before locking the buffer, pin the visibility map page if it appears to
* be necessary. Since we haven't got the lock yet, someone else might be
* in the middle of changing this, so we'll need to recheck after we have
* the lock.
*/
if (PageIsAllVisible(page))
visibilitymap_pin(relation, block, &vmbuffer);
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
lp = PageGetItemId(page, ItemPointerGetOffsetNumber(otid));
Redefine the lp_flags field of item pointers as having four states, rather than two independent bits (one of which was never used in heap pages anyway, or at least hadn't been in a very long time). This gives us flexibility to add the HOT notions of redirected and dead item pointers without requiring anything so klugy as magic values of lp_off and lp_len. The state values are chosen so that for the states currently in use (pre-HOT) there is no change in the physical representation.
2007-09-13 00:10:26 +02:00
Assert(ItemIdIsNormal(lp));
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Fill tuple before HeapSatisfiesHOTAndKeyUpdate Failing to do this results in almost all updates to system catalogs being non-HOT updates, because the OID column would differ (not having been set for the new tuple), which is an indexed column. While at it, make sure to set the tableoid early in both old and new tuples as well. This isn't of much consequence, since that column is seldom (never?) indexed. Report and patch from Andres Freund.
2013-02-01 14:43:09 +01:00
/*
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
* Fill in enough data in oldtup for HeapDetermineColumnsInfo to work
Fill tuple before HeapSatisfiesHOTAndKeyUpdate Failing to do this results in almost all updates to system catalogs being non-HOT updates, because the OID column would differ (not having been set for the new tuple), which is an indexed column. While at it, make sure to set the tableoid early in both old and new tuples as well. This isn't of much consequence, since that column is seldom (never?) indexed. Report and patch from Andres Freund.
2013-02-01 14:43:09 +01:00
* properly.
*/
oldtup.t_tableOid = RelationGetRelid(relation);
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
oldtup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
New HeapTuple structure/interface.
1998-11-27 20:52:36 +01:00
oldtup.t_len = ItemIdGetLength(lp);
oldtup.t_self = *otid;
pgindent run. Make it all clean.
2001-03-22 05:01:46 +01:00
Fill tuple before HeapSatisfiesHOTAndKeyUpdate Failing to do this results in almost all updates to system catalogs being non-HOT updates, because the OID column would differ (not having been set for the new tuple), which is an indexed column. While at it, make sure to set the tableoid early in both old and new tuples as well. This isn't of much consequence, since that column is seldom (never?) indexed. Report and patch from Andres Freund.
2013-02-01 14:43:09 +01:00
/* the new tuple is ready, except for this: */
newtup->t_tableOid = RelationGetRelid(relation);
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
/*
* Determine columns modified by the update. Additionally, identify
* whether any of the unmodified replica identity key attributes in the
* old tuple is externally stored or not. This is required because for
* such attributes the flattened value won't be WAL logged as part of the
* new tuple so we must include it as part of the old_key_tuple. See
* ExtractReplicaIdentity.
*/
modified_attrs = HeapDetermineColumnsInfo(relation, interesting_attrs,
id_attrs, &oldtup,
newtup, &id_has_external);
Simplify check of modified attributes in heap_update The old coding was getting more complicated as new things were added, and it would be barely tolerable with upcoming WARM updates and other future features such as indirect indexes. The new coding incurs a small performance cost in synthetic benchmark cases, and is barely measurable in normal cases. A much larger benefit is expected from WARM, which could actually bolt its needs on top of the existing coding, but it is much uglier and bug-prone than doing it on this new code. Additional optimization can be applied on top of this, if need be. Reviewed-by: Pavan Deolasee, Amit Kapila, Mithun CY Discussion: https://postgr.es/m/20161228232018.4hc66ndrzpz4g4wn@alvherre.pgsql https://postgr.es/m/CABOikdMJfz69dBNRTOZcB6s5A0tf8OMCyQVYQyR-WFFdoEwKMQ@mail.gmail.com
2017-03-29 17:18:48 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* If we're not updating any "key" column, we can grab a weaker lock type.
* This allows for more concurrency when we are running simultaneously
* with foreign key checks.
*
* Note that if a column gets detoasted while executing the update, but
* the value ends up being the same, this test will fail and we will use
* the stronger lock. This is acceptable; the important case to optimize
* is updates that don't manipulate key columns, not those that
Fix inconsistencies and typos in the tree, take 10 This addresses some issues with unnecessary code comments, fixes various typos in docs and comments, and removes some orphaned structures and definitions. Author: Alexander Lakhin Discussion: https://postgr.es/m/9aabc775-5494-b372-8bcb-4dfc0bd37c68@gmail.com
2019-08-13 06:53:41 +02:00
* serendipitously arrive at the same key values.
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
*/
Simplify check of modified attributes in heap_update The old coding was getting more complicated as new things were added, and it would be barely tolerable with upcoming WARM updates and other future features such as indirect indexes. The new coding incurs a small performance cost in synthetic benchmark cases, and is barely measurable in normal cases. A much larger benefit is expected from WARM, which could actually bolt its needs on top of the existing coding, but it is much uglier and bug-prone than doing it on this new code. Additional optimization can be applied on top of this, if need be. Reviewed-by: Pavan Deolasee, Amit Kapila, Mithun CY Discussion: https://postgr.es/m/20161228232018.4hc66ndrzpz4g4wn@alvherre.pgsql https://postgr.es/m/CABOikdMJfz69dBNRTOZcB6s5A0tf8OMCyQVYQyR-WFFdoEwKMQ@mail.gmail.com
2017-03-29 17:18:48 +02:00
if (!bms_overlap(modified_attrs, key_attrs))
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
Revert MERGE patch This reverts commits d204ef63776b8a00ca220adec23979091564e465, 83454e3c2b28141c0db01c7d2027e01040df5249 and a few more commits thereafter (complete list at the end) related to MERGE feature. While the feature was fully functional, with sufficient test coverage and necessary documentation, it was felt that some parts of the executor and parse-analyzer can use a different design and it wasn't possible to do that in the available time. So it was decided to revert the patch for PG11 and retry again in the future. Thanks again to all reviewers and bug reporters. List of commits reverted, in reverse chronological order: f1464c5380 Improve parse representation for MERGE ddb4158579 MERGE syntax diagram correction 530e69e59b Allow cpluspluscheck to pass by renaming variable 01b88b4df5 MERGE minor errata 3af7b2b0d4 MERGE fix variable warning in non-assert builds a5d86181ec MERGE INSERT allows only one VALUES clause 4b2d44031f MERGE post-commit review 4923550c20 Tab completion for MERGE aa3faa3c7a WITH support in MERGE 83454e3c2b New files for MERGE d204ef6377 MERGE SQL Command following SQL:2016 Author: Pavan Deolasee Reviewed-by: Michael Paquier
2018-04-12 12:22:56 +02:00
*lockmode = LockTupleNoKeyExclusive;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
mxact_status = MultiXactStatusNoKeyUpdate;
key_intact = true;
/*
* If this is the first possibly-multixact-able operation in the
* current transaction, set my per-backend OldestMemberMXactId
* setting. We can be certain that the transaction will never become a
* member of any older MultiXactIds than that. (We have to do this
* even if we end up just using our own TransactionId below, since
* some other backend could incorporate our XID into a MultiXact
* immediately afterwards.)
*/
MultiXactIdSetOldestMember();
}
else
{
Revert MERGE patch This reverts commits d204ef63776b8a00ca220adec23979091564e465, 83454e3c2b28141c0db01c7d2027e01040df5249 and a few more commits thereafter (complete list at the end) related to MERGE feature. While the feature was fully functional, with sufficient test coverage and necessary documentation, it was felt that some parts of the executor and parse-analyzer can use a different design and it wasn't possible to do that in the available time. So it was decided to revert the patch for PG11 and retry again in the future. Thanks again to all reviewers and bug reporters. List of commits reverted, in reverse chronological order: f1464c5380 Improve parse representation for MERGE ddb4158579 MERGE syntax diagram correction 530e69e59b Allow cpluspluscheck to pass by renaming variable 01b88b4df5 MERGE minor errata 3af7b2b0d4 MERGE fix variable warning in non-assert builds a5d86181ec MERGE INSERT allows only one VALUES clause 4b2d44031f MERGE post-commit review 4923550c20 Tab completion for MERGE aa3faa3c7a WITH support in MERGE 83454e3c2b New files for MERGE d204ef6377 MERGE SQL Command following SQL:2016 Author: Pavan Deolasee Reviewed-by: Michael Paquier
2018-04-12 12:22:56 +02:00
*lockmode = LockTupleExclusive;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
mxact_status = MultiXactStatusUpdate;
key_intact = false;
}
Correct nasty error in heap_update: it was releasing the buffer refcount before calling RelationInvalidateHeapTuple(), which is bad because the latter needs to look at the tuple data, which is in the shared disk buffer. If another backend manages to recycle the buffer while this is going on, we will compute the wrong hashindex for the tuple or maybe even crash outright. Must hold buffer refcount until afterwards. (This bug is not in 7.0.*; seems to be have introduced during WAL changes.)
2001-01-07 23:14:31 +01:00
/*
* Note: beyond this point, use oldtup not otid to refer to old tuple.
* otid may very well point at newtup->t_self, which we will overwrite
* with the new tuple's location, so there's great risk of confusion if we
* use otid anymore.
*/
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
l2:
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
checked_lockers = false;
locker_remains = false;
Adjust HeapTupleSatisfies* routines to take a HeapTuple. Previously, these functions took a HeapTupleHeader, but upcoming patches for logical replication will introduce new a new snapshot type under which the tuple's TID will be used to lookup (CMIN, CMAX) for visibility determination purposes. This makes that information available. Code churn is minimal since HeapTupleSatisfiesVisibility took the HeapTuple anyway, and deferenced it before calling the satisfies function. Independently of logical replication, this allows t_tableOid and t_self to be cross-checked via assertions in tqual.c. This seems like a useful way to make sure that all callers are setting these values properly, which has been previously put forward as desirable. Andres Freund, reviewed by Álvaro Herrera
2013-07-22 19:26:33 +02:00
result = HeapTupleSatisfiesUpdate(&oldtup, cid, buffer);
pgindent run over code.
1999-05-25 18:15:34 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/* see below about the "no wait" case */
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
Assert(result != TM_BeingModified || wait);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (result == TM_Invisible)
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
{
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
UnlockReleaseBuffer(buffer);
Fix a number of places that produced XX000 errors in the regression tests. It's against project policy to use elog() for user-facing errors, or to omit an errcode() selection for errors that aren't supposed to be "can't happen" cases. Fix all the violations of this policy that result in ERRCODE_INTERNAL_ERROR log entries during the standard regression tests, as errors that can reliably be triggered from SQL surely should be considered user-facing. I also looked through all the files touched by this commit and fixed other nearby problems of the same ilk. I do not claim to have fixed all violations of the policy, just the ones in these files. In a few places I also changed existing ERRCODE choices that didn't seem particularly appropriate; mainly replacing ERRCODE_SYNTAX_ERROR by something more specific. Back-patch to 9.5, but no further; changing ERRCODE assignments in stable branches doesn't seem like a good idea.
2015-08-03 05:49:19 +02:00
ereport(ERROR,
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
errmsg("attempted to update invisible tuple")));
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
}
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
else if (result == TM_BeingModified && wait)
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
{
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
TransactionId xwait;
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
uint16 infomask;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
bool can_continue = false;
/*
* XXX note that we don't consider the "no wait" case here. This
* isn't a problem currently because no caller uses that case, but it
* should be fixed if such a caller is introduced. It wasn't a
* problem previously because this code would always wait, but now
* that some tuple locks do not conflict with one of the lock modes we
* use, it is possible that this case is interesting to handle
* specially.
*
* This may cause failures with third-party code that calls
* heap_update directly.
*/
Ye-old pgindent run. Same 4-space tabs.
2000-04-12 19:17:23 +02:00
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
/* must copy state data before unlocking buffer */
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
infomask = oldtup.t_data->t_infomask;
1. Fix for elog(ERROR, "EvalPlanQual: t_xmin is uncommitted ?!") and possibly for other cases too: DO NOT cache status of transaction in unknown state (i.e. non-committed and non-aborted ones) Example: T1 reads row updated/inserted by running T2 and cache T2 status. T2 commits. Now T1 reads a row updated by T2 and with HEAP_XMAX_COMMITTED in t_infomask (so cached T2 status is not changed). Now T1 EvalPlanQual gets updated row version without HEAP_XMIN_COMMITTED -> TransactionIdDidCommit(t_xmin) and TransactionIdDidAbort(t_xmin) return FALSE and T2 decides that t_xmin is not committed and gets ERROR above. It's too late to find more smart way to handle such cases and so I just changed xact status caching and got rid TransactionIdFlushCache() from code. Changed: transam.c, xact.c, lmgr.c and transam.h - last three just because of TransactionIdFlushCache() is removed. 2. heapam.c: T1 marked a row for update. T2 waits for T1 commit/abort. T1 commits. T3 updates the row before T2 locks row page. Now T2 sees that new row t_xmax is different from xact id (T1) T2 was waiting for. Old code did Assert here. New one goes to HeapTupleSatisfiesUpdate. Obvious changes too. 3. Added Assert to vacuum.c 4. bufmgr.c: break Assert(buf->r_locks == 0 && !buf->ri_lock) into two Asserts.
1999-06-10 16:17:12 +02:00
/*
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* Now we have to do something about the existing locker. If it's a
* multi, sleep on it; we might be awakened before it is completely
* gone (or even not sleep at all in some cases); we need to preserve
* it as locker, unless it is gone completely.
*
* If it's not a multi, we need to check for sleeping conditions
* before actually going to sleep. If the update doesn't conflict
* with the locks, we just continue without sleeping (but making sure
* it is preserved).
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
*
* Before sleeping, we need to acquire tuple lock to establish our
* priority for the tuple (see heap_lock_tuple). LockTuple will
* release us when we are next-in-line for the tuple. Note we must
* not acquire the tuple lock until we're sure we're going to sleep;
* otherwise we're open for race conditions with other transactions
* holding the tuple lock which sleep on us.
*
* If we are forced to "start over" below, we keep the tuple lock;
* this arranges that we stay at the head of the line while rechecking
* tuple state.
1. Fix for elog(ERROR, "EvalPlanQual: t_xmin is uncommitted ?!") and possibly for other cases too: DO NOT cache status of transaction in unknown state (i.e. non-committed and non-aborted ones) Example: T1 reads row updated/inserted by running T2 and cache T2 status. T2 commits. Now T1 reads a row updated by T2 and with HEAP_XMAX_COMMITTED in t_infomask (so cached T2 status is not changed). Now T1 EvalPlanQual gets updated row version without HEAP_XMIN_COMMITTED -> TransactionIdDidCommit(t_xmin) and TransactionIdDidAbort(t_xmin) return FALSE and T2 decides that t_xmin is not committed and gets ERROR above. It's too late to find more smart way to handle such cases and so I just changed xact status caching and got rid TransactionIdFlushCache() from code. Changed: transam.c, xact.c, lmgr.c and transam.h - last three just because of TransactionIdFlushCache() is removed. 2. heapam.c: T1 marked a row for update. T2 waits for T1 commit/abort. T1 commits. T3 updates the row before T2 locks row page. Now T2 sees that new row t_xmax is different from xact id (T1) T2 was waiting for. Old code did Assert here. New one goes to HeapTupleSatisfiesUpdate. Obvious changes too. 3. Added Assert to vacuum.c 4. bufmgr.c: break Assert(buf->r_locks == 0 && !buf->ri_lock) into two Asserts.
1999-06-10 16:17:12 +02:00
*/
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
if (infomask & HEAP_XMAX_IS_MULTI)
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
{
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
TransactionId update_xact;
int remain;
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
bool current_is_member = false;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
*lockmode, &current_is_member))
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
{
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
/*
* Acquire the lock, if necessary (but skip it when we're
* requesting a lock and already have one; avoids deadlock).
*/
if (!current_is_member)
heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
LockWaitBlock, &have_tuple_lock);
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
/* wait for multixact */
MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
relation, &oldtup.t_self, XLTW_Update,
&remain);
checked_lockers = true;
locker_remains = remain != 0;
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
/*
* If xwait had just locked the tuple then some other xact
* could update this tuple before we get to this point. Check
* for xmax change, and start over if so.
*/
if (xmax_infomask_changed(oldtup.t_data->t_infomask,
infomask) ||
!TransactionIdEquals(HeapTupleHeaderGetRawXmax(oldtup.t_data),
xwait))
goto l2;
}
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
/*
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* Note that the multixact may not be done by now. It could have
* surviving members; our own xact or other subxacts of this
* backend, and also any other concurrent transaction that locked
Fix bogus comment Author: Alexander Lakhin Discussion: https://postgr.es/m/20190819072244.GE18166@paquier.xyz
2019-08-20 22:04:09 +02:00
* the tuple with LockTupleKeyShare if we only got
* LockTupleNoKeyExclusive. If this is the case, we have to be
* careful to mark the updated tuple with the surviving members in
* Xmax.
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
*
* Note that there could have been another update in the
* MultiXact. In that case, we need to check whether it committed
* or aborted. If it aborted we are safe to update it again;
* otherwise there is an update conflict, and we have to return
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
* TableTuple{Deleted, Updated} below.
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
*
* In the LockTupleExclusive case, we still need to preserve the
* surviving members: those would include the tuple locks we had
* before this one, which are important to keep in case this
* subxact aborts.
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
*/
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if (!HEAP_XMAX_IS_LOCKED_ONLY(oldtup.t_data->t_infomask))
update_xact = HeapTupleGetUpdateXid(oldtup.t_data);
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
else
update_xact = InvalidTransactionId;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Don't TransactionIdDidAbort in HeapTupleGetUpdateXid It is dangerous to do so, because some code expects to be able to see what's the true Xmax even if it is aborted (particularly while traversing HOT chains). So don't do it, and instead rely on the callers to verify for abortedness, if necessary. Several race conditions and bugs fixed in the process. One isolation test changes the expected output due to these. This also reverts commit c235a6a589b, which is no longer necessary. Backpatch to 9.3, where this function was introduced. Andres Freund
2013-11-29 20:08:06 +01:00
/*
* There was no UPDATE in the MultiXact; or it aborted. No
* TransactionIdIsInProgress() call needed here, since we called
* MultiXactIdWait() above.
*/
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if (!TransactionIdIsValid(update_xact) ||
TransactionIdDidAbort(update_xact))
can_continue = true;
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
}
else if (TransactionIdIsCurrentTransactionId(xwait))
{
/*
* The only locker is ourselves; we can avoid grabbing the tuple
* lock here, but must preserve our locking information.
*/
checked_lockers = true;
locker_remains = true;
can_continue = true;
}
else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) && key_intact)
{
/*
* If it's just a key-share locker, and we're not changing the key
* columns, we don't need to wait for it to end; but we need to
* preserve it as locker.
*/
checked_lockers = true;
locker_remains = true;
can_continue = true;
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
}
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
else
{
/*
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
* Wait for regular transaction to end; but first, acquire tuple
* lock.
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
*/
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
Revert MERGE patch This reverts commits d204ef63776b8a00ca220adec23979091564e465, 83454e3c2b28141c0db01c7d2027e01040df5249 and a few more commits thereafter (complete list at the end) related to MERGE feature. While the feature was fully functional, with sufficient test coverage and necessary documentation, it was felt that some parts of the executor and parse-analyzer can use a different design and it wasn't possible to do that in the available time. So it was decided to revert the patch for PG11 and retry again in the future. Thanks again to all reviewers and bug reporters. List of commits reverted, in reverse chronological order: f1464c5380 Improve parse representation for MERGE ddb4158579 MERGE syntax diagram correction 530e69e59b Allow cpluspluscheck to pass by renaming variable 01b88b4df5 MERGE minor errata 3af7b2b0d4 MERGE fix variable warning in non-assert builds a5d86181ec MERGE INSERT allows only one VALUES clause 4b2d44031f MERGE post-commit review 4923550c20 Tab completion for MERGE aa3faa3c7a WITH support in MERGE 83454e3c2b New files for MERGE d204ef6377 MERGE SQL Command following SQL:2016 Author: Pavan Deolasee Reviewed-by: Michael Paquier
2018-04-12 12:22:56 +02:00
heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
LockWaitBlock, &have_tuple_lock);
XactLockTableWait(xwait, relation, &oldtup.t_self,
XLTW_Update);
checked_lockers = true;
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
/*
* xwait is done, but if xwait had just locked the tuple then some
* other xact could update this tuple before we get to this point.
* Check for xmax change, and start over if so.
*/
if (xmax_infomask_changed(oldtup.t_data->t_infomask, infomask) ||
!TransactionIdEquals(xwait,
HeapTupleHeaderGetRawXmax(oldtup.t_data)))
goto l2;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
/* Otherwise check if it committed or aborted */
UpdateXmaxHintBits(oldtup.t_data, buffer, xwait);
if (oldtup.t_data->t_infomask & HEAP_XMAX_INVALID)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
can_continue = true;
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
}
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (can_continue)
result = TM_Ok;
Remove pointless HeapTupleHeaderIndicatesMovedPartitions calls Pavan Deolasee recently noted that a few of the HeapTupleHeaderIndicatesMovedPartitions calls added by commit 5db6df0c0117 are useless, since they are done after comparing t_self with t_ctid. But because t_self can never be set to the magical values that indicate that the tuple moved partition, this can never succeed: if the first test fails (so we know t_self equals t_ctid), necessarily the second test will also fail. So these checks can be removed and no harm is done. There's no bug here, just a code legibility issue. Reported-by: Pavan Deolasee <pavan.deolasee@gmail.com> Discussion: https://postgr.es/m/20200929164411.GA15497@alvherre.pgsql
2021-02-22 20:51:44 +01:00
else if (!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid))
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Updated;
else
result = TM_Deleted;
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
}
Repair RI trigger visibility problems (this time for sure ;-)) per recent discussion on pgsql-hackers: in READ COMMITTED mode we just have to force a QuerySnapshot update in the trigger, but in SERIALIZABLE mode we have to run the scan under a current snapshot and then complain if any rows would be updated/deleted that are not visible in the transaction snapshot.
2003-10-01 23:30:53 +02:00
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (crosscheck != InvalidSnapshot && result == TM_Ok)
Repair RI trigger visibility problems (this time for sure ;-)) per recent discussion on pgsql-hackers: in READ COMMITTED mode we just have to force a QuerySnapshot update in the trigger, but in SERIALIZABLE mode we have to run the scan under a current snapshot and then complain if any rows would be updated/deleted that are not visible in the transaction snapshot.
2003-10-01 23:30:53 +02:00
{
SERIALIZABLE transactions are actually implemented beneath the covers with transaction snapshots, i.e. a snapshot registered at the beginning of a transaction. Change variable naming and comments to reflect this reality in preparation for a future, truly serializable mode, e.g. Serializable Snapshot Isolation (SSI). For the moment transaction snapshots are still used to implement SERIALIZABLE, but hopefully not for too much longer. Patch by Kevin Grittner and Dan Ports with review and some minor wording changes by me.
2010-09-11 20:38:58 +02:00
/* Perform additional check for transaction-snapshot mode RI updates */
Clean up the representation of special snapshots by including a "method pointer" in every Snapshot struct. This allows removal of the case-by-case tests in HeapTupleSatisfiesVisibility, which should make it a bit faster (I didn't try any performance tests though). More importantly, we are no longer violating portable C practices by assuming that small integers are distinct from all pointer values, and HeapTupleSatisfiesDirty no longer has a non-reentrant API involving side-effects on a global variable. There were a couple of places calling HeapTupleSatisfiesXXX routines directly rather than through the HeapTupleSatisfiesVisibility macro. Since these places had to be changed anyway, I chose to make them go through the macro for uniformity. Along the way I renamed HeapTupleSatisfiesSnapshot to HeapTupleSatisfiesMVCC to emphasize that it's only used with MVCC-type snapshots. I was sorely tempted to rename HeapTupleSatisfiesVisibility to HeapTupleSatisfiesSnapshot, but forebore for the moment to avoid confusion and reduce the likelihood that this patch breaks some of the pending patches. Might want to reconsider doing that later.
2007-03-25 21:45:14 +02:00
if (!HeapTupleSatisfiesVisibility(&oldtup, crosscheck, buffer))
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
{
result = TM_Updated;
Assert(!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
}
Repair RI trigger visibility problems (this time for sure ;-)) per recent discussion on pgsql-hackers: in READ COMMITTED mode we just have to force a QuerySnapshot update in the trigger, but in SERIALIZABLE mode we have to run the scan under a current snapshot and then complain if any rows would be updated/deleted that are not visible in the transaction snapshot.
2003-10-01 23:30:53 +02:00
}
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (result != TM_Ok)
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
Assert(result == TM_SelfModified ||
result == TM_Updated ||
result == TM_Deleted ||
result == TM_BeingModified);
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
Assert(!(oldtup.t_data->t_infomask & HEAP_XMAX_INVALID));
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
Assert(result != TM_Updated ||
!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
tmfd->ctid = oldtup.t_data->t_ctid;
tmfd->xmax = HeapTupleHeaderGetUpdateXid(oldtup.t_data);
if (result == TM_SelfModified)
tmfd->cmax = HeapTupleHeaderGetCmax(oldtup.t_data);
Throw error if expiring tuple is again updated or deleted. This prevents surprising behavior when a FOR EACH ROW trigger BEFORE UPDATE or BEFORE DELETE directly or indirectly updates or deletes the the old row. Prior to this patch the requested action on the row could be silently ignored while all triggered actions based on the occurence of the requested action could be committed. One example of how this could happen is if the BEFORE DELETE trigger for a "parent" row deleted "children" which had trigger functions to update summary or status data on the parent. This also prevents similar surprising problems if the query has a volatile function which updates a target row while it is already being updated. There are related issues present in FOR UPDATE cursors and READ COMMITTED queries which are not handled by this patch. These issues need further evalution to determine what change, if any, is needed. Where the new error messages are generated, in most cases the best fix will be to move code from the BEFORE trigger to an AFTER trigger. Where this is not feasible, the trigger can avoid the error by re-issuing the triggering statement and returning NULL. Documentation changes will be submitted in a separate patch. Kevin Grittner and Tom Lane with input from Florian Pflug and Robert Haas, based on problems encountered during conversion of Wisconsin Circuit Court trigger logic to plpgsql triggers.
2012-10-26 21:55:36 +02:00
else
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
tmfd->cmax = InvalidCommandId;
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
UnlockReleaseBuffer(buffer);
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
if (have_tuple_lock)
Revert MERGE patch This reverts commits d204ef63776b8a00ca220adec23979091564e465, 83454e3c2b28141c0db01c7d2027e01040df5249 and a few more commits thereafter (complete list at the end) related to MERGE feature. While the feature was fully functional, with sufficient test coverage and necessary documentation, it was felt that some parts of the executor and parse-analyzer can use a different design and it wasn't possible to do that in the available time. So it was decided to revert the patch for PG11 and retry again in the future. Thanks again to all reviewers and bug reporters. List of commits reverted, in reverse chronological order: f1464c5380 Improve parse representation for MERGE ddb4158579 MERGE syntax diagram correction 530e69e59b Allow cpluspluscheck to pass by renaming variable 01b88b4df5 MERGE minor errata 3af7b2b0d4 MERGE fix variable warning in non-assert builds a5d86181ec MERGE INSERT allows only one VALUES clause 4b2d44031f MERGE post-commit review 4923550c20 Tab completion for MERGE aa3faa3c7a WITH support in MERGE 83454e3c2b New files for MERGE d204ef6377 MERGE SQL Command following SQL:2016 Author: Pavan Deolasee Reviewed-by: Michael Paquier
2018-04-12 12:22:56 +02:00
UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
if (vmbuffer != InvalidBuffer)
ReleaseBuffer(vmbuffer);
Ignore BRIN indexes when checking for HOT updates When determining whether an index update may be skipped by using HOT, we can ignore attributes indexed by block summarizing indexes without references to individual tuples that need to be cleaned up. A new type TU_UpdateIndexes provides a signal to the executor to determine which indexes to update - no indexes, all indexes, or only the summarizing indexes. This also removes rd_indexattr list, and replaces it with rd_attrsvalid flag. The list was not used anywhere, and a simple flag is sufficient. This was originally committed as 5753d4ee32, but then got reverted by e3fcca0d0d because of correctness issues. Original patch by Josef Simanek, various fixes and improvements by Tomas Vondra and me. Authors: Matthias van de Meent, Josef Simanek, Tomas Vondra Reviewed-by: Tomas Vondra, Alvaro Herrera Discussion: https://postgr.es/m/05ebcb44-f383-86e3-4f31-0a97a55634cf@enterprisedb.com Discussion: https://postgr.es/m/CAFp7QwpMRGcDAQumN7onN9HjrJ3u4X3ZRXdGFT0K5G2JWvnbWg%40mail.gmail.com
2023-03-20 10:34:07 +01:00
*update_indexes = TU_None;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
bms_free(hot_attrs);
Ignore BRIN indexes when checking for HOT updates When determining whether an index update may be skipped by using HOT, we can ignore attributes indexed by block summarizing indexes without references to individual tuples that need to be cleaned up. A new type TU_UpdateIndexes provides a signal to the executor to determine which indexes to update - no indexes, all indexes, or only the summarizing indexes. This also removes rd_indexattr list, and replaces it with rd_attrsvalid flag. The list was not used anywhere, and a simple flag is sufficient. This was originally committed as 5753d4ee32, but then got reverted by e3fcca0d0d because of correctness issues. Original patch by Josef Simanek, various fixes and improvements by Tomas Vondra and me. Authors: Matthias van de Meent, Josef Simanek, Tomas Vondra Reviewed-by: Tomas Vondra, Alvaro Herrera Discussion: https://postgr.es/m/05ebcb44-f383-86e3-4f31-0a97a55634cf@enterprisedb.com Discussion: https://postgr.es/m/CAFp7QwpMRGcDAQumN7onN9HjrJ3u4X3ZRXdGFT0K5G2JWvnbWg%40mail.gmail.com
2023-03-20 10:34:07 +01:00
bms_free(sum_attrs);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
bms_free(key_attrs);
Fix small query-lifespan memory leak in bulk updates. When there is an identifiable REPLICA IDENTITY index on the target table, heap_update leaks the id_attrs bitmapset. That's not many bytes, but it adds up over enough rows, since the code typically runs in a query-lifespan context. Bug introduced in commit e55704d8b, which did a rather poor job of cloning the existing use-pattern for RelationGetIndexAttrBitmap(). Per bug #14293 from Zhou Digoal. Back-patch to 9.4 where the bug was introduced. Report: <20160824114320.15676.45171@wrigleys.postgresql.org>
2016-08-25 04:20:01 +02:00
bms_free(id_attrs);
Simplify check of modified attributes in heap_update The old coding was getting more complicated as new things were added, and it would be barely tolerable with upcoming WARM updates and other future features such as indirect indexes. The new coding incurs a small performance cost in synthetic benchmark cases, and is barely measurable in normal cases. A much larger benefit is expected from WARM, which could actually bolt its needs on top of the existing coding, but it is much uglier and bug-prone than doing it on this new code. Additional optimization can be applied on top of this, if need be. Reviewed-by: Pavan Deolasee, Amit Kapila, Mithun CY Discussion: https://postgr.es/m/20161228232018.4hc66ndrzpz4g4wn@alvherre.pgsql https://postgr.es/m/CABOikdMJfz69dBNRTOZcB6s5A0tf8OMCyQVYQyR-WFFdoEwKMQ@mail.gmail.com
2017-03-29 17:18:48 +02:00
bms_free(modified_attrs);
bms_free(interesting_attrs);
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
return result;
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
}
Try again to make the visibility map crash safe. My previous attempt was quite a bit less than half-baked with respect to heap_update().
2011-06-27 19:55:55 +02:00
/*
* If we didn't pin the visibility map page and the page has become all
* visible while we were busy locking the buffer, or during some
* subsequent window during which we had it unlocked, we'll have to unlock
* and re-lock, to avoid holding the buffer lock across an I/O. That's a
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* bit unfortunate, especially since we'll now have to recheck whether the
heap_update() must recheck tuple after unlocking and relocking buffer. Bug found by Alvaro Herrera, fix suggested by Heikki Linnakangas and reviewed by Tom Lane.
2011-09-27 14:24:18 +02:00
* tuple has been locked or updated under us, but hopefully it won't
* happen very often.
Try again to make the visibility map crash safe. My previous attempt was quite a bit less than half-baked with respect to heap_update().
2011-06-27 19:55:55 +02:00
*/
if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
{
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
visibilitymap_pin(relation, block, &vmbuffer);
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
heap_update() must recheck tuple after unlocking and relocking buffer. Bug found by Alvaro Herrera, fix suggested by Heikki Linnakangas and reviewed by Tom Lane.
2011-09-27 14:24:18 +02:00
goto l2;
Try again to make the visibility map crash safe. My previous attempt was quite a bit less than half-baked with respect to heap_update().
2011-06-27 19:55:55 +02:00
}
Fill tuple before HeapSatisfiesHOTAndKeyUpdate Failing to do this results in almost all updates to system catalogs being non-HOT updates, because the OID column would differ (not having been set for the new tuple), which is an indexed column. While at it, make sure to set the tableoid early in both old and new tuples as well. This isn't of much consequence, since that column is seldom (never?) indexed. Report and patch from Andres Freund.
2013-02-01 14:43:09 +01:00
/* Fill in transaction status data */
Code review for HeapTupleHeader changes. Add version number to page headers (overlaying low byte of page size) and add HEAP_HASOID bit to t_infomask, per earlier discussion. Simplify scheme for overlaying fields in tuple header (no need for cmax to live in more than one place). Don't try to clear infomask status bits in tqual.c --- not safe to do it there. Don't try to force output table of a SELECT INTO to have OIDs, either. Get rid of unnecessarily complex three-state scheme for TupleDesc.tdhasoids, which has already caused one recent failure. Improve documentation.
2002-09-02 03:05:06 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* If the tuple we're updating is locked, we need to preserve the locking
* info in the old tuple's Xmax. Prepare a new Xmax value for this.
*/
compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
oldtup.t_data->t_infomask,
oldtup.t_data->t_infomask2,
Revert MERGE patch This reverts commits d204ef63776b8a00ca220adec23979091564e465, 83454e3c2b28141c0db01c7d2027e01040df5249 and a few more commits thereafter (complete list at the end) related to MERGE feature. While the feature was fully functional, with sufficient test coverage and necessary documentation, it was felt that some parts of the executor and parse-analyzer can use a different design and it wasn't possible to do that in the available time. So it was decided to revert the patch for PG11 and retry again in the future. Thanks again to all reviewers and bug reporters. List of commits reverted, in reverse chronological order: f1464c5380 Improve parse representation for MERGE ddb4158579 MERGE syntax diagram correction 530e69e59b Allow cpluspluscheck to pass by renaming variable 01b88b4df5 MERGE minor errata 3af7b2b0d4 MERGE fix variable warning in non-assert builds a5d86181ec MERGE INSERT allows only one VALUES clause 4b2d44031f MERGE post-commit review 4923550c20 Tab completion for MERGE aa3faa3c7a WITH support in MERGE 83454e3c2b New files for MERGE d204ef6377 MERGE SQL Command following SQL:2016 Author: Pavan Deolasee Reviewed-by: Michael Paquier
2018-04-12 12:22:56 +02:00
xid, *lockmode, true,
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
&xmax_old_tuple, &infomask_old_tuple,
&infomask2_old_tuple);
Restrict infomask bits to set on multixacts We must only set the bit(s) for the strongest lock held in the tuple; otherwise, a multixact containing members with exclusive lock and key-share lock will behave as though only a share lock is held. This bug was introduced in commit 0ac5ad5134, somewhere along development, when we allowed a singleton FOR SHARE lock to be implemented without a MultiXact by using a multi-bit pattern. I overlooked that GetMultiXactIdHintBits() needed to be tweaked as well. Previously, we could have the bits for FOR KEY SHARE and FOR UPDATE simultaneously set and it wouldn't cause a problem. Per report from digoal@126.com
2013-01-31 23:12:35 +01:00
/*
* And also prepare an Xmax value for the new copy of the tuple. If there
* was no xmax previously, or there was one but all lockers are now gone,
Replace occurrences of InvalidXid with InvalidTransactionId While Xid is a known shortening of TransactionId, InvalidXid is not defined in the code. Fix comments which mistakenly were using the shorter version. Author: Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com> Discussion: https://postgr.es/m/CALj2ACUQzdigML868nV4cojfELPkEzNLNOk7b91Pho4JB90fng@mail.gmail.com
2021-10-04 10:31:01 +02:00
* then use InvalidTransactionId; otherwise, get the xmax from the old
* tuple. (In rare cases that might also be InvalidTransactionId and yet
* not have the HEAP_XMAX_INVALID bit set; that's fine.)
Restrict infomask bits to set on multixacts We must only set the bit(s) for the strongest lock held in the tuple; otherwise, a multixact containing members with exclusive lock and key-share lock will behave as though only a share lock is held. This bug was introduced in commit 0ac5ad5134, somewhere along development, when we allowed a singleton FOR SHARE lock to be implemented without a MultiXact by using a multi-bit pattern. I overlooked that GetMultiXactIdHintBits() needed to be tweaked as well. Previously, we could have the bits for FOR KEY SHARE and FOR UPDATE simultaneously set and it wouldn't cause a problem. Per report from digoal@126.com
2013-01-31 23:12:35 +01:00
*/
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if ((oldtup.t_data->t_infomask & HEAP_XMAX_INVALID) ||
Fix handling of multixacts predating pg_upgrade After pg_upgrade, it is possible that some tuples' Xmax have multixacts corresponding to the old installation; such multixacts cannot have running members anymore. In many code sites we already know not to read them and clobber them silently, but at least when VACUUM tries to freeze a multixact or determine whether one needs freezing, there's an attempt to resolve it to its member transactions by calling GetMultiXactIdMembers, and if the multixact value is "in the future" with regards to the current valid multixact range, an error like this is raised: ERROR: MultiXactId 123 has not been created yet -- apparent wraparound and vacuuming fails. Per discussion with Andrew Gierth, it is completely bogus to try to resolve multixacts coming from before a pg_upgrade, regardless of where they stand with regards to the current valid multixact range. It's possible to get from under this problem by doing SELECT FOR UPDATE of the problem tuples, but if tables are large, this is slow and tedious, so a more thorough solution is desirable. To fix, we realize that multixacts in xmax created in 9.2 and previous have a specific bit pattern that is never used in 9.3 and later (we already knew this, per comments and infomask tests sprinkled in various places, but we weren't leveraging this knowledge appropriately). Whenever the infomask of the tuple matches that bit pattern, we just ignore the multixact completely as if Xmax wasn't set; or, in the case of tuple freezing, we act as if an unwanted value is set and clobber it without decoding. This guarantees that no errors will be raised, and that the values will be progressively removed until all tables are clean. Most callers of GetMultiXactIdMembers are patched to recognize directly that the value is a removable "empty" multixact and avoid calling GetMultiXactIdMembers altogether. To avoid changing the signature of GetMultiXactIdMembers() in back branches, we keep the "allow_old" boolean flag but rename it to "from_pgupgrade"; if the flag is true, we always return an empty set instead of looking up the multixact. (I suppose we could remove the argument in the master branch, but I chose not to do so in this commit). This was broken all along, but the error-facing message appeared first because of commit 8e9a16ab8f7f and was partially fixed in a25c2b7c4db3. This fix, backpatched all the way back to 9.3, goes approximately in the same direction as a25c2b7c4db3 but should cover all cases. Bug analysis by Andrew Gierth and Álvaro Herrera. A number of public reports match this bug: https://www.postgresql.org/message-id/20140330040029.GY4582@tamriel.snowman.net https://www.postgresql.org/message-id/538F3D70.6080902@publicrelay.com https://www.postgresql.org/message-id/556439CF.7070109@pscs.co.uk https://www.postgresql.org/message-id/SG2PR06MB0760098A111C88E31BD4D96FB3540@SG2PR06MB0760.apcprd06.prod.outlook.com https://www.postgresql.org/message-id/20160615203829.5798.4594@wrigleys.postgresql.org
2016-06-25 00:29:28 +02:00
HEAP_LOCKED_UPGRADED(oldtup.t_data->t_infomask) ||
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
(checked_lockers && !locker_remains))
xmax_new_tuple = InvalidTransactionId;
else
xmax_new_tuple = HeapTupleHeaderGetRawXmax(oldtup.t_data);
if (!TransactionIdIsValid(xmax_new_tuple))
{
infomask_new_tuple = HEAP_XMAX_INVALID;
infomask2_new_tuple = 0;
}
else
{
Restrict infomask bits to set on multixacts We must only set the bit(s) for the strongest lock held in the tuple; otherwise, a multixact containing members with exclusive lock and key-share lock will behave as though only a share lock is held. This bug was introduced in commit 0ac5ad5134, somewhere along development, when we allowed a singleton FOR SHARE lock to be implemented without a MultiXact by using a multi-bit pattern. I overlooked that GetMultiXactIdHintBits() needed to be tweaked as well. Previously, we could have the bits for FOR KEY SHARE and FOR UPDATE simultaneously set and it wouldn't cause a problem. Per report from digoal@126.com
2013-01-31 23:12:35 +01:00
/*
* If we found a valid Xmax for the new tuple, then the infomask bits
* to use on the new tuple depend on what was there on the old one.
* Note that since we're doing an update, the only possibility is that
* the lockers had FOR KEY SHARE lock.
*/
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if (oldtup.t_data->t_infomask & HEAP_XMAX_IS_MULTI)
{
GetMultiXactIdHintBits(xmax_new_tuple, &infomask_new_tuple,
&infomask2_new_tuple);
}
else
{
infomask_new_tuple = HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_LOCK_ONLY;
infomask2_new_tuple = 0;
}
}
/*
* Prepare the new tuple with the appropriate initial values of Xmin and
* Xmax, as well as initial infomask bits as computed above.
*/
New HeapTuple structure/interface.
1998-11-27 20:52:36 +01:00
newtup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
newtup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
Restructure subtransaction handling to reduce resource consumption, as per recent discussions. Invent SubTransactionIds that are managed like CommandIds (ie, counter is reset at start of each top transaction), and use these instead of TransactionIds to keep track of subtransaction status in those modules that need it. This means that a subtransaction does not need an XID unless it actually inserts/modifies rows in the database. Accordingly, don't assign it an XID nor take a lock on the XID until it tries to do that. This saves a lot of overhead for subtransactions that are only used for error recovery (eg plpgsql exceptions). Also, arrange to release a subtransaction's XID lock as soon as the subtransaction exits, in both the commit and abort cases. This avoids holding many unique locks after a long series of subtransactions. The price is some additional overhead in XactLockTableWait, but that seems acceptable. Finally, restructure the state machine in xact.c to have a more orthogonal set of states for subtransactions.
2004-09-16 18:58:44 +02:00
HeapTupleHeaderSetXmin(newtup->t_data, xid);
This patch wraps all accesses to t_xmin, t_cmin, t_xmax, and t_cmax in HeapTupleHeaderData in setter and getter macros called HeapTupleHeaderGetXmin, HeapTupleHeaderSetXmin etc. It also introduces a "virtual" field xvac by defining HeapTupleHeaderGetXvac and HeapTupleHeaderSetXvac. Xvac is used by VACUUM, in fact it is stored in t_cmin. Manfred Koizar
2002-06-15 21:54:24 +02:00
HeapTupleHeaderSetCmin(newtup->t_data, cid);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
newtup->t_data->t_infomask |= HEAP_UPDATED | infomask_new_tuple;
newtup->t_data->t_infomask2 |= infomask2_new_tuple;
HeapTupleHeaderSetXmax(newtup->t_data, xmax_new_tuple);
Combine cmin and cmax fields of HeapTupleHeaders into a single field, by keeping private state in each backend that has inserted and deleted the same tuple during its current top-level transaction. This is sufficient since there is no need to be able to determine the cmin/cmax from any other transaction. This gets us back down to 23-byte headers, removing a penalty paid in 8.0 to support subtransactions. Patch by Heikki Linnakangas, with minor revisions by moi, following a design hashed out awhile back on the pghackers list.
2007-02-09 04:35:35 +01:00
/*
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
* Replace cid with a combo CID if necessary. Note that we already put
Combine cmin and cmax fields of HeapTupleHeaders into a single field, by keeping private state in each backend that has inserted and deleted the same tuple during its current top-level transaction. This is sufficient since there is no need to be able to determine the cmin/cmax from any other transaction. This gets us back down to 23-byte headers, removing a penalty paid in 8.0 to support subtransactions. Patch by Heikki Linnakangas, with minor revisions by moi, following a design hashed out awhile back on the pghackers list.
2007-02-09 04:35:35 +01:00
* the plain cid into the new tuple.
*/
HeapTupleHeaderAdjustCmax(oldtup.t_data, &cid, &iscombo);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
Tweak heap_update/delete so that we do not hold the buffer context lock on the old tuple's page while we are doing TOAST pushups.
2001-01-15 06:29:19 +01:00
/*
* If the toaster needs to be activated, OR if the new tuple will not fit
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
* on the same page as the old, then we need to release the content lock
Tweak heap_update/delete so that we do not hold the buffer context lock on the old tuple's page while we are doing TOAST pushups.
2001-01-15 06:29:19 +01:00
* (but not the pin!) on the old tuple's buffer while we are off doing
* TOAST and/or table-file-extension work. We must mark the old tuple to
Fix torn-page, unlogged xid and further risks from heap_update(). When heap_update needs to look for a page for the new tuple version, because the current one doesn't have sufficient free space, or when columns have to be processed by the tuple toaster, it has to release the lock on the old page during that. Otherwise there'd be lock ordering and lock nesting issues. To avoid concurrent sessions from trying to update / delete / lock the tuple while the page's content lock is released, the tuple's xmax is set to the current session's xid. That unfortunately was done without any WAL logging, thereby violating the rule that no XIDs may appear on disk, without an according WAL record. If the database were to crash / fail over when the page level lock is released, and some activity lead to the page being written out to disk, the xid could end up being reused; potentially leading to the row becoming invisible. There might be additional risks by not having t_ctid point at the tuple itself, without having set the appropriate lock infomask fields. To fix, compute the appropriate xmax/infomask combination for locking the tuple, and perform WAL logging using the existing XLOG_HEAP_LOCK record. That allows the fix to be backpatched. This issue has existed for a long time. There appears to have been partial attempts at preventing dangers, but these never have fully been implemented, and were removed a long time ago, in 11919160 (cf. HEAP_XMAX_UNLOGGED). In master / 9.6, there's an additional issue, namely that the visibilitymap's freeze bit isn't reset at that point yet. Since that's a new issue, introduced only in a892234f830, that'll be fixed in a separate commit. Author: Masahiko Sawada and Andres Freund Reported-By: Different aspects by Thomas Munro, Noah Misch, and others Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: 9.1/all supported versions
2016-07-16 02:49:48 +02:00
* show that it's locked, else other processes may try to update it
* themselves.
Tweak heap_update/delete so that we do not hold the buffer context lock on the old tuple's page while we are doing TOAST pushups.
2001-01-15 06:29:19 +01:00
*
Tighten short-circuit tests for deciding whether we need to invoke tuptoaster.c --- fields that are compressed in-line are not a reason to invoke the toaster. Along the way, add a couple more htup.h macros to eliminate confusing negated tests, and get rid of the already vestigial TUPLE_TOASTER_ACTIVE symbol.
2004-01-16 21:51:30 +01:00
* We need to invoke the toaster if there are already any out-of-line
* toasted values present, or if the new tuple is over-threshold.
Moving toaster out of NO ELOG area in heap_update(). Jan
2000-07-04 19:11:40 +02:00
*/
Add a materialized view relations. A materialized view has a rule just like a view and a heap and other physical properties like a table. The rule is only used to populate the table, references in queries refer to the materialized data. This is a minimal implementation, but should still be useful in many cases. Currently data is only populated "on demand" by the CREATE MATERIALIZED VIEW and REFRESH MATERIALIZED VIEW statements. It is expected that future releases will add incremental updates with various timings, and that a more refined concept of defining what is "fresh" data will be developed. At some point it may even be possible to have queries use a materialized in place of references to underlying tables, but that requires the other above-mentioned features to be working first. Much of the documentation work by Robert Haas. Review by Noah Misch, Thom Brown, Robert Haas, Marko Tiikkaja Security review by KaiGai Kohei, with a decision on how best to implement sepgsql still pending.
2013-03-04 01:23:31 +01:00
if (relation->rd_rel->relkind != RELKIND_RELATION &&
relation->rd_rel->relkind != RELKIND_MATVIEW)
Decouple the values of TOAST_TUPLE_THRESHOLD and TOAST_MAX_CHUNK_SIZE. Add the latter to the values checked in pg_control, since it can't be changed without invalidating toast table content. This commit in itself shouldn't change any behavior, but it lays some necessary groundwork for experimentation with these toast-control numbers. Note: while TOAST_TUPLE_THRESHOLD can now be changed without initdb, some thought still needs to be given to needs_toast_table() in toasting.c before unleashing random changes.
2007-04-03 06:14:26 +02:00
{
/* toast table entries should never be recursively toasted */
Assert(!HeapTupleHasExternal(&oldtup));
Assert(!HeapTupleHasExternal(newtup));
need_toast = false;
}
else
need_toast = (HeapTupleHasExternal(&oldtup) ||
HeapTupleHasExternal(newtup) ||
newtup->t_len > TOAST_TUPLE_THRESHOLD);
Moving toaster out of NO ELOG area in heap_update(). Jan
2000-07-04 19:11:40 +02:00
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
pagefree = PageGetHeapFreeSpace(page);
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
Don't MAXALIGN in the checks to decide whether a tuple is over TOAST's threshold for tuple length. On 4-byte-MAXALIGN machines, the toast code creates tuples that have t_len exactly TOAST_TUPLE_THRESHOLD ... but this number is not itself maxaligned, so if heap_insert maxaligns t_len before comparing to TOAST_TUPLE_THRESHOLD, it'll uselessly recurse back to tuptoaster.c, wasting cycles. (It turns out that this does not happen on 8-byte-MAXALIGN machines, because for them the outer MAXALIGN in the TOAST_MAX_CHUNK_SIZE macro reduces TOAST_MAX_CHUNK_SIZE so that toast tuples will be less than TOAST_TUPLE_THRESHOLD in size. That MAXALIGN is really incorrect, but we can't remove it now, see below.) There isn't any particular value in maxaligning before comparing to the thresholds, so just don't do that, which saves a small number of cycles in itself. These numbers should be rejiggered to minimize wasted space on toast-relation pages, but we can't do that in the back branches because changing TOAST_MAX_CHUNK_SIZE would force an initdb (by changing the contents of toast tables). We can move the toast decision thresholds a bit, though, which is what this patch effectively does. Thanks to Pavan Deolasee for discovering the unintended recursion. Back-patch into 8.2, but not further, pending more testing. (HEAD is about to get a further patch modifying the thresholds, so it won't help much for testing this form of the patch.)
2007-02-04 21:00:37 +01:00
newtupsize = MAXALIGN(newtup->t_len);
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
if (need_toast || newtupsize > pagefree)
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
{
Fix torn-page, unlogged xid and further risks from heap_update(). When heap_update needs to look for a page for the new tuple version, because the current one doesn't have sufficient free space, or when columns have to be processed by the tuple toaster, it has to release the lock on the old page during that. Otherwise there'd be lock ordering and lock nesting issues. To avoid concurrent sessions from trying to update / delete / lock the tuple while the page's content lock is released, the tuple's xmax is set to the current session's xid. That unfortunately was done without any WAL logging, thereby violating the rule that no XIDs may appear on disk, without an according WAL record. If the database were to crash / fail over when the page level lock is released, and some activity lead to the page being written out to disk, the xid could end up being reused; potentially leading to the row becoming invisible. There might be additional risks by not having t_ctid point at the tuple itself, without having set the appropriate lock infomask fields. To fix, compute the appropriate xmax/infomask combination for locking the tuple, and perform WAL logging using the existing XLOG_HEAP_LOCK record. That allows the fix to be backpatched. This issue has existed for a long time. There appears to have been partial attempts at preventing dangers, but these never have fully been implemented, and were removed a long time ago, in 11919160 (cf. HEAP_XMAX_UNLOGGED). In master / 9.6, there's an additional issue, namely that the visibilitymap's freeze bit isn't reset at that point yet. Since that's a new issue, introduced only in a892234f830, that'll be fixed in a separate commit. Author: Masahiko Sawada and Andres Freund Reported-By: Different aspects by Thomas Munro, Noah Misch, and others Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: 9.1/all supported versions
2016-07-16 02:49:48 +02:00
TransactionId xmax_lock_old_tuple;
uint16 infomask_lock_old_tuple,
infomask2_lock_old_tuple;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
bool cleared_all_frozen = false;
Fix torn-page, unlogged xid and further risks from heap_update(). When heap_update needs to look for a page for the new tuple version, because the current one doesn't have sufficient free space, or when columns have to be processed by the tuple toaster, it has to release the lock on the old page during that. Otherwise there'd be lock ordering and lock nesting issues. To avoid concurrent sessions from trying to update / delete / lock the tuple while the page's content lock is released, the tuple's xmax is set to the current session's xid. That unfortunately was done without any WAL logging, thereby violating the rule that no XIDs may appear on disk, without an according WAL record. If the database were to crash / fail over when the page level lock is released, and some activity lead to the page being written out to disk, the xid could end up being reused; potentially leading to the row becoming invisible. There might be additional risks by not having t_ctid point at the tuple itself, without having set the appropriate lock infomask fields. To fix, compute the appropriate xmax/infomask combination for locking the tuple, and perform WAL logging using the existing XLOG_HEAP_LOCK record. That allows the fix to be backpatched. This issue has existed for a long time. There appears to have been partial attempts at preventing dangers, but these never have fully been implemented, and were removed a long time ago, in 11919160 (cf. HEAP_XMAX_UNLOGGED). In master / 9.6, there's an additional issue, namely that the visibilitymap's freeze bit isn't reset at that point yet. Since that's a new issue, introduced only in a892234f830, that'll be fixed in a separate commit. Author: Masahiko Sawada and Andres Freund Reported-By: Different aspects by Thomas Munro, Noah Misch, and others Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: 9.1/all supported versions
2016-07-16 02:49:48 +02:00
/*
* To prevent concurrent sessions from updating the tuple, we have to
Reword some unclear comments
2017-08-09 00:46:16 +02:00
* temporarily mark it locked, while we release the page-level lock.
Fix torn-page, unlogged xid and further risks from heap_update(). When heap_update needs to look for a page for the new tuple version, because the current one doesn't have sufficient free space, or when columns have to be processed by the tuple toaster, it has to release the lock on the old page during that. Otherwise there'd be lock ordering and lock nesting issues. To avoid concurrent sessions from trying to update / delete / lock the tuple while the page's content lock is released, the tuple's xmax is set to the current session's xid. That unfortunately was done without any WAL logging, thereby violating the rule that no XIDs may appear on disk, without an according WAL record. If the database were to crash / fail over when the page level lock is released, and some activity lead to the page being written out to disk, the xid could end up being reused; potentially leading to the row becoming invisible. There might be additional risks by not having t_ctid point at the tuple itself, without having set the appropriate lock infomask fields. To fix, compute the appropriate xmax/infomask combination for locking the tuple, and perform WAL logging using the existing XLOG_HEAP_LOCK record. That allows the fix to be backpatched. This issue has existed for a long time. There appears to have been partial attempts at preventing dangers, but these never have fully been implemented, and were removed a long time ago, in 11919160 (cf. HEAP_XMAX_UNLOGGED). In master / 9.6, there's an additional issue, namely that the visibilitymap's freeze bit isn't reset at that point yet. Since that's a new issue, introduced only in a892234f830, that'll be fixed in a separate commit. Author: Masahiko Sawada and Andres Freund Reported-By: Different aspects by Thomas Munro, Noah Misch, and others Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: 9.1/all supported versions
2016-07-16 02:49:48 +02:00
*
* To satisfy the rule that any xid potentially appearing in a buffer
* written out to disk, we unfortunately have to WAL log this
* temporary modification. We can reuse xl_heap_lock for this
* purpose. If we crash/error before following through with the
* actual update, xmax will be of an aborted transaction, allowing
* other sessions to proceed.
*/
/*
* Compute xmax / infomask appropriate for locking the tuple. This has
Reword some unclear comments
2017-08-09 00:46:16 +02:00
* to be done separately from the combo that's going to be used for
* updating, because the potentially created multixact would otherwise
* be wrong.
Fix torn-page, unlogged xid and further risks from heap_update(). When heap_update needs to look for a page for the new tuple version, because the current one doesn't have sufficient free space, or when columns have to be processed by the tuple toaster, it has to release the lock on the old page during that. Otherwise there'd be lock ordering and lock nesting issues. To avoid concurrent sessions from trying to update / delete / lock the tuple while the page's content lock is released, the tuple's xmax is set to the current session's xid. That unfortunately was done without any WAL logging, thereby violating the rule that no XIDs may appear on disk, without an according WAL record. If the database were to crash / fail over when the page level lock is released, and some activity lead to the page being written out to disk, the xid could end up being reused; potentially leading to the row becoming invisible. There might be additional risks by not having t_ctid point at the tuple itself, without having set the appropriate lock infomask fields. To fix, compute the appropriate xmax/infomask combination for locking the tuple, and perform WAL logging using the existing XLOG_HEAP_LOCK record. That allows the fix to be backpatched. This issue has existed for a long time. There appears to have been partial attempts at preventing dangers, but these never have fully been implemented, and were removed a long time ago, in 11919160 (cf. HEAP_XMAX_UNLOGGED). In master / 9.6, there's an additional issue, namely that the visibilitymap's freeze bit isn't reset at that point yet. Since that's a new issue, introduced only in a892234f830, that'll be fixed in a separate commit. Author: Masahiko Sawada and Andres Freund Reported-By: Different aspects by Thomas Munro, Noah Misch, and others Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: 9.1/all supported versions
2016-07-16 02:49:48 +02:00
*/
compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
oldtup.t_data->t_infomask,
oldtup.t_data->t_infomask2,
Revert MERGE patch This reverts commits d204ef63776b8a00ca220adec23979091564e465, 83454e3c2b28141c0db01c7d2027e01040df5249 and a few more commits thereafter (complete list at the end) related to MERGE feature. While the feature was fully functional, with sufficient test coverage and necessary documentation, it was felt that some parts of the executor and parse-analyzer can use a different design and it wasn't possible to do that in the available time. So it was decided to revert the patch for PG11 and retry again in the future. Thanks again to all reviewers and bug reporters. List of commits reverted, in reverse chronological order: f1464c5380 Improve parse representation for MERGE ddb4158579 MERGE syntax diagram correction 530e69e59b Allow cpluspluscheck to pass by renaming variable 01b88b4df5 MERGE minor errata 3af7b2b0d4 MERGE fix variable warning in non-assert builds a5d86181ec MERGE INSERT allows only one VALUES clause 4b2d44031f MERGE post-commit review 4923550c20 Tab completion for MERGE aa3faa3c7a WITH support in MERGE 83454e3c2b New files for MERGE d204ef6377 MERGE SQL Command following SQL:2016 Author: Pavan Deolasee Reviewed-by: Michael Paquier
2018-04-12 12:22:56 +02:00
xid, *lockmode, false,
Fix torn-page, unlogged xid and further risks from heap_update(). When heap_update needs to look for a page for the new tuple version, because the current one doesn't have sufficient free space, or when columns have to be processed by the tuple toaster, it has to release the lock on the old page during that. Otherwise there'd be lock ordering and lock nesting issues. To avoid concurrent sessions from trying to update / delete / lock the tuple while the page's content lock is released, the tuple's xmax is set to the current session's xid. That unfortunately was done without any WAL logging, thereby violating the rule that no XIDs may appear on disk, without an according WAL record. If the database were to crash / fail over when the page level lock is released, and some activity lead to the page being written out to disk, the xid could end up being reused; potentially leading to the row becoming invisible. There might be additional risks by not having t_ctid point at the tuple itself, without having set the appropriate lock infomask fields. To fix, compute the appropriate xmax/infomask combination for locking the tuple, and perform WAL logging using the existing XLOG_HEAP_LOCK record. That allows the fix to be backpatched. This issue has existed for a long time. There appears to have been partial attempts at preventing dangers, but these never have fully been implemented, and were removed a long time ago, in 11919160 (cf. HEAP_XMAX_UNLOGGED). In master / 9.6, there's an additional issue, namely that the visibilitymap's freeze bit isn't reset at that point yet. Since that's a new issue, introduced only in a892234f830, that'll be fixed in a separate commit. Author: Masahiko Sawada and Andres Freund Reported-By: Different aspects by Thomas Munro, Noah Misch, and others Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: 9.1/all supported versions
2016-07-16 02:49:48 +02:00
&xmax_lock_old_tuple, &infomask_lock_old_tuple,
&infomask2_lock_old_tuple);
Assert(HEAP_XMAX_IS_LOCKED_ONLY(infomask_lock_old_tuple));
START_CRIT_SECTION();
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
/* Clear obsolete visibility flags ... */
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
HeapTupleClearHotUpdated(&oldtup);
/* ... and store info about transaction updating this tuple */
Fix torn-page, unlogged xid and further risks from heap_update(). When heap_update needs to look for a page for the new tuple version, because the current one doesn't have sufficient free space, or when columns have to be processed by the tuple toaster, it has to release the lock on the old page during that. Otherwise there'd be lock ordering and lock nesting issues. To avoid concurrent sessions from trying to update / delete / lock the tuple while the page's content lock is released, the tuple's xmax is set to the current session's xid. That unfortunately was done without any WAL logging, thereby violating the rule that no XIDs may appear on disk, without an according WAL record. If the database were to crash / fail over when the page level lock is released, and some activity lead to the page being written out to disk, the xid could end up being reused; potentially leading to the row becoming invisible. There might be additional risks by not having t_ctid point at the tuple itself, without having set the appropriate lock infomask fields. To fix, compute the appropriate xmax/infomask combination for locking the tuple, and perform WAL logging using the existing XLOG_HEAP_LOCK record. That allows the fix to be backpatched. This issue has existed for a long time. There appears to have been partial attempts at preventing dangers, but these never have fully been implemented, and were removed a long time ago, in 11919160 (cf. HEAP_XMAX_UNLOGGED). In master / 9.6, there's an additional issue, namely that the visibilitymap's freeze bit isn't reset at that point yet. Since that's a new issue, introduced only in a892234f830, that'll be fixed in a separate commit. Author: Masahiko Sawada and Andres Freund Reported-By: Different aspects by Thomas Munro, Noah Misch, and others Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: 9.1/all supported versions
2016-07-16 02:49:48 +02:00
Assert(TransactionIdIsValid(xmax_lock_old_tuple));
HeapTupleHeaderSetXmax(oldtup.t_data, xmax_lock_old_tuple);
oldtup.t_data->t_infomask |= infomask_lock_old_tuple;
oldtup.t_data->t_infomask2 |= infomask2_lock_old_tuple;
Combine cmin and cmax fields of HeapTupleHeaders into a single field, by keeping private state in each backend that has inserted and deleted the same tuple during its current top-level transaction. This is sufficient since there is no need to be able to determine the cmin/cmax from any other transaction. This gets us back down to 23-byte headers, removing a penalty paid in 8.0 to support subtransactions. Patch by Heikki Linnakangas, with minor revisions by moi, following a design hashed out awhile back on the pghackers list.
2007-02-09 04:35:35 +01:00
HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
Fix torn-page, unlogged xid and further risks from heap_update(). When heap_update needs to look for a page for the new tuple version, because the current one doesn't have sufficient free space, or when columns have to be processed by the tuple toaster, it has to release the lock on the old page during that. Otherwise there'd be lock ordering and lock nesting issues. To avoid concurrent sessions from trying to update / delete / lock the tuple while the page's content lock is released, the tuple's xmax is set to the current session's xid. That unfortunately was done without any WAL logging, thereby violating the rule that no XIDs may appear on disk, without an according WAL record. If the database were to crash / fail over when the page level lock is released, and some activity lead to the page being written out to disk, the xid could end up being reused; potentially leading to the row becoming invisible. There might be additional risks by not having t_ctid point at the tuple itself, without having set the appropriate lock infomask fields. To fix, compute the appropriate xmax/infomask combination for locking the tuple, and perform WAL logging using the existing XLOG_HEAP_LOCK record. That allows the fix to be backpatched. This issue has existed for a long time. There appears to have been partial attempts at preventing dangers, but these never have fully been implemented, and were removed a long time ago, in 11919160 (cf. HEAP_XMAX_UNLOGGED). In master / 9.6, there's an additional issue, namely that the visibilitymap's freeze bit isn't reset at that point yet. Since that's a new issue, introduced only in a892234f830, that'll be fixed in a separate commit. Author: Masahiko Sawada and Andres Freund Reported-By: Different aspects by Thomas Munro, Noah Misch, and others Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: 9.1/all supported versions
2016-07-16 02:49:48 +02:00
/* temporarily make it look not-updated, but locked */
Modify tuptoaster's API so that it does not try to modify the passed tuple in-place, but instead passes back an all-new tuple structure if any changes are needed. This is a much cleaner and more robust solution for the bug discovered by Alexey Beschiokov; accordingly, revert the quick hack I installed yesterday. With this change, HeapTupleData.t_datamcxt is no longer needed; will remove it in a separate commit in HEAD only.
2005-11-20 19:38:20 +01:00
oldtup.t_data->t_ctid = oldtup.t_self;
Fix torn-page, unlogged xid and further risks from heap_update(). When heap_update needs to look for a page for the new tuple version, because the current one doesn't have sufficient free space, or when columns have to be processed by the tuple toaster, it has to release the lock on the old page during that. Otherwise there'd be lock ordering and lock nesting issues. To avoid concurrent sessions from trying to update / delete / lock the tuple while the page's content lock is released, the tuple's xmax is set to the current session's xid. That unfortunately was done without any WAL logging, thereby violating the rule that no XIDs may appear on disk, without an according WAL record. If the database were to crash / fail over when the page level lock is released, and some activity lead to the page being written out to disk, the xid could end up being reused; potentially leading to the row becoming invisible. There might be additional risks by not having t_ctid point at the tuple itself, without having set the appropriate lock infomask fields. To fix, compute the appropriate xmax/infomask combination for locking the tuple, and perform WAL logging using the existing XLOG_HEAP_LOCK record. That allows the fix to be backpatched. This issue has existed for a long time. There appears to have been partial attempts at preventing dangers, but these never have fully been implemented, and were removed a long time ago, in 11919160 (cf. HEAP_XMAX_UNLOGGED). In master / 9.6, there's an additional issue, namely that the visibilitymap's freeze bit isn't reset at that point yet. Since that's a new issue, introduced only in a892234f830, that'll be fixed in a separate commit. Author: Masahiko Sawada and Andres Freund Reported-By: Different aspects by Thomas Munro, Noah Misch, and others Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: 9.1/all supported versions
2016-07-16 02:49:48 +02:00
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
/*
* Clear all-frozen bit on visibility map if needed. We could
* immediately reset ALL_VISIBLE, but given that the WAL logging
* overhead would be unchanged, that doesn't seem necessarily
* worthwhile.
*/
Avoid improbable PANIC during heap_update. heap_update needs to clear any existing "all visible" flag on the old tuple's page (and on the new page too, if different). Per coding rules, to do this it must acquire pin on the appropriate visibility-map page while not holding exclusive buffer lock; which creates a race condition since someone else could set the flag whenever we're not holding the buffer lock. The code is supposed to handle that by re-checking the flag after acquiring buffer lock and retrying if it became set. However, one code path through heap_update itself, as well as one in its subroutine RelationGetBufferForTuple, failed to do this. The end result, in the unlikely event that a concurrent VACUUM did set the flag while we're transiently not holding lock, is a non-recurring "PANIC: wrong buffer passed to visibilitymap_clear" failure. This has been seen a few times in the buildfarm since recent VACUUM changes that added code paths that could set the all-visible flag while holding only exclusive buffer lock. Previously, the flag was (usually?) set only after doing LockBufferForCleanup, which would insist on buffer pin count zero, thus preventing the flag from becoming set partway through heap_update. However, it's clear that it's heap_update not VACUUM that's at fault here. What's less clear is whether there is any hazard from these bugs in released branches. heap_update is certainly violating API expectations, but if there is no code path that can set all-visible without a cleanup lock then it's only a latent bug. That's not 100% certain though, besides which we should worry about extensions or future back-patch fixes that could introduce such code paths. I chose to back-patch to v12. Fixing RelationGetBufferForTuple before that would require also back-patching portions of older fixes (notably 0d1fe9f74), which is more code churn than seems prudent to fix a hypothetical issue. Discussion: https://postgr.es/m/2247102.1618008027@sss.pgh.pa.us
2021-04-13 18:17:24 +02:00
if (PageIsAllVisible(page) &&
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
visibilitymap_clear(relation, block, vmbuffer,
VISIBILITYMAP_ALL_FROZEN))
cleared_all_frozen = true;
Fix torn-page, unlogged xid and further risks from heap_update(). When heap_update needs to look for a page for the new tuple version, because the current one doesn't have sufficient free space, or when columns have to be processed by the tuple toaster, it has to release the lock on the old page during that. Otherwise there'd be lock ordering and lock nesting issues. To avoid concurrent sessions from trying to update / delete / lock the tuple while the page's content lock is released, the tuple's xmax is set to the current session's xid. That unfortunately was done without any WAL logging, thereby violating the rule that no XIDs may appear on disk, without an according WAL record. If the database were to crash / fail over when the page level lock is released, and some activity lead to the page being written out to disk, the xid could end up being reused; potentially leading to the row becoming invisible. There might be additional risks by not having t_ctid point at the tuple itself, without having set the appropriate lock infomask fields. To fix, compute the appropriate xmax/infomask combination for locking the tuple, and perform WAL logging using the existing XLOG_HEAP_LOCK record. That allows the fix to be backpatched. This issue has existed for a long time. There appears to have been partial attempts at preventing dangers, but these never have fully been implemented, and were removed a long time ago, in 11919160 (cf. HEAP_XMAX_UNLOGGED). In master / 9.6, there's an additional issue, namely that the visibilitymap's freeze bit isn't reset at that point yet. Since that's a new issue, introduced only in a892234f830, that'll be fixed in a separate commit. Author: Masahiko Sawada and Andres Freund Reported-By: Different aspects by Thomas Munro, Noah Misch, and others Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: 9.1/all supported versions
2016-07-16 02:49:48 +02:00
MarkBufferDirty(buffer);
if (RelationNeedsWAL(relation))
{
xl_heap_lock xlrec;
XLogRecPtr recptr;
XLogBeginInsert();
XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
xlrec.offnum = ItemPointerGetOffsetNumber(&oldtup.t_self);
Fix xl_heap_lock WAL record field's data type. Make xl_heap_lock's infobits_set field of type uint8, not int8. Using int8 isn't appropriate given that the field just holds status bits. This fixes an oversight in commit 0ac5ad5134. In passing rename the nearby TransactionId field to "xmax" to make things consistency with related records, such as xl_heap_lock_updated. Deliberately avoid a bump in XLOG_PAGE_MAGIC. No backpatch, either. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkCd3kOS8b7Rfxw7Mh1_6jvX=Nzo-CWR1VBTiOtVZkWHA@mail.gmail.com
2023-04-11 23:07:54 +02:00
xlrec.xmax = xmax_lock_old_tuple;
Fix torn-page, unlogged xid and further risks from heap_update(). When heap_update needs to look for a page for the new tuple version, because the current one doesn't have sufficient free space, or when columns have to be processed by the tuple toaster, it has to release the lock on the old page during that. Otherwise there'd be lock ordering and lock nesting issues. To avoid concurrent sessions from trying to update / delete / lock the tuple while the page's content lock is released, the tuple's xmax is set to the current session's xid. That unfortunately was done without any WAL logging, thereby violating the rule that no XIDs may appear on disk, without an according WAL record. If the database were to crash / fail over when the page level lock is released, and some activity lead to the page being written out to disk, the xid could end up being reused; potentially leading to the row becoming invisible. There might be additional risks by not having t_ctid point at the tuple itself, without having set the appropriate lock infomask fields. To fix, compute the appropriate xmax/infomask combination for locking the tuple, and perform WAL logging using the existing XLOG_HEAP_LOCK record. That allows the fix to be backpatched. This issue has existed for a long time. There appears to have been partial attempts at preventing dangers, but these never have fully been implemented, and were removed a long time ago, in 11919160 (cf. HEAP_XMAX_UNLOGGED). In master / 9.6, there's an additional issue, namely that the visibilitymap's freeze bit isn't reset at that point yet. Since that's a new issue, introduced only in a892234f830, that'll be fixed in a separate commit. Author: Masahiko Sawada and Andres Freund Reported-By: Different aspects by Thomas Munro, Noah Misch, and others Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: 9.1/all supported versions
2016-07-16 02:49:48 +02:00
xlrec.infobits_set = compute_infobits(oldtup.t_data->t_infomask,
oldtup.t_data->t_infomask2);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
xlrec.flags =
cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
Fix torn-page, unlogged xid and further risks from heap_update(). When heap_update needs to look for a page for the new tuple version, because the current one doesn't have sufficient free space, or when columns have to be processed by the tuple toaster, it has to release the lock on the old page during that. Otherwise there'd be lock ordering and lock nesting issues. To avoid concurrent sessions from trying to update / delete / lock the tuple while the page's content lock is released, the tuple's xmax is set to the current session's xid. That unfortunately was done without any WAL logging, thereby violating the rule that no XIDs may appear on disk, without an according WAL record. If the database were to crash / fail over when the page level lock is released, and some activity lead to the page being written out to disk, the xid could end up being reused; potentially leading to the row becoming invisible. There might be additional risks by not having t_ctid point at the tuple itself, without having set the appropriate lock infomask fields. To fix, compute the appropriate xmax/infomask combination for locking the tuple, and perform WAL logging using the existing XLOG_HEAP_LOCK record. That allows the fix to be backpatched. This issue has existed for a long time. There appears to have been partial attempts at preventing dangers, but these never have fully been implemented, and were removed a long time ago, in 11919160 (cf. HEAP_XMAX_UNLOGGED). In master / 9.6, there's an additional issue, namely that the visibilitymap's freeze bit isn't reset at that point yet. Since that's a new issue, introduced only in a892234f830, that'll be fixed in a separate commit. Author: Masahiko Sawada and Andres Freund Reported-By: Different aspects by Thomas Munro, Noah Misch, and others Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: 9.1/all supported versions
2016-07-16 02:49:48 +02:00
XLogRegisterData((char *) &xlrec, SizeOfHeapLock);
recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
PageSetLSN(page, recptr);
}
END_CRIT_SECTION();
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
Tweak heap_update/delete so that we do not hold the buffer context lock on the old tuple's page while we are doing TOAST pushups.
2001-01-15 06:29:19 +01:00
Modify tuptoaster's API so that it does not try to modify the passed tuple in-place, but instead passes back an all-new tuple structure if any changes are needed. This is a much cleaner and more robust solution for the bug discovered by Alexey Beschiokov; accordingly, revert the quick hack I installed yesterday. With this change, HeapTupleData.t_datamcxt is no longer needed; will remove it in a separate commit in HEAD only.
2005-11-20 19:38:20 +01:00
/*
* Let the toaster do its thing, if needed.
*
* Note: below this point, heaptup is the data we actually intend to
* store into the relation; newtup is the caller's original untoasted
Don't MAXALIGN in the checks to decide whether a tuple is over TOAST's threshold for tuple length. On 4-byte-MAXALIGN machines, the toast code creates tuples that have t_len exactly TOAST_TUPLE_THRESHOLD ... but this number is not itself maxaligned, so if heap_insert maxaligns t_len before comparing to TOAST_TUPLE_THRESHOLD, it'll uselessly recurse back to tuptoaster.c, wasting cycles. (It turns out that this does not happen on 8-byte-MAXALIGN machines, because for them the outer MAXALIGN in the TOAST_MAX_CHUNK_SIZE macro reduces TOAST_MAX_CHUNK_SIZE so that toast tuples will be less than TOAST_TUPLE_THRESHOLD in size. That MAXALIGN is really incorrect, but we can't remove it now, see below.) There isn't any particular value in maxaligning before comparing to the thresholds, so just don't do that, which saves a small number of cycles in itself. These numbers should be rejiggered to minimize wasted space on toast-relation pages, but we can't do that in the back branches because changing TOAST_MAX_CHUNK_SIZE would force an initdb (by changing the contents of toast tables). We can move the toast decision thresholds a bit, though, which is what this patch effectively does. Thanks to Pavan Deolasee for discovering the unintended recursion. Back-patch into 8.2, but not further, pending more testing. (HEAD is about to get a further patch modifying the thresholds, so it won't help much for testing this form of the patch.)
2007-02-04 21:00:37 +01:00
* data.
Modify tuptoaster's API so that it does not try to modify the passed tuple in-place, but instead passes back an all-new tuple structure if any changes are needed. This is a much cleaner and more robust solution for the bug discovered by Alexey Beschiokov; accordingly, revert the quick hack I installed yesterday. With this change, HeapTupleData.t_datamcxt is no longer needed; will remove it in a separate commit in HEAD only.
2005-11-20 19:38:20 +01:00
*/
Tweak heap_update/delete so that we do not hold the buffer context lock on the old tuple's page while we are doing TOAST pushups.
2001-01-15 06:29:19 +01:00
if (need_toast)
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
{
Teach CLUSTER to skip writing WAL if not needed (ie, not using archiving) --- Simon. Also, code review and cleanup for the previous COPY-no-WAL patches --- Tom.
2007-03-29 02:15:39 +02:00
/* Note we always use WAL and FSM during updates */
Rename some toasting functions based on whether they are heap-specific. The old names for the attribute-detoasting functions names included the word "heap," which seems outdated now that the heap is only one of potentially many table access methods. On the other hand, toast_insert_or_update and toast_delete are heap-specific, so rename them by adding "heap_" as a prefix. Not all of the work of making the TOAST system fully accessible to AMs other than the heap is done yet, but there seems to be little harm in getting this renaming out of the way now. Commit 8b94dab06617ef80a0901ab103ebd8754427ef5a already divided up the functions among various files partially according to whether it was intended that they should be heap-specific or AM-agnostic, so this is just clarifying the division contemplated by that commit. Patch by me, reviewed and tested by Prabhat Sabu, Thomas Munro, Andres Freund, and Álvaro Herrera. Discussion: http://postgr.es/m/CA+TgmoZv-=2iWM4jcw5ZhJeL18HF96+W1yJeYrnGMYdkFFnEpQ@mail.gmail.com
2019-10-04 20:24:46 +02:00
heaptup = heap_toast_insert_or_update(relation, newtup, &oldtup, 0);
Modify tuptoaster's API so that it does not try to modify the passed tuple in-place, but instead passes back an all-new tuple structure if any changes are needed. This is a much cleaner and more robust solution for the bug discovered by Alexey Beschiokov; accordingly, revert the quick hack I installed yesterday. With this change, HeapTupleData.t_datamcxt is no longer needed; will remove it in a separate commit in HEAD only.
2005-11-20 19:38:20 +01:00
newtupsize = MAXALIGN(heaptup->t_len);
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
}
Modify tuptoaster's API so that it does not try to modify the passed tuple in-place, but instead passes back an all-new tuple structure if any changes are needed. This is a much cleaner and more robust solution for the bug discovered by Alexey Beschiokov; accordingly, revert the quick hack I installed yesterday. With this change, HeapTupleData.t_datamcxt is no longer needed; will remove it in a separate commit in HEAD only.
2005-11-20 19:38:20 +01:00
else
heaptup = newtup;
Tweak heap_update/delete so that we do not hold the buffer context lock on the old tuple's page while we are doing TOAST pushups.
2001-01-15 06:29:19 +01:00
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
/*
* Now, do we need a new page for the tuple, or not? This is a bit
* tricky since someone else could have added tuples to the page while
* we weren't looking. We have to recheck the available space after
* reacquiring the buffer lock. But don't bother to do that if the
* former amount of free space is still not enough; it's unlikely
* there's more free now than before.
*
* What's more, if we need to get a new page, we will need to acquire
* buffer locks on both old and new pages. To avoid deadlock against
* some other backend trying to get the same two locks in the other
* order, we must be consistent about the order we get the locks in.
Further work on connecting the free space map (which is still just a stub) into the rest of the system. Adopt a cleaner approach to preventing deadlock in concurrent heap_updates: allow RelationGetBufferForTuple to select any page of the rel, and put the onus on it to lock both buffers in a consistent order. Remove no-longer-needed isExtend hack from API of ReleaseAndReadBuffer.
2001-06-29 23:08:25 +02:00
* We use the rule "lock the lower-numbered page of the relation
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
* first". To implement this, we must do RelationGetBufferForTuple
Further work on connecting the free space map (which is still just a stub) into the rest of the system. Adopt a cleaner approach to preventing deadlock in concurrent heap_updates: allow RelationGetBufferForTuple to select any page of the rel, and put the onus on it to lock both buffers in a consistent order. Remove no-longer-needed isExtend hack from API of ReleaseAndReadBuffer.
2001-06-29 23:08:25 +02:00
* while not holding the lock on the old page, and we must rely on it
* to get the locks on both pages in the correct order.
Avoid improbable PANIC during heap_update. heap_update needs to clear any existing "all visible" flag on the old tuple's page (and on the new page too, if different). Per coding rules, to do this it must acquire pin on the appropriate visibility-map page while not holding exclusive buffer lock; which creates a race condition since someone else could set the flag whenever we're not holding the buffer lock. The code is supposed to handle that by re-checking the flag after acquiring buffer lock and retrying if it became set. However, one code path through heap_update itself, as well as one in its subroutine RelationGetBufferForTuple, failed to do this. The end result, in the unlikely event that a concurrent VACUUM did set the flag while we're transiently not holding lock, is a non-recurring "PANIC: wrong buffer passed to visibilitymap_clear" failure. This has been seen a few times in the buildfarm since recent VACUUM changes that added code paths that could set the all-visible flag while holding only exclusive buffer lock. Previously, the flag was (usually?) set only after doing LockBufferForCleanup, which would insist on buffer pin count zero, thus preventing the flag from becoming set partway through heap_update. However, it's clear that it's heap_update not VACUUM that's at fault here. What's less clear is whether there is any hazard from these bugs in released branches. heap_update is certainly violating API expectations, but if there is no code path that can set all-visible without a cleanup lock then it's only a latent bug. That's not 100% certain though, besides which we should worry about extensions or future back-patch fixes that could introduce such code paths. I chose to back-patch to v12. Fixing RelationGetBufferForTuple before that would require also back-patching portions of older fixes (notably 0d1fe9f74), which is more code churn than seems prudent to fix a hypothetical issue. Discussion: https://postgr.es/m/2247102.1618008027@sss.pgh.pa.us
2021-04-13 18:17:24 +02:00
*
* Another consideration is that we need visibility map page pin(s) if
* we will have to clear the all-visible flag on either page. If we
* call RelationGetBufferForTuple, we rely on it to acquire any such
* pins; but if we don't, we have to handle that here. Hence we need
* a loop.
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
*/
Avoid improbable PANIC during heap_update. heap_update needs to clear any existing "all visible" flag on the old tuple's page (and on the new page too, if different). Per coding rules, to do this it must acquire pin on the appropriate visibility-map page while not holding exclusive buffer lock; which creates a race condition since someone else could set the flag whenever we're not holding the buffer lock. The code is supposed to handle that by re-checking the flag after acquiring buffer lock and retrying if it became set. However, one code path through heap_update itself, as well as one in its subroutine RelationGetBufferForTuple, failed to do this. The end result, in the unlikely event that a concurrent VACUUM did set the flag while we're transiently not holding lock, is a non-recurring "PANIC: wrong buffer passed to visibilitymap_clear" failure. This has been seen a few times in the buildfarm since recent VACUUM changes that added code paths that could set the all-visible flag while holding only exclusive buffer lock. Previously, the flag was (usually?) set only after doing LockBufferForCleanup, which would insist on buffer pin count zero, thus preventing the flag from becoming set partway through heap_update. However, it's clear that it's heap_update not VACUUM that's at fault here. What's less clear is whether there is any hazard from these bugs in released branches. heap_update is certainly violating API expectations, but if there is no code path that can set all-visible without a cleanup lock then it's only a latent bug. That's not 100% certain though, besides which we should worry about extensions or future back-patch fixes that could introduce such code paths. I chose to back-patch to v12. Fixing RelationGetBufferForTuple before that would require also back-patching portions of older fixes (notably 0d1fe9f74), which is more code churn than seems prudent to fix a hypothetical issue. Discussion: https://postgr.es/m/2247102.1618008027@sss.pgh.pa.us
2021-04-13 18:17:24 +02:00
for (;;)
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
{
Avoid improbable PANIC during heap_update. heap_update needs to clear any existing "all visible" flag on the old tuple's page (and on the new page too, if different). Per coding rules, to do this it must acquire pin on the appropriate visibility-map page while not holding exclusive buffer lock; which creates a race condition since someone else could set the flag whenever we're not holding the buffer lock. The code is supposed to handle that by re-checking the flag after acquiring buffer lock and retrying if it became set. However, one code path through heap_update itself, as well as one in its subroutine RelationGetBufferForTuple, failed to do this. The end result, in the unlikely event that a concurrent VACUUM did set the flag while we're transiently not holding lock, is a non-recurring "PANIC: wrong buffer passed to visibilitymap_clear" failure. This has been seen a few times in the buildfarm since recent VACUUM changes that added code paths that could set the all-visible flag while holding only exclusive buffer lock. Previously, the flag was (usually?) set only after doing LockBufferForCleanup, which would insist on buffer pin count zero, thus preventing the flag from becoming set partway through heap_update. However, it's clear that it's heap_update not VACUUM that's at fault here. What's less clear is whether there is any hazard from these bugs in released branches. heap_update is certainly violating API expectations, but if there is no code path that can set all-visible without a cleanup lock then it's only a latent bug. That's not 100% certain though, besides which we should worry about extensions or future back-patch fixes that could introduce such code paths. I chose to back-patch to v12. Fixing RelationGetBufferForTuple before that would require also back-patching portions of older fixes (notably 0d1fe9f74), which is more code churn than seems prudent to fix a hypothetical issue. Discussion: https://postgr.es/m/2247102.1618008027@sss.pgh.pa.us
2021-04-13 18:17:24 +02:00
if (newtupsize > pagefree)
{
/* It doesn't fit, must use RelationGetBufferForTuple. */
newbuf = RelationGetBufferForTuple(relation, heaptup->t_len,
buffer, 0, NULL,
heapam: Pass number of required pages to RelationGetBufferForTuple() A future commit will use this information to determine how aggressively to extend the relation by. In heap_multi_insert() we know accurately how many pages we need once we need to extend the relation, providing an accurate lower bound for how much to extend. Reviewed-by: Melanie Plageman <melanieplageman@gmail.com> Discussion: https://postgr.es/m/20221029025420.eplyow6k7tgu6he3@awork3.anarazel.de
2023-04-06 23:18:24 +02:00
&vmbuffer_new, &vmbuffer,
0);
Avoid improbable PANIC during heap_update. heap_update needs to clear any existing "all visible" flag on the old tuple's page (and on the new page too, if different). Per coding rules, to do this it must acquire pin on the appropriate visibility-map page while not holding exclusive buffer lock; which creates a race condition since someone else could set the flag whenever we're not holding the buffer lock. The code is supposed to handle that by re-checking the flag after acquiring buffer lock and retrying if it became set. However, one code path through heap_update itself, as well as one in its subroutine RelationGetBufferForTuple, failed to do this. The end result, in the unlikely event that a concurrent VACUUM did set the flag while we're transiently not holding lock, is a non-recurring "PANIC: wrong buffer passed to visibilitymap_clear" failure. This has been seen a few times in the buildfarm since recent VACUUM changes that added code paths that could set the all-visible flag while holding only exclusive buffer lock. Previously, the flag was (usually?) set only after doing LockBufferForCleanup, which would insist on buffer pin count zero, thus preventing the flag from becoming set partway through heap_update. However, it's clear that it's heap_update not VACUUM that's at fault here. What's less clear is whether there is any hazard from these bugs in released branches. heap_update is certainly violating API expectations, but if there is no code path that can set all-visible without a cleanup lock then it's only a latent bug. That's not 100% certain though, besides which we should worry about extensions or future back-patch fixes that could introduce such code paths. I chose to back-patch to v12. Fixing RelationGetBufferForTuple before that would require also back-patching portions of older fixes (notably 0d1fe9f74), which is more code churn than seems prudent to fix a hypothetical issue. Discussion: https://postgr.es/m/2247102.1618008027@sss.pgh.pa.us
2021-04-13 18:17:24 +02:00
/* We're all done. */
break;
}
/* Acquire VM page pin if needed and we don't have it. */
if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
visibilitymap_pin(relation, block, &vmbuffer);
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
/* Re-acquire the lock on the old tuple's page. */
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
/* Re-check using the up-to-date free space */
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
pagefree = PageGetHeapFreeSpace(page);
Avoid improbable PANIC during heap_update. heap_update needs to clear any existing "all visible" flag on the old tuple's page (and on the new page too, if different). Per coding rules, to do this it must acquire pin on the appropriate visibility-map page while not holding exclusive buffer lock; which creates a race condition since someone else could set the flag whenever we're not holding the buffer lock. The code is supposed to handle that by re-checking the flag after acquiring buffer lock and retrying if it became set. However, one code path through heap_update itself, as well as one in its subroutine RelationGetBufferForTuple, failed to do this. The end result, in the unlikely event that a concurrent VACUUM did set the flag while we're transiently not holding lock, is a non-recurring "PANIC: wrong buffer passed to visibilitymap_clear" failure. This has been seen a few times in the buildfarm since recent VACUUM changes that added code paths that could set the all-visible flag while holding only exclusive buffer lock. Previously, the flag was (usually?) set only after doing LockBufferForCleanup, which would insist on buffer pin count zero, thus preventing the flag from becoming set partway through heap_update. However, it's clear that it's heap_update not VACUUM that's at fault here. What's less clear is whether there is any hazard from these bugs in released branches. heap_update is certainly violating API expectations, but if there is no code path that can set all-visible without a cleanup lock then it's only a latent bug. That's not 100% certain though, besides which we should worry about extensions or future back-patch fixes that could introduce such code paths. I chose to back-patch to v12. Fixing RelationGetBufferForTuple before that would require also back-patching portions of older fixes (notably 0d1fe9f74), which is more code churn than seems prudent to fix a hypothetical issue. Discussion: https://postgr.es/m/2247102.1618008027@sss.pgh.pa.us
2021-04-13 18:17:24 +02:00
if (newtupsize > pagefree ||
(vmbuffer == InvalidBuffer && PageIsAllVisible(page)))
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
{
/*
Avoid improbable PANIC during heap_update. heap_update needs to clear any existing "all visible" flag on the old tuple's page (and on the new page too, if different). Per coding rules, to do this it must acquire pin on the appropriate visibility-map page while not holding exclusive buffer lock; which creates a race condition since someone else could set the flag whenever we're not holding the buffer lock. The code is supposed to handle that by re-checking the flag after acquiring buffer lock and retrying if it became set. However, one code path through heap_update itself, as well as one in its subroutine RelationGetBufferForTuple, failed to do this. The end result, in the unlikely event that a concurrent VACUUM did set the flag while we're transiently not holding lock, is a non-recurring "PANIC: wrong buffer passed to visibilitymap_clear" failure. This has been seen a few times in the buildfarm since recent VACUUM changes that added code paths that could set the all-visible flag while holding only exclusive buffer lock. Previously, the flag was (usually?) set only after doing LockBufferForCleanup, which would insist on buffer pin count zero, thus preventing the flag from becoming set partway through heap_update. However, it's clear that it's heap_update not VACUUM that's at fault here. What's less clear is whether there is any hazard from these bugs in released branches. heap_update is certainly violating API expectations, but if there is no code path that can set all-visible without a cleanup lock then it's only a latent bug. That's not 100% certain though, besides which we should worry about extensions or future back-patch fixes that could introduce such code paths. I chose to back-patch to v12. Fixing RelationGetBufferForTuple before that would require also back-patching portions of older fixes (notably 0d1fe9f74), which is more code churn than seems prudent to fix a hypothetical issue. Discussion: https://postgr.es/m/2247102.1618008027@sss.pgh.pa.us
2021-04-13 18:17:24 +02:00
* Rats, it doesn't fit anymore, or somebody just now set the
* all-visible flag. We must now unlock and loop to avoid
* deadlock. Fortunately, this path should seldom be taken.
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
*/
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
}
else
{
Avoid improbable PANIC during heap_update. heap_update needs to clear any existing "all visible" flag on the old tuple's page (and on the new page too, if different). Per coding rules, to do this it must acquire pin on the appropriate visibility-map page while not holding exclusive buffer lock; which creates a race condition since someone else could set the flag whenever we're not holding the buffer lock. The code is supposed to handle that by re-checking the flag after acquiring buffer lock and retrying if it became set. However, one code path through heap_update itself, as well as one in its subroutine RelationGetBufferForTuple, failed to do this. The end result, in the unlikely event that a concurrent VACUUM did set the flag while we're transiently not holding lock, is a non-recurring "PANIC: wrong buffer passed to visibilitymap_clear" failure. This has been seen a few times in the buildfarm since recent VACUUM changes that added code paths that could set the all-visible flag while holding only exclusive buffer lock. Previously, the flag was (usually?) set only after doing LockBufferForCleanup, which would insist on buffer pin count zero, thus preventing the flag from becoming set partway through heap_update. However, it's clear that it's heap_update not VACUUM that's at fault here. What's less clear is whether there is any hazard from these bugs in released branches. heap_update is certainly violating API expectations, but if there is no code path that can set all-visible without a cleanup lock then it's only a latent bug. That's not 100% certain though, besides which we should worry about extensions or future back-patch fixes that could introduce such code paths. I chose to back-patch to v12. Fixing RelationGetBufferForTuple before that would require also back-patching portions of older fixes (notably 0d1fe9f74), which is more code churn than seems prudent to fix a hypothetical issue. Discussion: https://postgr.es/m/2247102.1618008027@sss.pgh.pa.us
2021-04-13 18:17:24 +02:00
/* We're all done. */
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
newbuf = buffer;
Avoid improbable PANIC during heap_update. heap_update needs to clear any existing "all visible" flag on the old tuple's page (and on the new page too, if different). Per coding rules, to do this it must acquire pin on the appropriate visibility-map page while not holding exclusive buffer lock; which creates a race condition since someone else could set the flag whenever we're not holding the buffer lock. The code is supposed to handle that by re-checking the flag after acquiring buffer lock and retrying if it became set. However, one code path through heap_update itself, as well as one in its subroutine RelationGetBufferForTuple, failed to do this. The end result, in the unlikely event that a concurrent VACUUM did set the flag while we're transiently not holding lock, is a non-recurring "PANIC: wrong buffer passed to visibilitymap_clear" failure. This has been seen a few times in the buildfarm since recent VACUUM changes that added code paths that could set the all-visible flag while holding only exclusive buffer lock. Previously, the flag was (usually?) set only after doing LockBufferForCleanup, which would insist on buffer pin count zero, thus preventing the flag from becoming set partway through heap_update. However, it's clear that it's heap_update not VACUUM that's at fault here. What's less clear is whether there is any hazard from these bugs in released branches. heap_update is certainly violating API expectations, but if there is no code path that can set all-visible without a cleanup lock then it's only a latent bug. That's not 100% certain though, besides which we should worry about extensions or future back-patch fixes that could introduce such code paths. I chose to back-patch to v12. Fixing RelationGetBufferForTuple before that would require also back-patching portions of older fixes (notably 0d1fe9f74), which is more code churn than seems prudent to fix a hypothetical issue. Discussion: https://postgr.es/m/2247102.1618008027@sss.pgh.pa.us
2021-04-13 18:17:24 +02:00
break;
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
}
}
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
}
Tweak heap_update/delete so that we do not hold the buffer context lock on the old tuple's page while we are doing TOAST pushups.
2001-01-15 06:29:19 +01:00
else
{
/* No TOAST work needed, and it'll fit on same page */
newbuf = buffer;
Modify tuptoaster's API so that it does not try to modify the passed tuple in-place, but instead passes back an all-new tuple structure if any changes are needed. This is a much cleaner and more robust solution for the bug discovered by Alexey Beschiokov; accordingly, revert the quick hack I installed yesterday. With this change, HeapTupleData.t_datamcxt is no longer needed; will remove it in a separate commit in HEAD only.
2005-11-20 19:38:20 +01:00
heaptup = newtup;
Tweak heap_update/delete so that we do not hold the buffer context lock on the old tuple's page while we are doing TOAST pushups.
2001-01-15 06:29:19 +01:00
}
heap' logging
2000-07-03 04:54:21 +02:00
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
/*
Fix serialization anomalies due to race conditions on INSERT. On insert the CheckForSerializableConflictIn() test was performed before the page(s) which were going to be modified had been locked (with an exclusive buffer content lock). If another process acquired a relation SIReadLock on the heap and scanned to a page on which an insert was going to occur before the page was so locked, a rw-conflict would be missed, which could allow a serialization anomaly to be missed. The window between the check and the page lock was small, so the bug was generally not noticed unless there was high concurrency with multiple processes inserting into the same table. This was reported by Peter Bailis as bug #11732, by Sean Chittenden as bug #13667, and by others. The race condition was eliminated in heap_insert() by moving the check down below the acquisition of the buffer lock, which had been the very next statement. Because of the loop locking and unlocking multiple buffers in heap_multi_insert() a check was added after all inserts were completed. The check before the start of the inserts was left because it might avoid a large amount of work to detect a serialization anomaly before performing the all of the inserts and the related WAL logging. While investigating this bug, other SSI bugs which were even harder to hit in practice were noticed and fixed, an unnecessary check (covered by another check, so redundant) was removed from heap_update(), and comments were improved. Back-patch to all supported branches. Kevin Grittner and Thomas Munro
2015-10-31 20:43:34 +01:00
* We're about to do the actual update -- check for conflict first, to
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
* avoid possibly having to roll back work we've just done.
*
Fix serialization anomalies due to race conditions on INSERT. On insert the CheckForSerializableConflictIn() test was performed before the page(s) which were going to be modified had been locked (with an exclusive buffer content lock). If another process acquired a relation SIReadLock on the heap and scanned to a page on which an insert was going to occur before the page was so locked, a rw-conflict would be missed, which could allow a serialization anomaly to be missed. The window between the check and the page lock was small, so the bug was generally not noticed unless there was high concurrency with multiple processes inserting into the same table. This was reported by Peter Bailis as bug #11732, by Sean Chittenden as bug #13667, and by others. The race condition was eliminated in heap_insert() by moving the check down below the acquisition of the buffer lock, which had been the very next statement. Because of the loop locking and unlocking multiple buffers in heap_multi_insert() a check was added after all inserts were completed. The check before the start of the inserts was left because it might avoid a large amount of work to detect a serialization anomaly before performing the all of the inserts and the related WAL logging. While investigating this bug, other SSI bugs which were even harder to hit in practice were noticed and fixed, an unnecessary check (covered by another check, so redundant) was removed from heap_update(), and comments were improved. Back-patch to all supported branches. Kevin Grittner and Thomas Munro
2015-10-31 20:43:34 +01:00
* This is safe without a recheck as long as there is no possibility of
* another process scanning the pages between this check and the update
* being visible to the scan (i.e., exclusive buffer content lock(s) are
* continuously held from this point until the tuple update is visible).
*
* For the new tuple the only check needed is at the relation level, but
* since both tuples are in the same relation and the check for oldtup
* will include checking the relation level, there is no benefit to a
* separate check for the new tuple.
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
*/
Fix potential SSI hazard in heap_update(). Commit 6f38d4dac38 failed to heed a warning about the stability of the value pointed to by "otid". The caller is allowed to pass in a pointer to newtup->t_self, which will be updated during the execution of the function. Instead, the SSI check should use the value we copy into oldtup.t_self near the top of the function. Not a live bug, because newtup->t_self doesn't really get updated until a bit later, but it was confusing and broke the rule established by the comment. Back-patch to 13. Reported-by: Tom Lane <tgl@sss.pgh.pa.us> Discussion: https://postgr.es/m/2689164.1618160085%40sss.pgh.pa.us
2021-04-13 02:34:25 +02:00
CheckForSerializableConflictIn(relation, &oldtup.t_self,
BufferGetBlockNumber(buffer));
Implement genuine serializable isolation level. Until now, our Serializable mode has in fact been what's called Snapshot Isolation, which allows some anomalies that could not occur in any serialized ordering of the transactions. This patch fixes that using a method called Serializable Snapshot Isolation, based on research papers by Michael J. Cahill (see README-SSI for full references). In Serializable Snapshot Isolation, transactions run like they do in Snapshot Isolation, but a predicate lock manager observes the reads and writes performed and aborts transactions if it detects that an anomaly might occur. This method produces some false positives, ie. it sometimes aborts transactions even though there is no anomaly. To track reads we implement predicate locking, see storage/lmgr/predicate.c. Whenever a tuple is read, a predicate lock is acquired on the tuple. Shared memory is finite, so when a transaction takes many tuple-level locks on a page, the locks are promoted to a single page-level lock, and further to a single relation level lock if necessary. To lock key values with no matching tuple, a sequential scan always takes a relation-level lock, and an index scan acquires a page-level lock that covers the search key, whether or not there are any matching keys at the moment. A predicate lock doesn't conflict with any regular locks or with another predicate locks in the normal sense. They're only used by the predicate lock manager to detect the danger of anomalies. Only serializable transactions participate in predicate locking, so there should be no extra overhead for for other transactions. Predicate locks can't be released at commit, but must be remembered until all the transactions that overlapped with it have completed. That means that we need to remember an unbounded amount of predicate locks, so we apply a lossy but conservative method of tracking locks for committed transactions. If we run short of shared memory, we overflow to a new "pg_serial" SLRU pool. We don't currently allow Serializable transactions in Hot Standby mode. That would be hard, because even read-only transactions can cause anomalies that wouldn't otherwise occur. Serializable isolation mode now means the new fully serializable level. Repeatable Read gives you the old Snapshot Isolation level that we have always had. Kevin Grittner and Dan Ports, reviewed by Jeff Davis, Heikki Linnakangas and Anssi Kääriäinen
2011-02-07 22:46:51 +01:00
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
/*
* At this point newbuf and buffer are both pinned and locked, and newbuf
Further work on connecting the free space map (which is still just a stub) into the rest of the system. Adopt a cleaner approach to preventing deadlock in concurrent heap_updates: allow RelationGetBufferForTuple to select any page of the rel, and put the onus on it to lock both buffers in a consistent order. Remove no-longer-needed isExtend hack from API of ReleaseAndReadBuffer.
2001-06-29 23:08:25 +02:00
* has enough space for the new tuple. If they are the same buffer, only
* one pin is held.
Repair race condition introduced into heap_update() in 7.1 --- PageGetFreeSpace() was being called while not holding the buffer lock, which not only could yield a garbage answer, but even if it's the right answer there might be less space available after we reacquire the buffer lock. Also repair potential deadlock introduced by my recent performance improvement in RelationGetBufferForTuple(): it was possible for two heap_updates to try to lock two buffers in opposite orders. The fix creates a global rule that buffers of a single heap relation should be locked in decreasing block number order. Currently, this only applies to heap_update; VACUUM can get away with ignoring the rule since it holds exclusive lock on the whole relation anyway. However, if we try to implement a VACUUM that can run in parallel with other transactions, VACUUM will also have to obey the lock order rule.
2001-05-17 00:35:12 +02:00
*/
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
if (newbuf == buffer)
{
/*
* Since the new tuple is going into the same page, we might be able
* to do a HOT update. Check if any of the index columns have been
Go back to considering HOT on pages marked full. Commit 2fd8685e7f simplified the checking of modified attributes that takes place within heap_update(). This included a micro-optimization affecting pages marked PD_PAGE_FULL: don't even try to use HOT to save a few cycles on determining HOT safety. The assumption was that it won't work out this time around, since it can't have worked out last time around. Remove the micro-optimization. It could only ever save cycles that are consumed by the vast majority of heap_update() calls, which hardly seems worth the added complexity. It also seems quite possible that there are workloads that will do worse over time by repeated application of the micro-optimization, despite saving some cycles on average, in the short term. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Álvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/CAH2-WznU1L3+DMPr1F7o2eJBT7=3bAJoY6ZkWABAxNt+-afyTA@mail.gmail.com
2021-11-26 19:58:38 +01:00
* changed.
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
*/
Go back to considering HOT on pages marked full. Commit 2fd8685e7f simplified the checking of modified attributes that takes place within heap_update(). This included a micro-optimization affecting pages marked PD_PAGE_FULL: don't even try to use HOT to save a few cycles on determining HOT safety. The assumption was that it won't work out this time around, since it can't have worked out last time around. Remove the micro-optimization. It could only ever save cycles that are consumed by the vast majority of heap_update() calls, which hardly seems worth the added complexity. It also seems quite possible that there are workloads that will do worse over time by repeated application of the micro-optimization, despite saving some cycles on average, in the short term. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Álvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/CAH2-WznU1L3+DMPr1F7o2eJBT7=3bAJoY6ZkWABAxNt+-afyTA@mail.gmail.com
2021-11-26 19:58:38 +01:00
if (!bms_overlap(modified_attrs, hot_attrs))
Ignore BRIN indexes when checking for HOT updates When determining whether an index update may be skipped by using HOT, we can ignore attributes indexed by block summarizing indexes without references to individual tuples that need to be cleaned up. A new type TU_UpdateIndexes provides a signal to the executor to determine which indexes to update - no indexes, all indexes, or only the summarizing indexes. This also removes rd_indexattr list, and replaces it with rd_attrsvalid flag. The list was not used anywhere, and a simple flag is sufficient. This was originally committed as 5753d4ee32, but then got reverted by e3fcca0d0d because of correctness issues. Original patch by Josef Simanek, various fixes and improvements by Tomas Vondra and me. Authors: Matthias van de Meent, Josef Simanek, Tomas Vondra Reviewed-by: Tomas Vondra, Alvaro Herrera Discussion: https://postgr.es/m/05ebcb44-f383-86e3-4f31-0a97a55634cf@enterprisedb.com Discussion: https://postgr.es/m/CAFp7QwpMRGcDAQumN7onN9HjrJ3u4X3ZRXdGFT0K5G2JWvnbWg%40mail.gmail.com
2023-03-20 10:34:07 +01:00
{
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
use_hot_update = true;
Ignore BRIN indexes when checking for HOT updates When determining whether an index update may be skipped by using HOT, we can ignore attributes indexed by block summarizing indexes without references to individual tuples that need to be cleaned up. A new type TU_UpdateIndexes provides a signal to the executor to determine which indexes to update - no indexes, all indexes, or only the summarizing indexes. This also removes rd_indexattr list, and replaces it with rd_attrsvalid flag. The list was not used anywhere, and a simple flag is sufficient. This was originally committed as 5753d4ee32, but then got reverted by e3fcca0d0d because of correctness issues. Original patch by Josef Simanek, various fixes and improvements by Tomas Vondra and me. Authors: Matthias van de Meent, Josef Simanek, Tomas Vondra Reviewed-by: Tomas Vondra, Alvaro Herrera Discussion: https://postgr.es/m/05ebcb44-f383-86e3-4f31-0a97a55634cf@enterprisedb.com Discussion: https://postgr.es/m/CAFp7QwpMRGcDAQumN7onN9HjrJ3u4X3ZRXdGFT0K5G2JWvnbWg%40mail.gmail.com
2023-03-20 10:34:07 +01:00
/*
* If none of the columns that are used in hot-blocking indexes
* were updated, we can apply HOT, but we do still need to check
* if we need to update the summarizing indexes, and update those
* indexes if the columns were updated, or we may fail to detect
* e.g. value bound changes in BRIN minmax indexes.
*/
if (bms_overlap(modified_attrs, sum_attrs))
summarized_update = true;
}
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
}
else
{
/* Set a hint that the old page could use prune/defrag */
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
PageSetFull(page);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
}
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
/*
* Compute replica identity tuple before entering the critical section so
* we don't PANIC upon a memory allocation failure.
* ExtractReplicaIdentity() will return NULL if nothing needs to be
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
* logged. Pass old key required as true only if the replica identity key
* columns are modified or it has external data.
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
*/
Simplify check of modified attributes in heap_update The old coding was getting more complicated as new things were added, and it would be barely tolerable with upcoming WARM updates and other future features such as indirect indexes. The new coding incurs a small performance cost in synthetic benchmark cases, and is barely measurable in normal cases. A much larger benefit is expected from WARM, which could actually bolt its needs on top of the existing coding, but it is much uglier and bug-prone than doing it on this new code. Additional optimization can be applied on top of this, if need be. Reviewed-by: Pavan Deolasee, Amit Kapila, Mithun CY Discussion: https://postgr.es/m/20161228232018.4hc66ndrzpz4g4wn@alvherre.pgsql https://postgr.es/m/CABOikdMJfz69dBNRTOZcB6s5A0tf8OMCyQVYQyR-WFFdoEwKMQ@mail.gmail.com
2017-03-29 17:18:48 +02:00
old_key_tuple = ExtractReplicaIdentity(relation, &oldtup,
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
bms_overlap(modified_attrs, id_attrs) ||
id_has_external,
Simplify check of modified attributes in heap_update The old coding was getting more complicated as new things were added, and it would be barely tolerable with upcoming WARM updates and other future features such as indirect indexes. The new coding incurs a small performance cost in synthetic benchmark cases, and is barely measurable in normal cases. A much larger benefit is expected from WARM, which could actually bolt its needs on top of the existing coding, but it is much uglier and bug-prone than doing it on this new code. Additional optimization can be applied on top of this, if need be. Reviewed-by: Pavan Deolasee, Amit Kapila, Mithun CY Discussion: https://postgr.es/m/20161228232018.4hc66ndrzpz4g4wn@alvherre.pgsql https://postgr.es/m/CABOikdMJfz69dBNRTOZcB6s5A0tf8OMCyQVYQyR-WFFdoEwKMQ@mail.gmail.com
2017-03-29 17:18:48 +02:00
&old_key_copied);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
Error message editing in backend/access.
2003-07-21 22:29:40 +02:00
/* NO EREPORT(ERROR) from here till changes are logged */
Add more critical-section calls: all code sections that hold spinlocks are now critical sections, so as to ensure die() won't interrupt us while we are munging shared-memory data structures. Avoid insecure intermediate states in some code that proc_exit will call, like palloc/pfree. Rename START/END_CRIT_CODE to START/END_CRIT_SECTION, since that seems to be what people tend to call them anyway, and make them be called with () like a function call, in hopes of not confusing pg_indent. I doubt that this is sufficient to make SIGTERM safe anywhere; there's just too much code that could get invoked during proc_exit().
2001-01-12 22:54:01 +01:00
START_CRIT_SECTION();
heap' logging
2000-07-03 04:54:21 +02:00
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
/*
* If this transaction commits, the old tuple will become DEAD sooner or
Improve handling of prune/no-prune decisions by storing a page's oldest unpruned XMAX in its header. At the cost of 4 bytes per page, this keeps us from performing heap_page_prune when there's no chance of pruning anything. Seems to be necessary per Heikki's preliminary performance testing.
2007-09-21 23:25:42 +02:00
* later. Set flag that this page is a candidate for pruning once our xid
* falls below the OldestXmin horizon. If the transaction finally aborts,
* the subsequent page pruning will be a no-op and the hint will be
* cleared.
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
*
* XXX Should we set hint on newbuf as well? If the transaction aborts,
* there would be a prunable tuple in the newbuf; but for now we choose
* not to optimize for aborts. Note that heap_xlog_update must be kept in
Improve handling of prune/no-prune decisions by storing a page's oldest unpruned XMAX in its header. At the cost of 4 bytes per page, this keeps us from performing heap_page_prune when there's no chance of pruning anything. Seems to be necessary per Heikki's preliminary performance testing.
2007-09-21 23:25:42 +02:00
* sync if this decision changes.
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
*/
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
PageSetPrunable(page, xid);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
if (use_hot_update)
{
/* Mark the old tuple as HOT-updated */
HeapTupleSetHotUpdated(&oldtup);
/* And mark the new tuple as heap-only */
HeapTupleSetHeapOnly(heaptup);
/* Mark the caller's copy too, in case different from heaptup */
HeapTupleSetHeapOnly(newtup);
}
else
{
/* Make sure tuples are correctly marked as not-HOT */
HeapTupleClearHotUpdated(&oldtup);
HeapTupleClearHeapOnly(heaptup);
HeapTupleClearHeapOnly(newtup);
}
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
RelationPutHeapTuple(relation, newbuf, heaptup, false); /* insert new tuple */
Tweak heap_update/delete so that we do not hold the buffer context lock on the old tuple's page while we are doing TOAST pushups.
2001-01-15 06:29:19 +01:00
Fix torn-page, unlogged xid and further risks from heap_update(). When heap_update needs to look for a page for the new tuple version, because the current one doesn't have sufficient free space, or when columns have to be processed by the tuple toaster, it has to release the lock on the old page during that. Otherwise there'd be lock ordering and lock nesting issues. To avoid concurrent sessions from trying to update / delete / lock the tuple while the page's content lock is released, the tuple's xmax is set to the current session's xid. That unfortunately was done without any WAL logging, thereby violating the rule that no XIDs may appear on disk, without an according WAL record. If the database were to crash / fail over when the page level lock is released, and some activity lead to the page being written out to disk, the xid could end up being reused; potentially leading to the row becoming invisible. There might be additional risks by not having t_ctid point at the tuple itself, without having set the appropriate lock infomask fields. To fix, compute the appropriate xmax/infomask combination for locking the tuple, and perform WAL logging using the existing XLOG_HEAP_LOCK record. That allows the fix to be backpatched. This issue has existed for a long time. There appears to have been partial attempts at preventing dangers, but these never have fully been implemented, and were removed a long time ago, in 11919160 (cf. HEAP_XMAX_UNLOGGED). In master / 9.6, there's an additional issue, namely that the visibilitymap's freeze bit isn't reset at that point yet. Since that's a new issue, introduced only in a892234f830, that'll be fixed in a separate commit. Author: Masahiko Sawada and Andres Freund Reported-By: Different aspects by Thomas Munro, Noah Misch, and others Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: 9.1/all supported versions
2016-07-16 02:49:48 +02:00
/* Clear obsolete visibility flags, possibly set by ourselves above... */
oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
/* ... and store info about transaction updating this tuple */
Assert(TransactionIdIsValid(xmax_old_tuple));
HeapTupleHeaderSetXmax(oldtup.t_data, xmax_old_tuple);
oldtup.t_data->t_infomask |= infomask_old_tuple;
oldtup.t_data->t_infomask2 |= infomask2_old_tuple;
HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
heap' logging
2000-07-03 04:54:21 +02:00
/* record address of new tuple in t_ctid of old one */
Modify tuptoaster's API so that it does not try to modify the passed tuple in-place, but instead passes back an all-new tuple structure if any changes are needed. This is a much cleaner and more robust solution for the bug discovered by Alexey Beschiokov; accordingly, revert the quick hack I installed yesterday. With this change, HeapTupleData.t_datamcxt is no longer needed; will remove it in a separate commit in HEAD only.
2005-11-20 19:38:20 +01:00
oldtup.t_data->t_ctid = heaptup->t_self;
heap' logging
2000-07-03 04:54:21 +02:00
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
/* clear PD_ALL_VISIBLE flags, reset all visibilitymap bits */
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
if (PageIsAllVisible(BufferGetPage(buffer)))
Fix a violation of WAL coding rules in the recent patch to include an "all tuples visible" flag in heap page headers. The flag update *must* be applied before calling XLogInsert, but heap_update and the tuple moving routines in VACUUM FULL were ignoring this rule. A crash and replay could therefore leave the flag incorrectly set, causing rows to appear visible in seqscans when they should not be. This might explain recent reports of data corruption from Jeff Ross and others. In passing, do a bit of editorialization on comments in visibilitymap.c.
2009-08-24 04:18:32 +02:00
{
all_visible_cleared = true;
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
PageClearAllVisible(BufferGetPage(buffer));
Try again to make the visibility map crash safe. My previous attempt was quite a bit less than half-baked with respect to heap_update().
2011-06-27 19:55:55 +02:00
visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
vmbuffer, VISIBILITYMAP_VALID_BITS);
Fix a violation of WAL coding rules in the recent patch to include an "all tuples visible" flag in heap page headers. The flag update *must* be applied before calling XLogInsert, but heap_update and the tuple moving routines in VACUUM FULL were ignoring this rule. A crash and replay could therefore leave the flag incorrectly set, causing rows to appear visible in seqscans when they should not be. This might explain recent reports of data corruption from Jeff Ross and others. In passing, do a bit of editorialization on comments in visibilitymap.c.
2009-08-24 04:18:32 +02:00
}
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
if (newbuf != buffer && PageIsAllVisible(BufferGetPage(newbuf)))
Fix a violation of WAL coding rules in the recent patch to include an "all tuples visible" flag in heap page headers. The flag update *must* be applied before calling XLogInsert, but heap_update and the tuple moving routines in VACUUM FULL were ignoring this rule. A crash and replay could therefore leave the flag incorrectly set, causing rows to appear visible in seqscans when they should not be. This might explain recent reports of data corruption from Jeff Ross and others. In passing, do a bit of editorialization on comments in visibilitymap.c.
2009-08-24 04:18:32 +02:00
{
all_visible_cleared_new = true;
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
PageClearAllVisible(BufferGetPage(newbuf));
Try again to make the visibility map crash safe. My previous attempt was quite a bit less than half-baked with respect to heap_update().
2011-06-27 19:55:55 +02:00
visibilitymap_clear(relation, BufferGetBlockNumber(newbuf),
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
vmbuffer_new, VISIBILITYMAP_VALID_BITS);
Fix a violation of WAL coding rules in the recent patch to include an "all tuples visible" flag in heap page headers. The flag update *must* be applied before calling XLogInsert, but heap_update and the tuple moving routines in VACUUM FULL were ignoring this rule. A crash and replay could therefore leave the flag incorrectly set, causing rows to appear visible in seqscans when they should not be. This might explain recent reports of data corruption from Jeff Ross and others. In passing, do a bit of editorialization on comments in visibilitymap.c.
2009-08-24 04:18:32 +02:00
}
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
if (newbuf != buffer)
MarkBufferDirty(newbuf);
MarkBufferDirty(buffer);
heap' logging
2000-07-03 04:54:21 +02:00
/* XLOG stuff */
Generalize concept of temporary relations to "relation persistence". This commit replaces pg_class.relistemp with pg_class.relpersistence; and also modifies the RangeVar node type to carry relpersistence rather than istemp. It also removes removes rd_istemp from RelationData and instead performs the correct computation based on relpersistence. For clarity, we add three new macros: RelationNeedsWAL(), RelationUsesLocalBuffers(), and RelationUsesTempNamespace(), so that we can clarify the purpose of each check that previous depended on rd_istemp. This is intended as infrastructure for the upcoming unlogged tables patch, as well as for future possible work on global temporary tables.
2010-12-13 18:34:26 +01:00
if (RelationNeedsWAL(relation))
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
{
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
XLogRecPtr recptr;
pgindent run over code.
1999-05-25 18:15:34 +02:00
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
/*
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
* For logical decoding we need combo CIDs to properly decode the
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
* catalog.
*/
if (RelationIsAccessibleInLogicalDecoding(relation))
{
log_heap_new_cid(relation, &oldtup);
log_heap_new_cid(relation, heaptup);
}
recptr = log_heap_update(relation, buffer,
newbuf, &oldtup, heaptup,
old_key_tuple,
all_visible_cleared,
all_visible_cleared_new);
heap' logging
2000-07-03 04:54:21 +02:00
if (newbuf != buffer)
{
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
PageSetLSN(BufferGetPage(newbuf), recptr);
heap' logging
2000-07-03 04:54:21 +02:00
}
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
PageSetLSN(BufferGetPage(buffer), recptr);
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
}
Tweak heap_update/delete so that we do not hold the buffer context lock on the old tuple's page while we are doing TOAST pushups.
2001-01-15 06:29:19 +01:00
Add more critical-section calls: all code sections that hold spinlocks are now critical sections, so as to ensure die() won't interrupt us while we are munging shared-memory data structures. Avoid insecure intermediate states in some code that proc_exit will call, like palloc/pfree. Rename START/END_CRIT_CODE to START/END_CRIT_SECTION, since that seems to be what people tend to call them anyway, and make them be called with () like a function call, in hopes of not confusing pg_indent. I doubt that this is sufficient to make SIGTERM safe anywhere; there's just too much code that could get invoked during proc_exit().
2001-01-12 22:54:01 +01:00
END_CRIT_SECTION();
Massive commit to run PGINDENT on all *.c and *.h files.
1997-09-07 07:04:48 +02:00
heap' logging
2000-07-03 04:54:21 +02:00
if (newbuf != buffer)
LockBuffer(newbuf, BUFFER_LOCK_UNLOCK);
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
Correct nasty error in heap_update: it was releasing the buffer refcount before calling RelationInvalidateHeapTuple(), which is bad because the latter needs to look at the tuple data, which is in the shared disk buffer. If another backend manages to recycle the buffer while this is going on, we will compute the wrong hashindex for the tuple or maybe even crash outright. Must hold buffer refcount until afterwards. (This bug is not in 7.0.*; seems to be have introduced during WAL changes.)
2001-01-07 23:14:31 +01:00
/*
* Mark old tuple for invalidation from system caches at next command
Revise sinval code to remove no-longer-used tuple TID from inval messages. This requires adjusting the API for syscache callback functions: they now get a hash value, not a TID, to identify the target tuple. Most of them weren't paying any attention to that argument anyway, but plancache did require a small amount of fixing. Also, improve performance a trifle by avoiding sending duplicate inval messages when a heap_update isn't changing the catcache lookup columns.
2011-08-17 01:27:46 +02:00
* boundary, and mark the new tuple for invalidation in case we abort. We
* have to do this before releasing the buffer because oldtup is in the
* buffer. (heaptup is all in local memory, but it's necessary to process
* both tuple versions in one call to inval.c so we can avoid redundant
* sinval messages.)
Correct nasty error in heap_update: it was releasing the buffer refcount before calling RelationInvalidateHeapTuple(), which is bad because the latter needs to look at the tuple data, which is in the shared disk buffer. If another backend manages to recycle the buffer while this is going on, we will compute the wrong hashindex for the tuple or maybe even crash outright. Must hold buffer refcount until afterwards. (This bug is not in 7.0.*; seems to be have introduced during WAL changes.)
2001-01-07 23:14:31 +01:00
*/
Revise sinval code to remove no-longer-used tuple TID from inval messages. This requires adjusting the API for syscache callback functions: they now get a hash value, not a TID, to identify the target tuple. Most of them weren't paying any attention to that argument anyway, but plancache did require a small amount of fixing. Also, improve performance a trifle by avoiding sending duplicate inval messages when a heap_update isn't changing the catcache lookup columns.
2011-08-17 01:27:46 +02:00
CacheInvalidateHeapTuple(relation, &oldtup, heaptup);
Correct nasty error in heap_update: it was releasing the buffer refcount before calling RelationInvalidateHeapTuple(), which is bad because the latter needs to look at the tuple data, which is in the shared disk buffer. If another backend manages to recycle the buffer while this is going on, we will compute the wrong hashindex for the tuple or maybe even crash outright. Must hold buffer refcount until afterwards. (This bug is not in 7.0.*; seems to be have introduced during WAL changes.)
2001-01-07 23:14:31 +01:00
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
/* Now we can release the buffer(s) */
Correct nasty error in heap_update: it was releasing the buffer refcount before calling RelationInvalidateHeapTuple(), which is bad because the latter needs to look at the tuple data, which is in the shared disk buffer. If another backend manages to recycle the buffer while this is going on, we will compute the wrong hashindex for the tuple or maybe even crash outright. Must hold buffer refcount until afterwards. (This bug is not in 7.0.*; seems to be have introduced during WAL changes.)
2001-01-07 23:14:31 +01:00
if (newbuf != buffer)
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
ReleaseBuffer(newbuf);
ReleaseBuffer(buffer);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
if (BufferIsValid(vmbuffer_new))
ReleaseBuffer(vmbuffer_new);
if (BufferIsValid(vmbuffer))
ReleaseBuffer(vmbuffer);
Correct nasty error in heap_update: it was releasing the buffer refcount before calling RelationInvalidateHeapTuple(), which is bad because the latter needs to look at the tuple data, which is in the shared disk buffer. If another backend manages to recycle the buffer while this is going on, we will compute the wrong hashindex for the tuple or maybe even crash outright. Must hold buffer refcount until afterwards. (This bug is not in 7.0.*; seems to be have introduced during WAL changes.)
2001-01-07 23:14:31 +01:00
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
/*
* Release the lmgr tuple lock, if we had it.
*/
if (have_tuple_lock)
Revert MERGE patch This reverts commits d204ef63776b8a00ca220adec23979091564e465, 83454e3c2b28141c0db01c7d2027e01040df5249 and a few more commits thereafter (complete list at the end) related to MERGE feature. While the feature was fully functional, with sufficient test coverage and necessary documentation, it was felt that some parts of the executor and parse-analyzer can use a different design and it wasn't possible to do that in the available time. So it was decided to revert the patch for PG11 and retry again in the future. Thanks again to all reviewers and bug reporters. List of commits reverted, in reverse chronological order: f1464c5380 Improve parse representation for MERGE ddb4158579 MERGE syntax diagram correction 530e69e59b Allow cpluspluscheck to pass by renaming variable 01b88b4df5 MERGE minor errata 3af7b2b0d4 MERGE fix variable warning in non-assert builds a5d86181ec MERGE INSERT allows only one VALUES clause 4b2d44031f MERGE post-commit review 4923550c20 Tab completion for MERGE aa3faa3c7a WITH support in MERGE 83454e3c2b New files for MERGE d204ef6377 MERGE SQL Command following SQL:2016 Author: Pavan Deolasee Reviewed-by: Michael Paquier
2018-04-12 12:22:56 +02:00
UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
Count updates that move row to a new page. Add pgstat counter to track row updates that result in the successor version going to a new heap page, leaving behind an original version whose t_ctid points to the new version. The current count is shown by the n_tup_newpage_upd column of each of the pg_stat_*_tables views. The new n_tup_newpage_upd column complements the existing n_tup_hot_upd and n_tup_upd columns. Tables that have high n_tup_newpage_upd values (relative to n_tup_upd) are good candidates for tuning heap fillfactor. Corey Huinker, with small tweaks by me. Author: Corey Huinker <corey.huinker@gmail.com> Reviewed-By: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CADkLM=ded21M9iZ36hHm-vj2rE2d=zcKpUQMds__Xm2pxLfHKA@mail.gmail.com
2023-03-23 19:16:17 +01:00
pgstat_count_heap_update(relation, use_hot_update, newbuf != buffer);
Revise pgstats stuff to fix the problems with not counting accesses generated by bitmap index scans. Along the way, simplify and speed up the code for counting sequential and index scans; it was both confusing and inefficient to be taking care of that in the per-tuple loops, IMHO. initdb forced because of internal changes in pg_stat view definitions.
2005-10-06 04:29:23 +02:00
Modify tuptoaster's API so that it does not try to modify the passed tuple in-place, but instead passes back an all-new tuple structure if any changes are needed. This is a much cleaner and more robust solution for the bug discovered by Alexey Beschiokov; accordingly, revert the quick hack I installed yesterday. With this change, HeapTupleData.t_datamcxt is no longer needed; will remove it in a separate commit in HEAD only.
2005-11-20 19:38:20 +01:00
/*
* If heaptup is a private copy, release it. Don't forget to copy t_self
* back to the caller's image, too.
*/
if (heaptup != newtup)
{
newtup->t_self = heaptup->t_self;
heap_freetuple(heaptup);
}
Ignore BRIN indexes when checking for HOT updates When determining whether an index update may be skipped by using HOT, we can ignore attributes indexed by block summarizing indexes without references to individual tuples that need to be cleaned up. A new type TU_UpdateIndexes provides a signal to the executor to determine which indexes to update - no indexes, all indexes, or only the summarizing indexes. This also removes rd_indexattr list, and replaces it with rd_attrsvalid flag. The list was not used anywhere, and a simple flag is sufficient. This was originally committed as 5753d4ee32, but then got reverted by e3fcca0d0d because of correctness issues. Original patch by Josef Simanek, various fixes and improvements by Tomas Vondra and me. Authors: Matthias van de Meent, Josef Simanek, Tomas Vondra Reviewed-by: Tomas Vondra, Alvaro Herrera Discussion: https://postgr.es/m/05ebcb44-f383-86e3-4f31-0a97a55634cf@enterprisedb.com Discussion: https://postgr.es/m/CAFp7QwpMRGcDAQumN7onN9HjrJ3u4X3ZRXdGFT0K5G2JWvnbWg%40mail.gmail.com
2023-03-20 10:34:07 +01:00
/*
* If it is a HOT update, the update may still need to update summarized
* indexes, lest we fail to update those summaries and get incorrect
* results (for example, minmax bounds of the block may change with this
* update).
*/
if (use_hot_update)
{
if (summarized_update)
*update_indexes = TU_Summarizing;
else
*update_indexes = TU_None;
}
else
*update_indexes = TU_All;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
if (old_key_tuple != NULL && old_key_copied)
heap_freetuple(old_key_tuple);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
bms_free(hot_attrs);
Ignore BRIN indexes when checking for HOT updates When determining whether an index update may be skipped by using HOT, we can ignore attributes indexed by block summarizing indexes without references to individual tuples that need to be cleaned up. A new type TU_UpdateIndexes provides a signal to the executor to determine which indexes to update - no indexes, all indexes, or only the summarizing indexes. This also removes rd_indexattr list, and replaces it with rd_attrsvalid flag. The list was not used anywhere, and a simple flag is sufficient. This was originally committed as 5753d4ee32, but then got reverted by e3fcca0d0d because of correctness issues. Original patch by Josef Simanek, various fixes and improvements by Tomas Vondra and me. Authors: Matthias van de Meent, Josef Simanek, Tomas Vondra Reviewed-by: Tomas Vondra, Alvaro Herrera Discussion: https://postgr.es/m/05ebcb44-f383-86e3-4f31-0a97a55634cf@enterprisedb.com Discussion: https://postgr.es/m/CAFp7QwpMRGcDAQumN7onN9HjrJ3u4X3ZRXdGFT0K5G2JWvnbWg%40mail.gmail.com
2023-03-20 10:34:07 +01:00
bms_free(sum_attrs);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
bms_free(key_attrs);
Fix small query-lifespan memory leak in bulk updates. When there is an identifiable REPLICA IDENTITY index on the target table, heap_update leaks the id_attrs bitmapset. That's not many bytes, but it adds up over enough rows, since the code typically runs in a query-lifespan context. Bug introduced in commit e55704d8b, which did a rather poor job of cloning the existing use-pattern for RelationGetIndexAttrBitmap(). Per bug #14293 from Zhou Digoal. Back-patch to 9.4 where the bug was introduced. Report: <20160824114320.15676.45171@wrigleys.postgresql.org>
2016-08-25 04:20:01 +02:00
bms_free(id_attrs);
Simplify check of modified attributes in heap_update The old coding was getting more complicated as new things were added, and it would be barely tolerable with upcoming WARM updates and other future features such as indirect indexes. The new coding incurs a small performance cost in synthetic benchmark cases, and is barely measurable in normal cases. A much larger benefit is expected from WARM, which could actually bolt its needs on top of the existing coding, but it is much uglier and bug-prone than doing it on this new code. Additional optimization can be applied on top of this, if need be. Reviewed-by: Pavan Deolasee, Amit Kapila, Mithun CY Discussion: https://postgr.es/m/20161228232018.4hc66ndrzpz4g4wn@alvherre.pgsql https://postgr.es/m/CABOikdMJfz69dBNRTOZcB6s5A0tf8OMCyQVYQyR-WFFdoEwKMQ@mail.gmail.com
2017-03-29 17:18:48 +02:00
bms_free(modified_attrs);
bms_free(interesting_attrs);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
return TM_Ok;
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
}
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
/*
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
* Check if the specified attribute's values are the same. Subroutine for
* HeapDetermineColumnsInfo.
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
*/
static bool
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
heap_attr_equals(TupleDesc tupdesc, int attrnum, Datum value1, Datum value2,
bool isnull1, bool isnull2)
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
{
Form_pg_attribute att;
/*
* If one value is NULL and other is not, then they are certainly not
* equal
*/
if (isnull1 != isnull2)
return false;
/*
* If both are NULL, they can be considered equal.
*/
if (isnull1)
return true;
/*
* We do simple binary comparison of the two datums. This may be overly
* strict because there can be multiple binary representations for the
* same logical value. But we should be OK as long as there are no false
* positives. Using a type-specific equality operator is messy because
* there could be multiple notions of equality in different operator
* classes; furthermore, we cannot safely invoke user-defined functions
* while holding exclusive buffer lock.
*/
if (attrnum <= 0)
{
/* The only allowed system columns are OIDs, so do this */
return (DatumGetObjectId(value1) == DatumGetObjectId(value2));
}
else
{
Assert(attrnum <= tupdesc->natts);
Change tupledesc->attrs[n] to TupleDescAttr(tupledesc, n). This is a mechanical change in preparation for a later commit that will change the layout of TupleDesc. Introducing a macro to abstract the details of where attributes are stored will allow us to change that in separate step and revise it in future. Author: Thomas Munro, editorialized by Andres Freund Reviewed-By: Andres Freund Discussion: https://postgr.es/m/CAEepm=0ZtQ-SpsgCyzzYpsXS6e=kZWqk3g5Ygn3MDV7A8dabUA@mail.gmail.com
2017-08-20 20:19:07 +02:00
att = TupleDescAttr(tupdesc, attrnum - 1);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
return datumIsEqual(value1, value2, att->attbyval, att->attlen);
}
}
/*
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* Check which columns are being updated.
*
Simplify check of modified attributes in heap_update The old coding was getting more complicated as new things were added, and it would be barely tolerable with upcoming WARM updates and other future features such as indirect indexes. The new coding incurs a small performance cost in synthetic benchmark cases, and is barely measurable in normal cases. A much larger benefit is expected from WARM, which could actually bolt its needs on top of the existing coding, but it is much uglier and bug-prone than doing it on this new code. Additional optimization can be applied on top of this, if need be. Reviewed-by: Pavan Deolasee, Amit Kapila, Mithun CY Discussion: https://postgr.es/m/20161228232018.4hc66ndrzpz4g4wn@alvherre.pgsql https://postgr.es/m/CABOikdMJfz69dBNRTOZcB6s5A0tf8OMCyQVYQyR-WFFdoEwKMQ@mail.gmail.com
2017-03-29 17:18:48 +02:00
* Given an updated tuple, determine (and return into the output bitmapset),
* from those listed as interesting, the set of columns that changed.
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
*
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
* has_external indicates if any of the unmodified attributes (from those
* listed as interesting) of the old tuple is a member of external_cols and is
* stored externally.
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
*/
Simplify check of modified attributes in heap_update The old coding was getting more complicated as new things were added, and it would be barely tolerable with upcoming WARM updates and other future features such as indirect indexes. The new coding incurs a small performance cost in synthetic benchmark cases, and is barely measurable in normal cases. A much larger benefit is expected from WARM, which could actually bolt its needs on top of the existing coding, but it is much uglier and bug-prone than doing it on this new code. Additional optimization can be applied on top of this, if need be. Reviewed-by: Pavan Deolasee, Amit Kapila, Mithun CY Discussion: https://postgr.es/m/20161228232018.4hc66ndrzpz4g4wn@alvherre.pgsql https://postgr.es/m/CABOikdMJfz69dBNRTOZcB6s5A0tf8OMCyQVYQyR-WFFdoEwKMQ@mail.gmail.com
2017-03-29 17:18:48 +02:00
static Bitmapset *
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
HeapDetermineColumnsInfo(Relation relation,
Bitmapset *interesting_cols,
Bitmapset *external_cols,
HeapTuple oldtup, HeapTuple newtup,
bool *has_external)
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
{
Remove bms_first_member(). This function has been semi-deprecated ever since we invented bms_next_member(). Its habit of scribbling on the input bitmapset isn't great, plus for sufficiently large bitmapsets it would take O(N^2) time to complete a loop. Now we have the additional problem that reducing the input to empty while leaving it still accessible would violate a planned invariant. So let's just get rid of it, after updating the few extant callers to use bms_next_member(). Patch by me; thanks to Nathan Bossart and Richard Guo for review. Discussion: https://postgr.es/m/1159933.1677621588@sss.pgh.pa.us
2023-03-02 17:34:29 +01:00
int attidx;
Simplify check of modified attributes in heap_update The old coding was getting more complicated as new things were added, and it would be barely tolerable with upcoming WARM updates and other future features such as indirect indexes. The new coding incurs a small performance cost in synthetic benchmark cases, and is barely measurable in normal cases. A much larger benefit is expected from WARM, which could actually bolt its needs on top of the existing coding, but it is much uglier and bug-prone than doing it on this new code. Additional optimization can be applied on top of this, if need be. Reviewed-by: Pavan Deolasee, Amit Kapila, Mithun CY Discussion: https://postgr.es/m/20161228232018.4hc66ndrzpz4g4wn@alvherre.pgsql https://postgr.es/m/CABOikdMJfz69dBNRTOZcB6s5A0tf8OMCyQVYQyR-WFFdoEwKMQ@mail.gmail.com
2017-03-29 17:18:48 +02:00
Bitmapset *modified = NULL;
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
TupleDesc tupdesc = RelationGetDescr(relation);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
Remove bms_first_member(). This function has been semi-deprecated ever since we invented bms_next_member(). Its habit of scribbling on the input bitmapset isn't great, plus for sufficiently large bitmapsets it would take O(N^2) time to complete a loop. Now we have the additional problem that reducing the input to empty while leaving it still accessible would violate a planned invariant. So let's just get rid of it, after updating the few extant callers to use bms_next_member(). Patch by me; thanks to Nathan Bossart and Richard Guo for review. Discussion: https://postgr.es/m/1159933.1677621588@sss.pgh.pa.us
2023-03-02 17:34:29 +01:00
attidx = -1;
while ((attidx = bms_next_member(interesting_cols, attidx)) >= 0)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
Remove bms_first_member(). This function has been semi-deprecated ever since we invented bms_next_member(). Its habit of scribbling on the input bitmapset isn't great, plus for sufficiently large bitmapsets it would take O(N^2) time to complete a loop. Now we have the additional problem that reducing the input to empty while leaving it still accessible would violate a planned invariant. So let's just get rid of it, after updating the few extant callers to use bms_next_member(). Patch by me; thanks to Nathan Bossart and Richard Guo for review. Discussion: https://postgr.es/m/1159933.1677621588@sss.pgh.pa.us
2023-03-02 17:34:29 +01:00
/* attidx is zero-based, attrnum is the normal attribute number */
AttrNumber attrnum = attidx + FirstLowInvalidHeapAttributeNumber;
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
Datum value1,
value2;
bool isnull1,
isnull2;
/*
* If it's a whole-tuple reference, say "not equal". It's not really
* worth supporting this case, since it could only succeed after a
* no-op update, which is hardly a case worth optimizing for.
*/
if (attrnum == 0)
{
Remove bms_first_member(). This function has been semi-deprecated ever since we invented bms_next_member(). Its habit of scribbling on the input bitmapset isn't great, plus for sufficiently large bitmapsets it would take O(N^2) time to complete a loop. Now we have the additional problem that reducing the input to empty while leaving it still accessible would violate a planned invariant. So let's just get rid of it, after updating the few extant callers to use bms_next_member(). Patch by me; thanks to Nathan Bossart and Richard Guo for review. Discussion: https://postgr.es/m/1159933.1677621588@sss.pgh.pa.us
2023-03-02 17:34:29 +01:00
modified = bms_add_member(modified, attidx);
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
continue;
}
/*
* Likewise, automatically say "not equal" for any system attribute
* other than tableOID; we cannot expect these to be consistent in a
* HOT chain, or even to be set correctly yet in the new tuple.
*/
if (attrnum < 0)
{
if (attrnum != TableOidAttributeNumber)
{
Remove bms_first_member(). This function has been semi-deprecated ever since we invented bms_next_member(). Its habit of scribbling on the input bitmapset isn't great, plus for sufficiently large bitmapsets it would take O(N^2) time to complete a loop. Now we have the additional problem that reducing the input to empty while leaving it still accessible would violate a planned invariant. So let's just get rid of it, after updating the few extant callers to use bms_next_member(). Patch by me; thanks to Nathan Bossart and Richard Guo for review. Discussion: https://postgr.es/m/1159933.1677621588@sss.pgh.pa.us
2023-03-02 17:34:29 +01:00
modified = bms_add_member(modified, attidx);
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
continue;
}
}
/*
* Extract the corresponding values. XXX this is pretty inefficient
* if there are many indexed columns. Should we do a single
* heap_deform_tuple call on each tuple, instead? But that doesn't
* work for system columns ...
*/
value1 = heap_getattr(oldtup, attrnum, tupdesc, &isnull1);
value2 = heap_getattr(newtup, attrnum, tupdesc, &isnull2);
if (!heap_attr_equals(tupdesc, attrnum, value1,
value2, isnull1, isnull2))
{
Remove bms_first_member(). This function has been semi-deprecated ever since we invented bms_next_member(). Its habit of scribbling on the input bitmapset isn't great, plus for sufficiently large bitmapsets it would take O(N^2) time to complete a loop. Now we have the additional problem that reducing the input to empty while leaving it still accessible would violate a planned invariant. So let's just get rid of it, after updating the few extant callers to use bms_next_member(). Patch by me; thanks to Nathan Bossart and Richard Guo for review. Discussion: https://postgr.es/m/1159933.1677621588@sss.pgh.pa.us
2023-03-02 17:34:29 +01:00
modified = bms_add_member(modified, attidx);
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
continue;
}
/*
* No need to check attributes that can't be stored externally. Note
* that system attributes can't be stored externally.
*/
if (attrnum < 0 || isnull1 ||
TupleDescAttr(tupdesc, attrnum - 1)->attlen != -1)
continue;
/*
* Check if the old tuple's attribute is stored externally and is a
* member of external_cols.
*/
if (VARATT_IS_EXTERNAL((struct varlena *) DatumGetPointer(value1)) &&
Remove bms_first_member(). This function has been semi-deprecated ever since we invented bms_next_member(). Its habit of scribbling on the input bitmapset isn't great, plus for sufficiently large bitmapsets it would take O(N^2) time to complete a loop. Now we have the additional problem that reducing the input to empty while leaving it still accessible would violate a planned invariant. So let's just get rid of it, after updating the few extant callers to use bms_next_member(). Patch by me; thanks to Nathan Bossart and Richard Guo for review. Discussion: https://postgr.es/m/1159933.1677621588@sss.pgh.pa.us
2023-03-02 17:34:29 +01:00
bms_is_member(attidx, external_cols))
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
*has_external = true;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
}
Simplify check of modified attributes in heap_update The old coding was getting more complicated as new things were added, and it would be barely tolerable with upcoming WARM updates and other future features such as indirect indexes. The new coding incurs a small performance cost in synthetic benchmark cases, and is barely measurable in normal cases. A much larger benefit is expected from WARM, which could actually bolt its needs on top of the existing coding, but it is much uglier and bug-prone than doing it on this new code. Additional optimization can be applied on top of this, if need be. Reviewed-by: Pavan Deolasee, Amit Kapila, Mithun CY Discussion: https://postgr.es/m/20161228232018.4hc66ndrzpz4g4wn@alvherre.pgsql https://postgr.es/m/CABOikdMJfz69dBNRTOZcB6s5A0tf8OMCyQVYQyR-WFFdoEwKMQ@mail.gmail.com
2017-03-29 17:18:48 +02:00
return modified;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
}
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
/*
* simple_heap_update - replace a tuple
*
* This routine may be used to update a tuple when concurrent updates of
* the target tuple are not expected (for example, because we have a lock
* on the relation associated with the tuple). Any failure is reported
Error message editing in backend/access.
2003-07-21 22:29:40 +02:00
* via ereport().
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
*/
void
Ignore BRIN indexes when checking for HOT updates When determining whether an index update may be skipped by using HOT, we can ignore attributes indexed by block summarizing indexes without references to individual tuples that need to be cleaned up. A new type TU_UpdateIndexes provides a signal to the executor to determine which indexes to update - no indexes, all indexes, or only the summarizing indexes. This also removes rd_indexattr list, and replaces it with rd_attrsvalid flag. The list was not used anywhere, and a simple flag is sufficient. This was originally committed as 5753d4ee32, but then got reverted by e3fcca0d0d because of correctness issues. Original patch by Josef Simanek, various fixes and improvements by Tomas Vondra and me. Authors: Matthias van de Meent, Josef Simanek, Tomas Vondra Reviewed-by: Tomas Vondra, Alvaro Herrera Discussion: https://postgr.es/m/05ebcb44-f383-86e3-4f31-0a97a55634cf@enterprisedb.com Discussion: https://postgr.es/m/CAFp7QwpMRGcDAQumN7onN9HjrJ3u4X3ZRXdGFT0K5G2JWvnbWg%40mail.gmail.com
2023-03-20 10:34:07 +01:00
simple_heap_update(Relation relation, ItemPointer otid, HeapTuple tup,
TU_UpdateIndexes *update_indexes)
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
TM_Result result;
TM_FailureData tmfd;
Revert MERGE patch This reverts commits d204ef63776b8a00ca220adec23979091564e465, 83454e3c2b28141c0db01c7d2027e01040df5249 and a few more commits thereafter (complete list at the end) related to MERGE feature. While the feature was fully functional, with sufficient test coverage and necessary documentation, it was felt that some parts of the executor and parse-analyzer can use a different design and it wasn't possible to do that in the available time. So it was decided to revert the patch for PG11 and retry again in the future. Thanks again to all reviewers and bug reporters. List of commits reverted, in reverse chronological order: f1464c5380 Improve parse representation for MERGE ddb4158579 MERGE syntax diagram correction 530e69e59b Allow cpluspluscheck to pass by renaming variable 01b88b4df5 MERGE minor errata 3af7b2b0d4 MERGE fix variable warning in non-assert builds a5d86181ec MERGE INSERT allows only one VALUES clause 4b2d44031f MERGE post-commit review 4923550c20 Tab completion for MERGE aa3faa3c7a WITH support in MERGE 83454e3c2b New files for MERGE d204ef6377 MERGE SQL Command following SQL:2016 Author: Pavan Deolasee Reviewed-by: Michael Paquier
2018-04-12 12:22:56 +02:00
LockTupleMode lockmode;
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
Fix LISTEN/NOTIFY race condition reported by Gavin Sherry. While a really general fix might be difficult, I believe the only case where AtCommit_Notify could see an uncommitted tuple is where the other guy has just unlistened and not yet committed. The best solution seems to be to just skip updating that tuple, on the assumption that the other guy does not want to hear about the notification anyway. This is not perfect --- if the other guy rolls back his unlisten instead of committing, then he really should have gotten this notify. But to do that, we'd have to wait to see if he commits or not, or make UNLISTEN hold exclusive lock on pg_listener until commit. Either of these answers is deadlock-prone, not to mention horrible for interactive performance. Do it this way for now. (What happened to that project to do LISTEN/NOTIFY in memory with no table, anyway?)
2003-09-16 01:33:43 +02:00
result = heap_update(relation, otid, tup,
Avoid incrementing the CommandCounter when CommandCounterIncrement is called but no database changes have been made since the last CommandCounterIncrement. This should result in a significant improvement in the number of "commands" that can typically be performed within a transaction before hitting the 2^32 CommandId size limit. In particular this buys back (and more) the possible adverse consequences of my previous patch to fix plan caching behavior. The implementation requires tracking whether the current CommandCounter value has been "used" to mark any tuples. CommandCounter values stored into snapshots are presumed not to be used for this purpose. This requires some small executor changes, since the executor used to conflate the curcid of the snapshot it was using with the command ID to mark output tuples with. Separating these concepts allows some small simplifications in executor APIs. Something for the TODO list: look into having CommandCounterIncrement not do AcceptInvalidationMessages. It seems fairly bogus to be doing it there, but exactly where to do it instead isn't clear, and I'm disinclined to mess with asynchronous behavior during late beta.
2007-11-30 22:22:54 +01:00
GetCurrentCommandId(true), InvalidSnapshot,
Throw error if expiring tuple is again updated or deleted. This prevents surprising behavior when a FOR EACH ROW trigger BEFORE UPDATE or BEFORE DELETE directly or indirectly updates or deletes the the old row. Prior to this patch the requested action on the row could be silently ignored while all triggered actions based on the occurence of the requested action could be committed. One example of how this could happen is if the BEFORE DELETE trigger for a "parent" row deleted "children" which had trigger functions to update summary or status data on the parent. This also prevents similar surprising problems if the query has a volatile function which updates a target row while it is already being updated. There are related issues present in FOR UPDATE cursors and READ COMMITTED queries which are not handled by this patch. These issues need further evalution to determine what change, if any, is needed. Where the new error messages are generated, in most cases the best fix will be to move code from the BEFORE trigger to an AFTER trigger. Where this is not feasible, the trigger can avoid the error by re-issuing the triggering statement and returning NULL. Documentation changes will be submitted in a separate patch. Kevin Grittner and Tom Lane with input from Florian Pflug and Robert Haas, based on problems encountered during conversion of Wisconsin Circuit Court trigger logic to plpgsql triggers.
2012-10-26 21:55:36 +02:00
true /* wait for commit */ ,
Ignore BRIN indexes when checking for HOT updates When determining whether an index update may be skipped by using HOT, we can ignore attributes indexed by block summarizing indexes without references to individual tuples that need to be cleaned up. A new type TU_UpdateIndexes provides a signal to the executor to determine which indexes to update - no indexes, all indexes, or only the summarizing indexes. This also removes rd_indexattr list, and replaces it with rd_attrsvalid flag. The list was not used anywhere, and a simple flag is sufficient. This was originally committed as 5753d4ee32, but then got reverted by e3fcca0d0d because of correctness issues. Original patch by Josef Simanek, various fixes and improvements by Tomas Vondra and me. Authors: Matthias van de Meent, Josef Simanek, Tomas Vondra Reviewed-by: Tomas Vondra, Alvaro Herrera Discussion: https://postgr.es/m/05ebcb44-f383-86e3-4f31-0a97a55634cf@enterprisedb.com Discussion: https://postgr.es/m/CAFp7QwpMRGcDAQumN7onN9HjrJ3u4X3ZRXdGFT0K5G2JWvnbWg%40mail.gmail.com
2023-03-20 10:34:07 +01:00
&tmfd, &lockmode, update_indexes);
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
switch (result)
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
case TM_SelfModified:
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
/* Tuple was already updated in current command? */
Error message editing in backend/access.
2003-07-21 22:29:40 +02:00
elog(ERROR, "tuple already updated by self");
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
break;
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
case TM_Ok:
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
/* done successfully */
break;
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
case TM_Updated:
Error message editing in backend/access.
2003-07-21 22:29:40 +02:00
elog(ERROR, "tuple concurrently updated");
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
break;
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
case TM_Deleted:
elog(ERROR, "tuple concurrently deleted");
break;
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
default:
Error message editing in backend/access.
2003-07-21 22:29:40 +02:00
elog(ERROR, "unrecognized heap_update status: %u", result);
Fix all the places that called heap_update() and heap_delete() without bothering to check the return value --- which meant that in case the update or delete failed because of a concurrent update, you'd not find out about it, except by observing later that the transaction produced the wrong outcome. There are now subroutines simple_heap_update and simple_heap_delete that should be used anyplace that you're not prepared to do the full nine yards of coping with concurrent updates. In practice, that seems to mean absolutely everywhere but the executor, because *noplace* else was checking.
2001-01-23 05:32:23 +01:00
break;
}
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* Return the MultiXactStatus corresponding to the given tuple lock mode.
*/
static MultiXactStatus
get_mxact_status_for_lock(LockTupleMode mode, bool is_update)
{
Avoid inconsistent type declaration Clang 3.3 correctly complains that a variable of type enum MultiXactStatus cannot hold a value of -1, which makes sense. Change the declared type of the variable to int instead, and apply casting as necessary to avoid the warning. Per notice from Andres Freund
2013-06-25 22:36:29 +02:00
int retval;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if (is_update)
retval = tupleLockExtraInfo[mode].updstatus;
else
retval = tupleLockExtraInfo[mode].lockstatus;
if (retval == -1)
elog(ERROR, "invalid lock tuple mode %d/%s", mode,
is_update ? "true" : "false");
Avoid inconsistent type declaration Clang 3.3 correctly complains that a variable of type enum MultiXactStatus cannot hold a value of -1, which makes sense. Change the declared type of the variable to int instead, and apply casting as necessary to avoid the warning. Per notice from Andres Freund
2013-06-25 22:36:29 +02:00
return (MultiXactStatus) retval;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
/*
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
* heap_lock_tuple - lock a tuple in shared or exclusive mode
*
* Note that this acquires a buffer pin, which the caller must release.
*
* Input parameters:
* relation: relation containing tuple (caller must hold suitable lock)
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
* tid: TID of tuple to lock
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
* cid: current command ID (used for visibility test, and stored into
* tuple's cmax if lock is successful)
* mode: indicates if shared or exclusive tuple lock is desired
Implement SKIP LOCKED for row-level locks This clause changes the behavior of SELECT locking clauses in the presence of locked rows: instead of causing a process to block waiting for the locks held by other processes (or raise an error, with NOWAIT), SKIP LOCKED makes the new reader skip over such rows. While this is not appropriate behavior for general purposes, there are some cases in which it is useful, such as queue-like tables. Catalog version bumped because this patch changes the representation of stored rules. Reviewed by Craig Ringer (based on a previous attempt at an implementation by Simon Riggs, who also provided input on the syntax used in the current patch), David Rowley, and Álvaro Herrera. Author: Thomas Munro
2014-10-07 22:23:34 +02:00
* wait_policy: what to do if tuple lock is not available
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* follow_updates: if true, follow the update chain to also lock descendant
* tuples.
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
*
* Output parameters:
* *tuple: all fields filled in
* *buffer: set to buffer holding tuple (pinned but not locked at exit)
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
* *tmfd: filled in failure cases (see below)
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
*
tableam: Rename wrapper functions to match callback names. Some of the wrapper functions didn't match the callback names. Many of them due to staying "consistent" with historic naming of the wrapped functionality. We decided that for most cases it's more important to be for tableam to be consistent going forward, than with the past. The one exception is beginscan/endscan/... because it'd have looked odd to have systable_beginscan/endscan/... with a different naming scheme, and changing the systable_* APIs would have caused way too much churn (including breaking a lot of external users). Author: Ashwin Agrawal, with some small additions by Andres Freund Reviewed-By: Andres Freund Discussion: https://postgr.es/m/CALfoeiugyrXZfX7n0ORCa4L-m834dzmaE8eFdbNR6PMpetU4Ww@mail.gmail.com
2019-05-24 01:25:48 +02:00
* Function results are the same as the ones for table_tuple_lock().
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
*
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
* In the failure cases other than TM_Invisible, the routine fills
* *tmfd with the tuple's t_ctid, t_xmax (resolving a possible MultiXact,
* if necessary), and t_cmax (the last only for TM_SelfModified,
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
* since we cannot obtain cmax from a combo CID generated by another
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
* transaction).
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
* See comments for struct TM_FailureData for additional info.
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
*
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* See README.tuplock for a thorough explanation of this mechanism.
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
TM_Result
Throw error if expiring tuple is again updated or deleted. This prevents surprising behavior when a FOR EACH ROW trigger BEFORE UPDATE or BEFORE DELETE directly or indirectly updates or deletes the the old row. Prior to this patch the requested action on the row could be silently ignored while all triggered actions based on the occurence of the requested action could be committed. One example of how this could happen is if the BEFORE DELETE trigger for a "parent" row deleted "children" which had trigger functions to update summary or status data on the parent. This also prevents similar surprising problems if the query has a volatile function which updates a target row while it is already being updated. There are related issues present in FOR UPDATE cursors and READ COMMITTED queries which are not handled by this patch. These issues need further evalution to determine what change, if any, is needed. Where the new error messages are generated, in most cases the best fix will be to move code from the BEFORE trigger to an AFTER trigger. Where this is not feasible, the trigger can avoid the error by re-issuing the triggering statement and returning NULL. Documentation changes will be submitted in a separate patch. Kevin Grittner and Tom Lane with input from Florian Pflug and Robert Haas, based on problems encountered during conversion of Wisconsin Circuit Court trigger logic to plpgsql triggers.
2012-10-26 21:55:36 +02:00
heap_lock_tuple(Relation relation, HeapTuple tuple,
Implement SKIP LOCKED for row-level locks This clause changes the behavior of SELECT locking clauses in the presence of locked rows: instead of causing a process to block waiting for the locks held by other processes (or raise an error, with NOWAIT), SKIP LOCKED makes the new reader skip over such rows. While this is not appropriate behavior for general purposes, there are some cases in which it is useful, such as queue-like tables. Catalog version bumped because this patch changes the representation of stored rules. Reviewed by Craig Ringer (based on a previous attempt at an implementation by Simon Riggs, who also provided input on the syntax used in the current patch), David Rowley, and Álvaro Herrera. Author: Thomas Munro
2014-10-07 22:23:34 +02:00
CommandId cid, LockTupleMode mode, LockWaitPolicy wait_policy,
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
bool follow_updates,
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
Buffer *buffer, TM_FailureData *tmfd)
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
TM_Result result;
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
ItemPointer tid = &(tuple->t_self);
ItemId lp;
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
Page page;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
Buffer vmbuffer = InvalidBuffer;
BlockNumber block;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
TransactionId xid,
xmax;
uint16 old_infomask,
new_infomask,
new_infomask2;
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
bool first_time = true;
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
bool skip_tuple_lock = false;
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
bool have_tuple_lock = false;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
bool cleared_all_frozen = false;
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
*buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
block = ItemPointerGetBlockNumber(tid);
/*
Fix hard to hit race condition in heapam's tuple locking code. As mentioned in its commit message, eca0f1db left open a race condition, where a page could be marked all-visible, after the code checked PageIsAllVisible() to pin the VM, but before the page is locked. Plug that hole. Reviewed-By: Robert Haas, Andres Freund Author: Amit Kapila Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-08-05 05:07:16 +02:00
* Before locking the buffer, pin the visibility map page if it appears to
* be necessary. Since we haven't got the lock yet, someone else might be
* in the middle of changing this, so we'll need to recheck after we have
* the lock.
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
*/
if (PageIsAllVisible(BufferGetPage(*buffer)))
visibilitymap_pin(relation, block, &vmbuffer);
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(*buffer);
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
Redefine the lp_flags field of item pointers as having four states, rather than two independent bits (one of which was never used in heap pages anyway, or at least hadn't been in a very long time). This gives us flexibility to add the HOT notions of redirected and dead item pointers without requiring anything so klugy as magic values of lp_off and lp_len. The state values are chosen so that for the states currently in use (pre-HOT) there is no change in the physical representation.
2007-09-13 00:10:26 +02:00
Assert(ItemIdIsNormal(lp));
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
tuple->t_len = ItemIdGetLength(lp);
Repair problems with VACUUM destroying t_ctid chains too soon, and with insufficient paranoia in code that follows t_ctid links. (We must do both because even with VACUUM doing it properly, the intermediate state with a dangling t_ctid link is visible concurrently during lazy VACUUM, and could be seen afterwards if either type of VACUUM crashes partway through.) Also try to improve documentation about what's going on. Patch is a bit bulky because passing the XMAX information around required changing the APIs of some low-level heapam.c routines, but it's not conceptually very complicated. Per trouble report from Teodor and subsequent analysis. This needs to be back-patched, but I'll do that after 8.1 beta is out.
2005-08-20 02:40:32 +02:00
tuple->t_tableOid = RelationGetRelid(relation);
pgindent run over code.
1999-05-25 18:15:34 +02:00
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
l3:
Adjust HeapTupleSatisfies* routines to take a HeapTuple. Previously, these functions took a HeapTupleHeader, but upcoming patches for logical replication will introduce new a new snapshot type under which the tuple's TID will be used to lookup (CMIN, CMAX) for visibility determination purposes. This makes that information available. Code churn is minimal since HeapTupleSatisfiesVisibility took the HeapTuple anyway, and deferenced it before calling the satisfies function. Independently of logical replication, this allows t_tableOid and t_self to be cross-checked via assertions in tqual.c. This seems like a useful way to make sure that all callers are setting these values properly, which has been previously put forward as desirable. Andres Freund, reviewed by Álvaro Herrera
2013-07-22 19:26:33 +02:00
result = HeapTupleSatisfiesUpdate(tuple, cid, *buffer);
pgindent run over code.
1999-05-25 18:15:34 +02:00
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (result == TM_Invisible)
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
{
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
/*
* This is possible, but only when locking a tuple for ON CONFLICT
* UPDATE. We return this value here rather than throwing an error in
* order to give that case the opportunity to throw a more specific
* error.
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Invisible;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
goto out_locked;
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
}
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
else if (result == TM_BeingModified ||
result == TM_Updated ||
result == TM_Deleted)
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
{
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
TransactionId xwait;
uint16 infomask;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
uint16 infomask2;
bool require_sleep;
ItemPointerData t_ctid;
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
/* must copy state data before unlocking buffer */
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
xwait = HeapTupleHeaderGetRawXmax(tuple->t_data);
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
infomask = tuple->t_data->t_infomask;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
infomask2 = tuple->t_data->t_infomask2;
ItemPointerCopy(&tuple->t_data->t_ctid, &t_ctid);
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
Repair two related errors in heap_lock_tuple: it was failing to recognize cases where we already hold the desired lock "indirectly", either via membership in a MultiXact or because the lock was originally taken by a different subtransaction of the current transaction. These cases must be accounted for to avoid needless deadlocks and/or inappropriate replacement of an exclusive lock with a shared lock. Per report from Clarence Gardner and subsequent investigation.
2006-11-17 19:00:15 +01:00
/*
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* If any subtransaction of the current top transaction already holds
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
* a lock as strong as or stronger than what we're requesting, we
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* effectively hold the desired lock already. We *must* succeed
* without trying to take the tuple lock, else we will deadlock
* against anyone wanting to acquire a stronger lock.
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
*
* Note we only do this the first time we loop on the HTSU result;
* there is no point in testing in subsequent passes, because
* evidently our own transaction cannot have acquired a new lock after
* the first time we checked.
Repair two related errors in heap_lock_tuple: it was failing to recognize cases where we already hold the desired lock "indirectly", either via membership in a MultiXact or because the lock was originally taken by a different subtransaction of the current transaction. These cases must be accounted for to avoid needless deadlocks and/or inappropriate replacement of an exclusive lock with a shared lock. Per report from Clarence Gardner and subsequent investigation.
2006-11-17 19:00:15 +01:00
*/
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
if (first_time)
Repair two related errors in heap_lock_tuple: it was failing to recognize cases where we already hold the desired lock "indirectly", either via membership in a MultiXact or because the lock was originally taken by a different subtransaction of the current transaction. These cases must be accounted for to avoid needless deadlocks and/or inappropriate replacement of an exclusive lock with a shared lock. Per report from Clarence Gardner and subsequent investigation.
2006-11-17 19:00:15 +01:00
{
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
first_time = false;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
if (infomask & HEAP_XMAX_IS_MULTI)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
int i;
int nmembers;
MultiXactMember *members;
/*
* We don't need to allow old multixacts here; if that had
* been the case, HeapTupleSatisfiesUpdate would have returned
* MayBeUpdated and we wouldn't be here.
*/
nmembers =
GetMultiXactIdMembers(xwait, &members, false,
HEAP_XMAX_IS_LOCKED_ONLY(infomask));
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
for (i = 0; i < nmembers; i++)
{
/* only consider members of our own transaction */
if (!TransactionIdIsCurrentTransactionId(members[i].xid))
continue;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
if (TUPLOCK_from_mxstatus(members[i].status) >= mode)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
pfree(members);
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Ok;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
goto out_unlocked;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
else
{
/*
* Disable acquisition of the heavyweight tuple lock.
* Otherwise, when promoting a weaker lock, we might
* deadlock with another locker that has acquired the
* heavyweight tuple lock and is waiting for our
* transaction to finish.
*
* Note that in this case we still need to wait for
* the multixact if required, to avoid acquiring
* conflicting locks.
*/
skip_tuple_lock = true;
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
if (members)
pfree(members);
}
else if (TransactionIdIsCurrentTransactionId(xwait))
{
switch (mode)
{
case LockTupleKeyShare:
Assert(HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) ||
HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
HEAP_XMAX_IS_EXCL_LOCKED(infomask));
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Ok;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
goto out_unlocked;
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
case LockTupleShare:
if (HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
HEAP_XMAX_IS_EXCL_LOCKED(infomask))
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Ok;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
goto out_unlocked;
}
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
break;
case LockTupleNoKeyExclusive:
if (HEAP_XMAX_IS_EXCL_LOCKED(infomask))
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Ok;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
goto out_unlocked;
}
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
break;
case LockTupleExclusive:
if (HEAP_XMAX_IS_EXCL_LOCKED(infomask) &&
infomask2 & HEAP_KEYS_UPDATED)
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Ok;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
goto out_unlocked;
}
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
break;
}
}
Repair two related errors in heap_lock_tuple: it was failing to recognize cases where we already hold the desired lock "indirectly", either via membership in a MultiXact or because the lock was originally taken by a different subtransaction of the current transaction. These cases must be accounted for to avoid needless deadlocks and/or inappropriate replacement of an exclusive lock with a shared lock. Per report from Clarence Gardner and subsequent investigation.
2006-11-17 19:00:15 +01:00
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* Initially assume that we will have to wait for the locking
* transaction(s) to finish. We check various cases below in which
* this can be turned off.
*/
require_sleep = true;
if (mode == LockTupleKeyShare)
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
{
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
/*
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* If we're requesting KeyShare, and there's no update present, we
* don't need to wait. Even if there is an update, we can still
* continue if the key hasn't been modified.
*
* However, if there are updates, we need to walk the update chain
* to mark future versions of the row as locked, too. That way,
* if somebody deletes that future version, we're protected
* against the key going away. This locking of future versions
* could block momentarily, if a concurrent transaction is
* deleting a key; or it could return a value to the effect that
* the transaction deleting the key has already committed. So we
* do this before re-locking the buffer; otherwise this would be
* prone to deadlocks.
*
* Note that the TID we're locking was grabbed before we unlocked
* the buffer. For it to change while we're not looking, the
* other properties we're testing for below after re-locking the
* buffer would also change, in which case we would restart this
* loop above.
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
*/
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if (!(infomask2 & HEAP_KEYS_UPDATED))
Add NOWAIT option to SELECT FOR UPDATE/SHARE. Original patch by Hans-Juergen Schoenig, revisions by Karel Zak and Tom Lane.
2005-08-01 22:31:16 +02:00
{
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
bool updated;
Add NOWAIT option to SELECT FOR UPDATE/SHARE. Original patch by Hans-Juergen Schoenig, revisions by Karel Zak and Tom Lane.
2005-08-01 22:31:16 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
updated = !HEAP_XMAX_IS_LOCKED_ONLY(infomask);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* If there are updates, follow the update chain; bail out if
* that cannot be done.
*/
if (follow_updates && updated)
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
TM_Result res;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
res = heap_lock_updated_tuple(relation, tuple, &t_ctid,
GetCurrentTransactionId(),
mode);
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (res != TM_Ok)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
result = res;
/* recovery code expects to have buffer lock held */
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
goto failed;
}
}
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
/*
* Make sure it's still an appropriate lock, else start over.
* Also, if it wasn't updated before we released the lock, but
* is updated now, we start over too; the reason is that we
* now need to follow the update chain to lock the new
* versions.
*/
if (!HeapTupleHeaderIsOnlyLocked(tuple->t_data) &&
((tuple->t_data->t_infomask2 & HEAP_KEYS_UPDATED) ||
!updated))
goto l3;
/* Things look okay, so we can skip sleeping */
require_sleep = false;
/*
* Note we allow Xmax to change here; other updaters/lockers
* could have modified it before we grabbed the buffer lock.
* However, this is not a problem, because with the recheck we
* just did we ensure that they still don't conflict with the
* lock we want.
*/
}
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
else if (mode == LockTupleShare)
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
{
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* If we're requesting Share, we can similarly avoid sleeping if
* there's no update and no exclusive lock present.
*/
if (HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
!HEAP_XMAX_IS_EXCL_LOCKED(infomask))
Add NOWAIT option to SELECT FOR UPDATE/SHARE. Original patch by Hans-Juergen Schoenig, revisions by Karel Zak and Tom Lane.
2005-08-01 22:31:16 +02:00
{
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* Make sure it's still an appropriate lock, else start over.
* See above about allowing xmax to change.
*/
if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
HEAP_XMAX_IS_EXCL_LOCKED(tuple->t_data->t_infomask))
goto l3;
require_sleep = false;
}
}
else if (mode == LockTupleNoKeyExclusive)
{
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
/*
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* If we're requesting NoKeyExclusive, we might also be able to
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
* avoid sleeping; just ensure that there no conflicting lock
* already acquired.
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
*/
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if (infomask & HEAP_XMAX_IS_MULTI)
{
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
mode, NULL))
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
/*
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
* No conflict, but if the xmax changed under us in the
* meantime, start over.
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
*/
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
!TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
xwait))
goto l3;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
/* otherwise, we're good */
require_sleep = false;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
}
else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
{
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
/* if the xmax changed in the meantime, start over */
Fix race when updating a tuple concurrently locked by another process If a tuple is locked, and this lock is later upgraded either to an update or to a stronger lock, and in the meantime some other process tries to lock, update or delete the same tuple, it (the tuple) could end up being updated twice, or having conflicting locks held. The reason for this is that the second updater checks for a change in Xmax value, or in the HEAP_XMAX_IS_MULTI infomask bit, after noticing the first lock; and if there's a change, it restarts and re-evaluates its ability to update the tuple. But it neglected to check for changes in lock strength or in lock-vs-update status when those two properties stayed the same. This would lead it to take the wrong decision and continue with its own update, when in reality it shouldn't do so but instead restart from the top. This could lead to either an assertion failure much later (when a multixact containing multiple updates is detected), or duplicate copies of tuples. To fix, make sure to compare the other relevant infomask bits alongside the Xmax value and HEAP_XMAX_IS_MULTI bit, and restart from the top if necessary. Also, in the belt-and-suspenders spirit, add a check to MultiXactCreateFromMembers that a multixact being created does not have two or more members that are claimed to be updates. This should protect against other bugs that might cause similar bogus situations. Backpatch to 9.3, where the possibility of multixacts containing updates was introduced. (In prior versions it was possible to have the tuple lock upgraded from shared to exclusive, and an update would not restart from the top; yet we're protected against a bug there because there's always a sleep to wait for the locking transaction to complete before continuing to do anything. Really, the fact that tuple locks always conflicted with concurrent updates is what protected against bugs here.) Per report from Andrew Dunstan and Josh Berkus in thread at http://www.postgresql.org/message-id/534C8B33.9050807@pgexperts.com Bug analysis by Andres Freund.
2014-04-24 20:41:55 +02:00
if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
Post-pgindent cleanup Make slightly better decisions about indentation than what pgindent is capable of. Mostly breaking out long function calls into one line per argument, with a few other minor adjustments. No functional changes- all whitespace. pgindent ran cleanly (didn't change anything) after. Passes all regressions.
2013-06-01 15:38:15 +02:00
!TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
xwait))
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
goto l3;
/* otherwise, we're good */
require_sleep = false;
}
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
}
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
/*
* As a check independent from those above, we can also avoid sleeping
* if the current transaction is the sole locker of the tuple. Note
* that the strength of the lock already held is irrelevant; this is
* not about recording the lock in Xmax (which will be done regardless
* of this optimization, below). Also, note that the cases where we
* hold a lock stronger than we are requesting are already handled
* above by not doing anything.
*
* Note we only deal with the non-multixact case here; MultiXactIdWait
* is well equipped to deal with this situation on its own.
*/
if (require_sleep && !(infomask & HEAP_XMAX_IS_MULTI) &&
TransactionIdIsCurrentTransactionId(xwait))
{
/* ... but if the xmax changed in the meantime, start over */
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
!TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
xwait))
goto l3;
Assert(HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask));
require_sleep = false;
}
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
/*
Avoid serializability errors when locking a tuple with a committed update When key-share locking a tuple that has been not-key-updated, and the update is a committed transaction, in some cases we raised serializability errors: ERROR: could not serialize access due to concurrent update Because the key-share doesn't conflict with the update, the error is unnecessary and inconsistent with the case that the update hasn't committed yet. This causes problems for some usage patterns, even if it can be claimed that it's sufficient to retry the aborted transaction: given a steady stream of updating transactions and a long locking transaction, the long transaction can be starved indefinitely despite multiple retries. To fix, we recognize that HeapTupleSatisfiesUpdate can return HeapTupleUpdated when an updating transaction has committed, and that we need to deal with that case exactly as if it were a non-committed update: verify whether the two operations conflict, and if not, carry on normally. If they do conflict, however, there is a difference: in the HeapTupleBeingUpdated case we can just sleep until the concurrent transaction is gone, while in the HeapTupleUpdated case this is not possible and we must raise an error instead. Per trouble report from Olivier Dony. In addition to a couple of test cases that verify the changed behavior, I added a test case to verify the behavior that remains unchanged, namely that errors are raised when a update that modifies the key is used. That must still generate serializability errors. One pre-existing test case changes behavior; per discussion, the new behavior is actually the desired one. Discussion: https://www.postgresql.org/message-id/560AA479.4080807@odoo.com https://www.postgresql.org/message-id/20151014164844.3019.25750@wrigleys.postgresql.org Backpatch to 9.3, where the problem appeared.
2016-07-15 20:17:20 +02:00
* Time to sleep on the other transaction/multixact, if necessary.
*
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
* If the other transaction is an update/delete that's already
* committed, then sleeping cannot possibly do any good: if we're
* required to sleep, get out to raise an error instead.
Avoid serializability errors when locking a tuple with a committed update When key-share locking a tuple that has been not-key-updated, and the update is a committed transaction, in some cases we raised serializability errors: ERROR: could not serialize access due to concurrent update Because the key-share doesn't conflict with the update, the error is unnecessary and inconsistent with the case that the update hasn't committed yet. This causes problems for some usage patterns, even if it can be claimed that it's sufficient to retry the aborted transaction: given a steady stream of updating transactions and a long locking transaction, the long transaction can be starved indefinitely despite multiple retries. To fix, we recognize that HeapTupleSatisfiesUpdate can return HeapTupleUpdated when an updating transaction has committed, and that we need to deal with that case exactly as if it were a non-committed update: verify whether the two operations conflict, and if not, carry on normally. If they do conflict, however, there is a difference: in the HeapTupleBeingUpdated case we can just sleep until the concurrent transaction is gone, while in the HeapTupleUpdated case this is not possible and we must raise an error instead. Per trouble report from Olivier Dony. In addition to a couple of test cases that verify the changed behavior, I added a test case to verify the behavior that remains unchanged, namely that errors are raised when a update that modifies the key is used. That must still generate serializability errors. One pre-existing test case changes behavior; per discussion, the new behavior is actually the desired one. Discussion: https://www.postgresql.org/message-id/560AA479.4080807@odoo.com https://www.postgresql.org/message-id/20151014164844.3019.25750@wrigleys.postgresql.org Backpatch to 9.3, where the problem appeared.
2016-07-15 20:17:20 +02:00
*
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* By here, we either have already acquired the buffer exclusive lock,
* or we must wait for the locking transaction or multixact; so below
* we ensure that we grab buffer lock after the sleep.
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (require_sleep && (result == TM_Updated || result == TM_Deleted))
Avoid serializability errors when locking a tuple with a committed update When key-share locking a tuple that has been not-key-updated, and the update is a committed transaction, in some cases we raised serializability errors: ERROR: could not serialize access due to concurrent update Because the key-share doesn't conflict with the update, the error is unnecessary and inconsistent with the case that the update hasn't committed yet. This causes problems for some usage patterns, even if it can be claimed that it's sufficient to retry the aborted transaction: given a steady stream of updating transactions and a long locking transaction, the long transaction can be starved indefinitely despite multiple retries. To fix, we recognize that HeapTupleSatisfiesUpdate can return HeapTupleUpdated when an updating transaction has committed, and that we need to deal with that case exactly as if it were a non-committed update: verify whether the two operations conflict, and if not, carry on normally. If they do conflict, however, there is a difference: in the HeapTupleBeingUpdated case we can just sleep until the concurrent transaction is gone, while in the HeapTupleUpdated case this is not possible and we must raise an error instead. Per trouble report from Olivier Dony. In addition to a couple of test cases that verify the changed behavior, I added a test case to verify the behavior that remains unchanged, namely that errors are raised when a update that modifies the key is used. That must still generate serializability errors. One pre-existing test case changes behavior; per discussion, the new behavior is actually the desired one. Discussion: https://www.postgresql.org/message-id/560AA479.4080807@odoo.com https://www.postgresql.org/message-id/20151014164844.3019.25750@wrigleys.postgresql.org Backpatch to 9.3, where the problem appeared.
2016-07-15 20:17:20 +02:00
{
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
goto failed;
}
else if (require_sleep)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
/*
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
* Acquire tuple lock to establish our priority for the tuple, or
* die trying. LockTuple will release us when we are next-in-line
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
* for the tuple. We must do this even if we are share-locking,
* but not if we already have a weaker lock on the tuple.
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
*
* If we are forced to "start over" below, we keep the tuple lock;
* this arranges that we stay at the head of the line while
* rechecking tuple state.
*/
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
if (!skip_tuple_lock &&
!heap_acquire_tuplock(relation, tid, mode, wait_policy,
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
&have_tuple_lock))
{
/*
* This can only happen if wait_policy is Skip and the lock
* couldn't be obtained.
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_WouldBlock;
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
/* recovery code expects to have buffer lock held */
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
goto failed;
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if (infomask & HEAP_XMAX_IS_MULTI)
{
MultiXactStatus status = get_mxact_status_for_lock(mode, false);
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/* We only ever lock tuples, never update them */
if (status >= MultiXactStatusNoKeyUpdate)
elog(ERROR, "invalid lock mode in heap_lock_tuple");
Repair two related errors in heap_lock_tuple: it was failing to recognize cases where we already hold the desired lock "indirectly", either via membership in a MultiXact or because the lock was originally taken by a different subtransaction of the current transaction. These cases must be accounted for to avoid needless deadlocks and/or inappropriate replacement of an exclusive lock with a shared lock. Per report from Clarence Gardner and subsequent investigation.
2006-11-17 19:00:15 +01:00
Implement SKIP LOCKED for row-level locks This clause changes the behavior of SELECT locking clauses in the presence of locked rows: instead of causing a process to block waiting for the locks held by other processes (or raise an error, with NOWAIT), SKIP LOCKED makes the new reader skip over such rows. While this is not appropriate behavior for general purposes, there are some cases in which it is useful, such as queue-like tables. Catalog version bumped because this patch changes the representation of stored rules. Reviewed by Craig Ringer (based on a previous attempt at an implementation by Simon Riggs, who also provided input on the syntax used in the current patch), David Rowley, and Álvaro Herrera. Author: Thomas Munro
2014-10-07 22:23:34 +02:00
/* wait for multixact to end, or die trying */
switch (wait_policy)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
Implement SKIP LOCKED for row-level locks This clause changes the behavior of SELECT locking clauses in the presence of locked rows: instead of causing a process to block waiting for the locks held by other processes (or raise an error, with NOWAIT), SKIP LOCKED makes the new reader skip over such rows. While this is not appropriate behavior for general purposes, there are some cases in which it is useful, such as queue-like tables. Catalog version bumped because this patch changes the representation of stored rules. Reviewed by Craig Ringer (based on a previous attempt at an implementation by Simon Riggs, who also provided input on the syntax used in the current patch), David Rowley, and Álvaro Herrera. Author: Thomas Munro
2014-10-07 22:23:34 +02:00
case LockWaitBlock:
MultiXactIdWait((MultiXactId) xwait, status, infomask,
Fix reference-after-free when waiting for another xact due to constraint. If an insertion or update had to wait for another transaction to finish, because there was another insertion with conflicting key in progress, we would pass a just-free'd item pointer to XactLockTableWait(). All calls to XactLockTableWait() and MultiXactIdWait() had similar issues. Some passed a pointer to a buffer in the buffer cache, after already releasing the lock. The call in EvalPlanQualFetch had already released the pin too. All but the call in execUtils.c would merely lead to reporting a bogus ctid, however (or an assertion failure, if enabled). All the callers that passed HeapTuple->t_data->t_ctid were slightly bogus anyway: if the tuple was updated (again) in the same transaction, its ctid field would point to the next tuple in the chain, not the tuple itself. Backpatch to 9.4, where the 'ctid' argument to XactLockTableWait was added (in commit f88d4cfc)
2015-02-04 15:00:34 +01:00
relation, &tuple->t_self, XLTW_Lock, NULL);
Implement SKIP LOCKED for row-level locks This clause changes the behavior of SELECT locking clauses in the presence of locked rows: instead of causing a process to block waiting for the locks held by other processes (or raise an error, with NOWAIT), SKIP LOCKED makes the new reader skip over such rows. While this is not appropriate behavior for general purposes, there are some cases in which it is useful, such as queue-like tables. Catalog version bumped because this patch changes the representation of stored rules. Reviewed by Craig Ringer (based on a previous attempt at an implementation by Simon Riggs, who also provided input on the syntax used in the current patch), David Rowley, and Álvaro Herrera. Author: Thomas Munro
2014-10-07 22:23:34 +02:00
break;
case LockWaitSkip:
if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
status, infomask, relation,
NULL))
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_WouldBlock;
Implement SKIP LOCKED for row-level locks This clause changes the behavior of SELECT locking clauses in the presence of locked rows: instead of causing a process to block waiting for the locks held by other processes (or raise an error, with NOWAIT), SKIP LOCKED makes the new reader skip over such rows. While this is not appropriate behavior for general purposes, there are some cases in which it is useful, such as queue-like tables. Catalog version bumped because this patch changes the representation of stored rules. Reviewed by Craig Ringer (based on a previous attempt at an implementation by Simon Riggs, who also provided input on the syntax used in the current patch), David Rowley, and Álvaro Herrera. Author: Thomas Munro
2014-10-07 22:23:34 +02:00
/* recovery code expects to have buffer lock held */
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
goto failed;
}
break;
case LockWaitError:
if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
status, infomask, relation,
NULL))
ereport(ERROR,
(errcode(ERRCODE_LOCK_NOT_AVAILABLE),
errmsg("could not obtain lock on row in relation \"%s\"",
RelationGetRelationName(relation))));
break;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
Repair two related errors in heap_lock_tuple: it was failing to recognize cases where we already hold the desired lock "indirectly", either via membership in a MultiXact or because the lock was originally taken by a different subtransaction of the current transaction. These cases must be accounted for to avoid needless deadlocks and/or inappropriate replacement of an exclusive lock with a shared lock. Per report from Clarence Gardner and subsequent investigation.
2006-11-17 19:00:15 +01:00
/*
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* Of course, the multixact might not be done here: if we're
* requesting a light lock mode, other transactions with light
* locks could still be alive, as well as locks owned by our
* own xact or other subxacts of this backend. We need to
* preserve the surviving MultiXact members. Note that it
* isn't absolutely necessary in the latter case, but doing so
* is simpler.
Repair two related errors in heap_lock_tuple: it was failing to recognize cases where we already hold the desired lock "indirectly", either via membership in a MultiXact or because the lock was originally taken by a different subtransaction of the current transaction. These cases must be accounted for to avoid needless deadlocks and/or inappropriate replacement of an exclusive lock with a shared lock. Per report from Clarence Gardner and subsequent investigation.
2006-11-17 19:00:15 +01:00
*/
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
}
else
{
Implement SKIP LOCKED for row-level locks This clause changes the behavior of SELECT locking clauses in the presence of locked rows: instead of causing a process to block waiting for the locks held by other processes (or raise an error, with NOWAIT), SKIP LOCKED makes the new reader skip over such rows. While this is not appropriate behavior for general purposes, there are some cases in which it is useful, such as queue-like tables. Catalog version bumped because this patch changes the representation of stored rules. Reviewed by Craig Ringer (based on a previous attempt at an implementation by Simon Riggs, who also provided input on the syntax used in the current patch), David Rowley, and Álvaro Herrera. Author: Thomas Munro
2014-10-07 22:23:34 +02:00
/* wait for regular transaction to end, or die trying */
switch (wait_policy)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
Implement SKIP LOCKED for row-level locks This clause changes the behavior of SELECT locking clauses in the presence of locked rows: instead of causing a process to block waiting for the locks held by other processes (or raise an error, with NOWAIT), SKIP LOCKED makes the new reader skip over such rows. While this is not appropriate behavior for general purposes, there are some cases in which it is useful, such as queue-like tables. Catalog version bumped because this patch changes the representation of stored rules. Reviewed by Craig Ringer (based on a previous attempt at an implementation by Simon Riggs, who also provided input on the syntax used in the current patch), David Rowley, and Álvaro Herrera. Author: Thomas Munro
2014-10-07 22:23:34 +02:00
case LockWaitBlock:
Fix reference-after-free when waiting for another xact due to constraint. If an insertion or update had to wait for another transaction to finish, because there was another insertion with conflicting key in progress, we would pass a just-free'd item pointer to XactLockTableWait(). All calls to XactLockTableWait() and MultiXactIdWait() had similar issues. Some passed a pointer to a buffer in the buffer cache, after already releasing the lock. The call in EvalPlanQualFetch had already released the pin too. All but the call in execUtils.c would merely lead to reporting a bogus ctid, however (or an assertion failure, if enabled). All the callers that passed HeapTuple->t_data->t_ctid were slightly bogus anyway: if the tuple was updated (again) in the same transaction, its ctid field would point to the next tuple in the chain, not the tuple itself. Backpatch to 9.4, where the 'ctid' argument to XactLockTableWait was added (in commit f88d4cfc)
2015-02-04 15:00:34 +01:00
XactLockTableWait(xwait, relation, &tuple->t_self,
Implement SKIP LOCKED for row-level locks This clause changes the behavior of SELECT locking clauses in the presence of locked rows: instead of causing a process to block waiting for the locks held by other processes (or raise an error, with NOWAIT), SKIP LOCKED makes the new reader skip over such rows. While this is not appropriate behavior for general purposes, there are some cases in which it is useful, such as queue-like tables. Catalog version bumped because this patch changes the representation of stored rules. Reviewed by Craig Ringer (based on a previous attempt at an implementation by Simon Riggs, who also provided input on the syntax used in the current patch), David Rowley, and Álvaro Herrera. Author: Thomas Munro
2014-10-07 22:23:34 +02:00
XLTW_Lock);
break;
case LockWaitSkip:
if (!ConditionalXactLockTableWait(xwait))
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_WouldBlock;
Implement SKIP LOCKED for row-level locks This clause changes the behavior of SELECT locking clauses in the presence of locked rows: instead of causing a process to block waiting for the locks held by other processes (or raise an error, with NOWAIT), SKIP LOCKED makes the new reader skip over such rows. While this is not appropriate behavior for general purposes, there are some cases in which it is useful, such as queue-like tables. Catalog version bumped because this patch changes the representation of stored rules. Reviewed by Craig Ringer (based on a previous attempt at an implementation by Simon Riggs, who also provided input on the syntax used in the current patch), David Rowley, and Álvaro Herrera. Author: Thomas Munro
2014-10-07 22:23:34 +02:00
/* recovery code expects to have buffer lock held */
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
goto failed;
}
break;
case LockWaitError:
if (!ConditionalXactLockTableWait(xwait))
ereport(ERROR,
(errcode(ERRCODE_LOCK_NOT_AVAILABLE),
errmsg("could not obtain lock on row in relation \"%s\"",
RelationGetRelationName(relation))));
break;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
/* if there are updates, follow the update chain */
if (follow_updates && !HEAP_XMAX_IS_LOCKED_ONLY(infomask))
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
TM_Result res;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
res = heap_lock_updated_tuple(relation, tuple, &t_ctid,
GetCurrentTransactionId(),
mode);
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (res != TM_Ok)
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
{
result = res;
/* recovery code expects to have buffer lock held */
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
goto failed;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
/*
* xwait is done, but if xwait had just locked the tuple then some
* other xact could update this tuple before we get to this point.
* Check for xmax change, and start over if so.
*/
if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
!TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
xwait))
goto l3;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
if (!(infomask & HEAP_XMAX_IS_MULTI))
{
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* Otherwise check if it committed or aborted. Note we cannot
* be here if the tuple was only locked by somebody who didn't
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
* conflict with us; that would have been handled above. So
* that transaction must necessarily be gone by now. But
* don't check for this in the multixact case, because some
* locker transactions might still be running.
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
*/
UpdateXmaxHintBits(tuple->t_data, *buffer, xwait);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
}
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/* By here, we're certain that we hold buffer exclusive lock again */
/*
* We may lock if previous xmax aborted, or if it committed but only
* locked the tuple without updating it; or if we didn't have to wait
* at all for whatever reason.
*/
if (!require_sleep ||
(tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
HeapTupleHeaderIsOnlyLocked(tuple->t_data))
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Ok;
Remove pointless HeapTupleHeaderIndicatesMovedPartitions calls Pavan Deolasee recently noted that a few of the HeapTupleHeaderIndicatesMovedPartitions calls added by commit 5db6df0c0117 are useless, since they are done after comparing t_self with t_ctid. But because t_self can never be set to the magical values that indicate that the tuple moved partition, this can never succeed: if the first test fails (so we know t_self equals t_ctid), necessarily the second test will also fail. So these checks can be removed and no harm is done. There's no bug here, just a code legibility issue. Reported-by: Pavan Deolasee <pavan.deolasee@gmail.com> Discussion: https://postgr.es/m/20200929164411.GA15497@alvherre.pgsql
2021-02-22 20:51:44 +01:00
else if (!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Updated;
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
else
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Deleted;
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
failed:
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (result != TM_Ok)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
Assert(result == TM_SelfModified || result == TM_Updated ||
result == TM_Deleted || result == TM_WouldBlock);
Allow special SKIP LOCKED condition in Assert() Under concurrency, it is possible for two sessions to be merrily locking and releasing a tuple and marking it again as HEAP_XMAX_INVALID all the while a third session attempts to lock it, miserably fails at it, and then contemplates life, the universe and everything only to eventually fail an assertion that said bit is not set. Before SKIP LOCKED that was indeed a reasonable expectation, but alas! commit df630b0dd5ea falsified it. This bug is as old as time itself, and even older, if you think time begins with the oldest supported branch. Therefore, backpatch to all supported branches. Author: Simon Riggs <simon.riggs@enterprisedb.com> Discussion: https://postgr.es/m/CANbhV-FeEwMnN8yuMyss7if1ZKjOKfjcgqB26n8pqu1e=q0ebg@mail.gmail.com
2022-01-04 17:01:05 +01:00
/*
* When locking a tuple under LockWaitSkip semantics and we fail with
* TM_WouldBlock above, it's possible for concurrent transactions to
* release the lock and set HEAP_XMAX_INVALID in the meantime. So
* this assert is slightly different from the equivalent one in
* heap_delete and heap_update.
*/
Fix silly mistake in Assert
2022-01-04 17:21:23 +01:00
Assert((result == TM_WouldBlock) ||
!(tuple->t_data->t_infomask & HEAP_XMAX_INVALID));
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
Assert(result != TM_Updated ||
!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid));
tmfd->ctid = tuple->t_data->t_ctid;
tmfd->xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
if (result == TM_SelfModified)
tmfd->cmax = HeapTupleHeaderGetCmax(tuple->t_data);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
else
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
tmfd->cmax = InvalidCommandId;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
goto out_locked;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
Fix hard to hit race condition in heapam's tuple locking code. As mentioned in its commit message, eca0f1db left open a race condition, where a page could be marked all-visible, after the code checked PageIsAllVisible() to pin the VM, but before the page is locked. Plug that hole. Reviewed-By: Robert Haas, Andres Freund Author: Amit Kapila Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-08-05 05:07:16 +02:00
/*
* If we didn't pin the visibility map page and the page has become all
* visible while we were busy locking the buffer, or during some
* subsequent window during which we had it unlocked, we'll have to unlock
* and re-lock, to avoid holding the buffer lock across I/O. That's a bit
* unfortunate, especially since we'll now have to recheck whether the
* tuple has been locked or updated under us, but hopefully it won't
* happen very often.
*/
if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
{
LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
visibilitymap_pin(relation, block, &vmbuffer);
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
goto l3;
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
xmax = HeapTupleHeaderGetRawXmax(tuple->t_data);
old_infomask = tuple->t_data->t_infomask;
/*
* If this is the first possibly-multixact-able operation in the current
* transaction, set my per-backend OldestMemberMXactId setting. We can be
* certain that the transaction will never become a member of any older
* MultiXactIds than that. (We have to do this even if we end up just
* using our own TransactionId below, since some other backend could
* incorporate our XID into a MultiXact immediately afterwards.)
*/
MultiXactIdSetOldestMember();
/*
* Compute the new xmax and infomask to store into the tuple. Note we do
* not modify the tuple just yet, because that would leave it in the wrong
* state if multixact.c elogs.
*/
compute_new_xmax_infomask(xmax, old_infomask, tuple->t_data->t_infomask2,
GetCurrentTransactionId(), mode, false,
&xid, &new_infomask, &new_infomask2);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
START_CRIT_SECTION();
heap' logging
2000-07-03 04:54:21 +02:00
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
/*
* Store transaction information of xact locking the tuple.
*
Combine cmin and cmax fields of HeapTupleHeaders into a single field, by keeping private state in each backend that has inserted and deleted the same tuple during its current top-level transaction. This is sufficient since there is no need to be able to determine the cmin/cmax from any other transaction. This gets us back down to 23-byte headers, removing a penalty paid in 8.0 to support subtransactions. Patch by Heikki Linnakangas, with minor revisions by moi, following a design hashed out awhile back on the pghackers list.
2007-02-09 04:35:35 +01:00
* Note: Cmax is meaningless in this context, so don't set it; this avoids
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* possibly generating a useless combo CID. Moreover, if we're locking a
* previously updated tuple, it's important to preserve the Cmax.
*
* Also reset the HOT UPDATE bit, but only if there's no update; otherwise
* we would break the HOT chain.
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
*/
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
tuple->t_data->t_infomask &= ~HEAP_XMAX_BITS;
tuple->t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
tuple->t_data->t_infomask |= new_infomask;
tuple->t_data->t_infomask2 |= new_infomask2;
if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
HeapTupleHeaderClearHotUpdated(tuple->t_data);
Restructure subtransaction handling to reduce resource consumption, as per recent discussions. Invent SubTransactionIds that are managed like CommandIds (ie, counter is reset at start of each top transaction), and use these instead of TransactionIds to keep track of subtransaction status in those modules that need it. This means that a subtransaction does not need an XID unless it actually inserts/modifies rows in the database. Accordingly, don't assign it an XID nor take a lock on the XID until it tries to do that. This saves a lot of overhead for subtransactions that are only used for error recovery (eg plpgsql exceptions). Also, arrange to release a subtransaction's XID lock as soon as the subtransaction exits, in both the commit and abort cases. This avoids holding many unique locks after a long series of subtransactions. The price is some additional overhead in XactLockTableWait, but that seems acceptable. Finally, restructure the state machine in xact.c to have a more orthogonal set of states for subtransactions.
2004-09-16 18:58:44 +02:00
HeapTupleHeaderSetXmax(tuple->t_data, xid);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* Make sure there is no forward chain link in t_ctid. Note that in the
* cases where the tuple has been updated, we must not overwrite t_ctid,
* because it was set by the updater. Moreover, if the tuple has been
* updated, we need to follow the update chain to lock the new versions of
* the tuple as well.
*/
if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
tuple->t_data->t_ctid = *tid;
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
/* Clear only the all-frozen bit on visibility map if needed */
if (PageIsAllVisible(page) &&
visibilitymap_clear(relation, block, vmbuffer,
VISIBILITYMAP_ALL_FROZEN))
cleared_all_frozen = true;
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
MarkBufferDirty(*buffer);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
/*
* XLOG stuff. You might think that we don't need an XLOG record because
* there is no state change worth restoring after a crash. You would be
* wrong however: we have just written either a TransactionId or a
* MultiXactId that may never have been seen on disk before, and we need
* to make sure that there are XLOG entries covering those ID numbers.
* Else the same IDs might be re-used after a crash, which would be
* disastrous if this page made it to disk before the crash. Essentially
* we have to enforce the WAL log-before-data rule even in this case.
Change WAL-logging scheme for multixacts to be more like regular transaction IDs, rather than like subtrans; in particular, the information now survives a database restart. Per previous discussion, this is essential for PITR log shipping and for 2PC.
2005-06-08 17:50:28 +02:00
* (Also, in a PITR log-shipping or 2PC environment, we have to have XLOG
* entries for everything anyway.)
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
*/
Generalize concept of temporary relations to "relation persistence". This commit replaces pg_class.relistemp with pg_class.relpersistence; and also modifies the RangeVar node type to carry relpersistence rather than istemp. It also removes removes rd_istemp from RelationData and instead performs the correct computation based on relpersistence. For clarity, we add three new macros: RelationNeedsWAL(), RelationUsesLocalBuffers(), and RelationUsesTempNamespace(), so that we can clarify the purpose of each check that previous depended on rd_istemp. This is intended as infrastructure for the upcoming unlogged tables patch, as well as for future possible work on global temporary tables.
2010-12-13 18:34:26 +01:00
if (RelationNeedsWAL(relation))
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
{
xl_heap_lock xlrec;
XLogRecPtr recptr;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogBeginInsert();
XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD);
xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
Fix xl_heap_lock WAL record field's data type. Make xl_heap_lock's infobits_set field of type uint8, not int8. Using int8 isn't appropriate given that the field just holds status bits. This fixes an oversight in commit 0ac5ad5134. In passing rename the nearby TransactionId field to "xmax" to make things consistency with related records, such as xl_heap_lock_updated. Deliberately avoid a bump in XLOG_PAGE_MAGIC. No backpatch, either. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkCd3kOS8b7Rfxw7Mh1_6jvX=Nzo-CWR1VBTiOtVZkWHA@mail.gmail.com
2023-04-11 23:07:54 +02:00
xlrec.xmax = xid;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
xlrec.infobits_set = compute_infobits(new_infomask,
tuple->t_data->t_infomask2);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
xlrec.flags = cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterData((char *) &xlrec, SizeOfHeapLock);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
Introduce replication progress tracking infrastructure. When implementing a replication solution ontop of logical decoding, two related problems exist: * How to safely keep track of replication progress * How to change replication behavior, based on the origin of a row; e.g. to avoid loops in bi-directional replication setups The solution to these problems, as implemented here, consist out of three parts: 1) 'replication origins', which identify nodes in a replication setup. 2) 'replication progress tracking', which remembers, for each replication origin, how far replay has progressed in a efficient and crash safe manner. 3) The ability to filter out changes performed on the behest of a replication origin during logical decoding; this allows complex replication topologies. E.g. by filtering all replayed changes out. Most of this could also be implemented in "userspace", e.g. by inserting additional rows contain origin information, but that ends up being much less efficient and more complicated. We don't want to require various replication solutions to reimplement logic for this independently. The infrastructure is intended to be generic enough to be reusable. This infrastructure also replaces the 'nodeid' infrastructure of commit timestamps. It is intended to provide all the former capabilities, except that there's only 2^16 different origins; but now they integrate with logical decoding. Additionally more functionality is accessible via SQL. Since the commit timestamp infrastructure has also been introduced in 9.5 (commit 73c986add) changing the API is not a problem. For now the number of origins for which the replication progress can be tracked simultaneously is determined by the max_replication_slots GUC. That GUC is not a perfect match to configure this, but there doesn't seem to be sufficient reason to introduce a separate new one. Bumps both catversion and wal page magic. Author: Andres Freund, with contributions from Petr Jelinek and Craig Ringer Reviewed-By: Heikki Linnakangas, Petr Jelinek, Robert Haas, Steve Singer Discussion: 20150216002155.GI15326@awork2.anarazel.de, 20140923182422.GA15776@alap3.anarazel.de, 20131114172632.GE7522@alap2.anarazel.de
2015-04-29 19:30:53 +02:00
/* we don't decode row locks atm, so no need to log the origin */
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
Clean up the use of some page-header-access macros: principally, use SizeOfPageHeaderData instead of sizeof(PageHeaderData) in places where that makes the code clearer, and avoid casting between Page and PageHeader where possible. Zdenek Kotala, with some additional cleanup by Heikki Linnakangas. I did not apply the parts of the proposed patch that would have resulted in slightly changing the on-disk format of hash indexes; it seems to me that's not a win as long as there's any chance of having in-place upgrade for 8.4.
2008-07-13 22:45:47 +02:00
PageSetLSN(page, recptr);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
}
END_CRIT_SECTION();
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Ok;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
out_locked:
Initial MVCC code. New code for locking buffer' context.
1998-12-15 13:47:01 +01:00
LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
out_unlocked:
if (BufferIsValid(vmbuffer))
ReleaseBuffer(vmbuffer);
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
/*
* Don't update the visibility map here. Locking a tuple doesn't change
* visibility info.
*/
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
/*
* Now that we have successfully marked the tuple as locked, we can
* release the lmgr tuple lock, if we had it.
*/
if (have_tuple_lock)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
UnlockTupleTuplock(relation, tid, mode);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
return result;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
/*
* Acquire heavyweight lock on the given tuple, in preparation for acquiring
* its normal, Xmax-based tuple lock.
*
* have_tuple_lock is an input and output parameter: on input, it indicates
* whether the lock has previously been acquired (and this function does
* nothing in that case). If this function returns success, have_tuple_lock
* has been flipped to true.
*
* Returns false if it was unable to obtain the lock; this can only happen if
* wait_policy is Skip.
*/
static bool
heap_acquire_tuplock(Relation relation, ItemPointer tid, LockTupleMode mode,
LockWaitPolicy wait_policy, bool *have_tuple_lock)
{
if (*have_tuple_lock)
return true;
switch (wait_policy)
{
case LockWaitBlock:
LockTupleTuplock(relation, tid, mode);
break;
case LockWaitSkip:
if (!ConditionalLockTupleTuplock(relation, tid, mode))
return false;
break;
case LockWaitError:
if (!ConditionalLockTupleTuplock(relation, tid, mode))
ereport(ERROR,
(errcode(ERRCODE_LOCK_NOT_AVAILABLE),
errmsg("could not obtain lock on row in relation \"%s\"",
RelationGetRelationName(relation))));
break;
}
*have_tuple_lock = true;
return true;
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* Given an original set of Xmax and infomask, and a transaction (identified by
* add_to_xmax) acquiring a new lock of some mode, compute the new Xmax and
* corresponding infomasks to use on the tuple.
*
* Note that this might have side effects such as creating a new MultiXactId.
*
* Most callers will have called HeapTupleSatisfiesUpdate before this function;
* that will have set the HEAP_XMAX_INVALID bit if the xmax was a MultiXactId
* but it was not running anymore. There is a race condition, which is that the
* MultiXactId may have finished since then, but that uncommon case is handled
* either here, or within MultiXactIdExpand.
*
* There is a similar race condition possible when the old xmax was a regular
* TransactionId. We test TransactionIdIsInProgress again just to narrow the
* window, but it's still possible to end up creating an unnecessary
* MultiXactId. Fortunately this is harmless.
*/
static void
compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
uint16 old_infomask2, TransactionId add_to_xmax,
LockTupleMode mode, bool is_update,
TransactionId *result_xmax, uint16 *result_infomask,
uint16 *result_infomask2)
{
TransactionId new_xmax;
uint16 new_infomask,
new_infomask2;
Optimize updating a row that's locked by same xid Updating or locking a row that was already locked by the same transaction under the same Xid caused a MultiXact to be created; but this is unnecessary, because there's no usefulness in being able to differentiate two locks by the same transaction. In particular, if a transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't modify columns of the key, we would dutifully represent the resulting combination as a multixact -- even though a single key-update is sufficient. Optimize the case so that only the strongest of both locks/updates is represented in Xmax. This can save some Xmax's from becoming MultiXacts, which can be a significant optimization. This missed optimization opportunity was spotted by Andres Freund while investigating a bug reported by Oliver Seemann in message CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com and also directly as a performance regression reported by Dong Ye in message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com Reportedly, this patch fixes the performance regression. Since the missing optimization was reported as a significant performance regression from 9.2, backpatch to 9.3. Andres Freund, tweaked by Álvaro Herrera
2013-12-19 20:39:59 +01:00
Assert(TransactionIdIsCurrentTransactionId(add_to_xmax));
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
l5:
new_infomask = 0;
new_infomask2 = 0;
if (old_infomask & HEAP_XMAX_INVALID)
{
/*
* No previous locker; we just insert our own TransactionId.
Optimize updating a row that's locked by same xid Updating or locking a row that was already locked by the same transaction under the same Xid caused a MultiXact to be created; but this is unnecessary, because there's no usefulness in being able to differentiate two locks by the same transaction. In particular, if a transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't modify columns of the key, we would dutifully represent the resulting combination as a multixact -- even though a single key-update is sufficient. Optimize the case so that only the strongest of both locks/updates is represented in Xmax. This can save some Xmax's from becoming MultiXacts, which can be a significant optimization. This missed optimization opportunity was spotted by Andres Freund while investigating a bug reported by Oliver Seemann in message CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com and also directly as a performance regression reported by Dong Ye in message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com Reportedly, this patch fixes the performance regression. Since the missing optimization was reported as a significant performance regression from 9.2, backpatch to 9.3. Andres Freund, tweaked by Álvaro Herrera
2013-12-19 20:39:59 +01:00
*
* Note that it's critical that this case be the first one checked,
* because there are several blocks below that come back to this one
* to implement certain optimizations; old_infomask might contain
* other dirty bits in those cases, but we don't really care.
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
*/
if (is_update)
{
new_xmax = add_to_xmax;
if (mode == LockTupleExclusive)
new_infomask2 |= HEAP_KEYS_UPDATED;
}
else
{
new_infomask |= HEAP_XMAX_LOCK_ONLY;
switch (mode)
{
case LockTupleKeyShare:
new_xmax = add_to_xmax;
new_infomask |= HEAP_XMAX_KEYSHR_LOCK;
break;
case LockTupleShare:
new_xmax = add_to_xmax;
new_infomask |= HEAP_XMAX_SHR_LOCK;
break;
case LockTupleNoKeyExclusive:
new_xmax = add_to_xmax;
new_infomask |= HEAP_XMAX_EXCL_LOCK;
break;
case LockTupleExclusive:
new_xmax = add_to_xmax;
new_infomask |= HEAP_XMAX_EXCL_LOCK;
new_infomask2 |= HEAP_KEYS_UPDATED;
break;
default:
new_xmax = InvalidTransactionId; /* silence compiler */
elog(ERROR, "invalid lock mode");
}
}
}
else if (old_infomask & HEAP_XMAX_IS_MULTI)
{
MultiXactStatus new_status;
/*
* Currently we don't allow XMAX_COMMITTED to be set for multis, so
* cross-check.
*/
Assert(!(old_infomask & HEAP_XMAX_COMMITTED));
/*
* A multixact together with LOCK_ONLY set but neither lock bit set
* (i.e. a pg_upgraded share locked tuple) cannot possibly be running
* anymore. This check is critical for databases upgraded by
* pg_upgrade; both MultiXactIdIsRunning and MultiXactIdExpand assume
* that such multis are never passed.
*/
Fix handling of multixacts predating pg_upgrade After pg_upgrade, it is possible that some tuples' Xmax have multixacts corresponding to the old installation; such multixacts cannot have running members anymore. In many code sites we already know not to read them and clobber them silently, but at least when VACUUM tries to freeze a multixact or determine whether one needs freezing, there's an attempt to resolve it to its member transactions by calling GetMultiXactIdMembers, and if the multixact value is "in the future" with regards to the current valid multixact range, an error like this is raised: ERROR: MultiXactId 123 has not been created yet -- apparent wraparound and vacuuming fails. Per discussion with Andrew Gierth, it is completely bogus to try to resolve multixacts coming from before a pg_upgrade, regardless of where they stand with regards to the current valid multixact range. It's possible to get from under this problem by doing SELECT FOR UPDATE of the problem tuples, but if tables are large, this is slow and tedious, so a more thorough solution is desirable. To fix, we realize that multixacts in xmax created in 9.2 and previous have a specific bit pattern that is never used in 9.3 and later (we already knew this, per comments and infomask tests sprinkled in various places, but we weren't leveraging this knowledge appropriately). Whenever the infomask of the tuple matches that bit pattern, we just ignore the multixact completely as if Xmax wasn't set; or, in the case of tuple freezing, we act as if an unwanted value is set and clobber it without decoding. This guarantees that no errors will be raised, and that the values will be progressively removed until all tables are clean. Most callers of GetMultiXactIdMembers are patched to recognize directly that the value is a removable "empty" multixact and avoid calling GetMultiXactIdMembers altogether. To avoid changing the signature of GetMultiXactIdMembers() in back branches, we keep the "allow_old" boolean flag but rename it to "from_pgupgrade"; if the flag is true, we always return an empty set instead of looking up the multixact. (I suppose we could remove the argument in the master branch, but I chose not to do so in this commit). This was broken all along, but the error-facing message appeared first because of commit 8e9a16ab8f7f and was partially fixed in a25c2b7c4db3. This fix, backpatched all the way back to 9.3, goes approximately in the same direction as a25c2b7c4db3 but should cover all cases. Bug analysis by Andrew Gierth and Álvaro Herrera. A number of public reports match this bug: https://www.postgresql.org/message-id/20140330040029.GY4582@tamriel.snowman.net https://www.postgresql.org/message-id/538F3D70.6080902@publicrelay.com https://www.postgresql.org/message-id/556439CF.7070109@pscs.co.uk https://www.postgresql.org/message-id/SG2PR06MB0760098A111C88E31BD4D96FB3540@SG2PR06MB0760.apcprd06.prod.outlook.com https://www.postgresql.org/message-id/20160615203829.5798.4594@wrigleys.postgresql.org
2016-06-25 00:29:28 +02:00
if (HEAP_LOCKED_UPGRADED(old_infomask))
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
old_infomask &= ~HEAP_XMAX_IS_MULTI;
old_infomask |= HEAP_XMAX_INVALID;
goto l5;
}
/*
* If the XMAX is already a MultiXactId, then we need to expand it to
* include add_to_xmax; but if all the members were lockers and are
* all gone, we can do away with the IS_MULTI bit and just set
* add_to_xmax as the only locker/updater. If all lockers are gone
* and we have an updater that aborted, we can also do without a
* multi.
*
* The cost of doing GetMultiXactIdMembers would be paid by
* MultiXactIdExpand if we weren't to do this, so this check is not
* incurring extra work anyhow.
*/
Avoid uselessly looking up old LOCK_ONLY multixacts Commit 0ac5ad5134f2 removed an optimization in multixact.c that skipped fetching members of MultiXactId that were older than our OldestVisibleMXactId value. The reason this was removed is that it is possible for multixacts that contain updates to be older than that value. However, if the caller is certain that the multi does not contain an update (because the infomask bits say so), it can pass this info down to GetMultiXactIdMembers, enabling it to use the old optimization. Pointed out by Andres Freund in 20131121200517.GM7240@alap2.anarazel.de
2014-07-29 21:41:06 +02:00
if (!MultiXactIdIsRunning(xmax, HEAP_XMAX_IS_LOCKED_ONLY(old_infomask)))
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) ||
Optimize locking a tuple already locked by another subxact Locking and updating the same tuple repeatedly led to some strange multixacts being created which had several subtransactions of the same parent transaction holding locks of the same strength. However, once a subxact of the current transaction holds a lock of a given strength, it's not necessary to acquire the same lock again. This made some coding patterns much slower than required. The fix is twofold. First we change HeapTupleSatisfiesUpdate to return HeapTupleBeingUpdated for the case where the current transaction is already a single-xid locker for the given tuple; it used to return HeapTupleMayBeUpdated for that case. The new logic is simpler, and the change to pgrowlocks is a testament to that: previously we needed to check for the single-xid locker separately in a very ugly way. That test is simpler now. As fallout from the HTSU change, some of its callers need to be amended so that tuple-locked-by-own-transaction is taken into account in the BeingUpdated case rather than the MayBeUpdated case. For many of them there is no difference; but heap_delete() and heap_update now check explicitely and do not grab tuple lock in that case. The HTSU change also means that routine MultiXactHasRunningRemoteMembers introduced in commit 11ac4c73cb895 is no longer necessary and can be removed; the case that used to require it is now handled naturally as result of the changes to heap_delete and heap_update. The second part of the fix to the performance issue is to adjust heap_lock_tuple to avoid the slowness: 1. Previously we checked for the case that our own transaction already held a strong enough lock and returned MayBeUpdated, but only in the multixact case. Now we do it for the plain Xid case as well, which saves having to LockTuple. 2. If the current transaction is the only locker of the tuple (but with a lock not as strong as what we need; otherwise it would have been caught in the check mentioned above), we can skip sleeping on the multixact, and instead go straight to create an updated multixact with the additional lock strength. 3. Most importantly, make sure that both the single-xid-locker case and the multixact-locker case optimization are applied always. We do this by checking both in a single place, rather than them appearing in two separate portions of the routine -- something that is made possible by the HeapTupleSatisfiesUpdate API change. Previously we would only check for the single-xid case when HTSU returned MayBeUpdated, and only checked for the multixact case when HTSU returned BeingUpdated. This was at odds with what HTSU actually returned in one case: if our own transaction was locker in a multixact, it returned MayBeUpdated, so the optimization never applied. This is what led to the large multixacts in the first place. Per bug report #8470 by Oskari Saarenmaa.
2015-04-10 18:47:15 +02:00
!TransactionIdDidCommit(MultiXactIdGetUpdateXid(xmax,
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
old_infomask)))
{
/*
* Reset these bits and restart; otherwise fall through to
* create a new multi below.
*/
old_infomask &= ~HEAP_XMAX_IS_MULTI;
old_infomask |= HEAP_XMAX_INVALID;
goto l5;
}
}
new_status = get_mxact_status_for_lock(mode, is_update);
new_xmax = MultiXactIdExpand((MultiXactId) xmax, add_to_xmax,
new_status);
GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
}
else if (old_infomask & HEAP_XMAX_COMMITTED)
{
/*
* It's a committed update, so we need to preserve him as updater of
* the tuple.
*/
MultiXactStatus status;
MultiXactStatus new_status;
if (old_infomask2 & HEAP_KEYS_UPDATED)
status = MultiXactStatusUpdate;
else
status = MultiXactStatusNoKeyUpdate;
new_status = get_mxact_status_for_lock(mode, is_update);
pgindent run for release 9.3 This is the first run of the Perl-based pgindent script. Also update pgindent instructions.
2013-05-29 22:58:43 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* since it's not running, it's obviously impossible for the old
* updater to be identical to the current one, so we need not check
* for that case as we do in the block above.
*/
new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
}
else if (TransactionIdIsInProgress(xmax))
{
/*
* If the XMAX is a valid, in-progress TransactionId, then we need to
* create a new MultiXactId that includes both the old locker or
* updater and our own TransactionId.
*/
MultiXactStatus new_status;
Optimize updating a row that's locked by same xid Updating or locking a row that was already locked by the same transaction under the same Xid caused a MultiXact to be created; but this is unnecessary, because there's no usefulness in being able to differentiate two locks by the same transaction. In particular, if a transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't modify columns of the key, we would dutifully represent the resulting combination as a multixact -- even though a single key-update is sufficient. Optimize the case so that only the strongest of both locks/updates is represented in Xmax. This can save some Xmax's from becoming MultiXacts, which can be a significant optimization. This missed optimization opportunity was spotted by Andres Freund while investigating a bug reported by Oliver Seemann in message CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com and also directly as a performance regression reported by Dong Ye in message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com Reportedly, this patch fixes the performance regression. Since the missing optimization was reported as a significant performance regression from 9.2, backpatch to 9.3. Andres Freund, tweaked by Álvaro Herrera
2013-12-19 20:39:59 +01:00
MultiXactStatus old_status;
LockTupleMode old_mode;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
{
if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
Optimize updating a row that's locked by same xid Updating or locking a row that was already locked by the same transaction under the same Xid caused a MultiXact to be created; but this is unnecessary, because there's no usefulness in being able to differentiate two locks by the same transaction. In particular, if a transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't modify columns of the key, we would dutifully represent the resulting combination as a multixact -- even though a single key-update is sufficient. Optimize the case so that only the strongest of both locks/updates is represented in Xmax. This can save some Xmax's from becoming MultiXacts, which can be a significant optimization. This missed optimization opportunity was spotted by Andres Freund while investigating a bug reported by Oliver Seemann in message CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com and also directly as a performance regression reported by Dong Ye in message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com Reportedly, this patch fixes the performance regression. Since the missing optimization was reported as a significant performance regression from 9.2, backpatch to 9.3. Andres Freund, tweaked by Álvaro Herrera
2013-12-19 20:39:59 +01:00
old_status = MultiXactStatusForKeyShare;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
Optimize updating a row that's locked by same xid Updating or locking a row that was already locked by the same transaction under the same Xid caused a MultiXact to be created; but this is unnecessary, because there's no usefulness in being able to differentiate two locks by the same transaction. In particular, if a transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't modify columns of the key, we would dutifully represent the resulting combination as a multixact -- even though a single key-update is sufficient. Optimize the case so that only the strongest of both locks/updates is represented in Xmax. This can save some Xmax's from becoming MultiXacts, which can be a significant optimization. This missed optimization opportunity was spotted by Andres Freund while investigating a bug reported by Oliver Seemann in message CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com and also directly as a performance regression reported by Dong Ye in message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com Reportedly, this patch fixes the performance regression. Since the missing optimization was reported as a significant performance regression from 9.2, backpatch to 9.3. Andres Freund, tweaked by Álvaro Herrera
2013-12-19 20:39:59 +01:00
old_status = MultiXactStatusForShare;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
{
if (old_infomask2 & HEAP_KEYS_UPDATED)
Optimize updating a row that's locked by same xid Updating or locking a row that was already locked by the same transaction under the same Xid caused a MultiXact to be created; but this is unnecessary, because there's no usefulness in being able to differentiate two locks by the same transaction. In particular, if a transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't modify columns of the key, we would dutifully represent the resulting combination as a multixact -- even though a single key-update is sufficient. Optimize the case so that only the strongest of both locks/updates is represented in Xmax. This can save some Xmax's from becoming MultiXacts, which can be a significant optimization. This missed optimization opportunity was spotted by Andres Freund while investigating a bug reported by Oliver Seemann in message CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com and also directly as a performance regression reported by Dong Ye in message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com Reportedly, this patch fixes the performance regression. Since the missing optimization was reported as a significant performance regression from 9.2, backpatch to 9.3. Andres Freund, tweaked by Álvaro Herrera
2013-12-19 20:39:59 +01:00
old_status = MultiXactStatusForUpdate;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
else
Optimize updating a row that's locked by same xid Updating or locking a row that was already locked by the same transaction under the same Xid caused a MultiXact to be created; but this is unnecessary, because there's no usefulness in being able to differentiate two locks by the same transaction. In particular, if a transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't modify columns of the key, we would dutifully represent the resulting combination as a multixact -- even though a single key-update is sufficient. Optimize the case so that only the strongest of both locks/updates is represented in Xmax. This can save some Xmax's from becoming MultiXacts, which can be a significant optimization. This missed optimization opportunity was spotted by Andres Freund while investigating a bug reported by Oliver Seemann in message CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com and also directly as a performance regression reported by Dong Ye in message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com Reportedly, this patch fixes the performance regression. Since the missing optimization was reported as a significant performance regression from 9.2, backpatch to 9.3. Andres Freund, tweaked by Álvaro Herrera
2013-12-19 20:39:59 +01:00
old_status = MultiXactStatusForNoKeyUpdate;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
else
{
/*
* LOCK_ONLY can be present alone only when a page has been
* upgraded by pg_upgrade. But in that case,
* TransactionIdIsInProgress() should have returned false. We
* assume it's no longer locked in this case.
*/
elog(WARNING, "LOCK_ONLY found for Xid in progress %u", xmax);
old_infomask |= HEAP_XMAX_INVALID;
old_infomask &= ~HEAP_XMAX_LOCK_ONLY;
goto l5;
}
}
else
{
/* it's an update, but which kind? */
if (old_infomask2 & HEAP_KEYS_UPDATED)
Optimize updating a row that's locked by same xid Updating or locking a row that was already locked by the same transaction under the same Xid caused a MultiXact to be created; but this is unnecessary, because there's no usefulness in being able to differentiate two locks by the same transaction. In particular, if a transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't modify columns of the key, we would dutifully represent the resulting combination as a multixact -- even though a single key-update is sufficient. Optimize the case so that only the strongest of both locks/updates is represented in Xmax. This can save some Xmax's from becoming MultiXacts, which can be a significant optimization. This missed optimization opportunity was spotted by Andres Freund while investigating a bug reported by Oliver Seemann in message CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com and also directly as a performance regression reported by Dong Ye in message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com Reportedly, this patch fixes the performance regression. Since the missing optimization was reported as a significant performance regression from 9.2, backpatch to 9.3. Andres Freund, tweaked by Álvaro Herrera
2013-12-19 20:39:59 +01:00
old_status = MultiXactStatusUpdate;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
else
Optimize updating a row that's locked by same xid Updating or locking a row that was already locked by the same transaction under the same Xid caused a MultiXact to be created; but this is unnecessary, because there's no usefulness in being able to differentiate two locks by the same transaction. In particular, if a transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't modify columns of the key, we would dutifully represent the resulting combination as a multixact -- even though a single key-update is sufficient. Optimize the case so that only the strongest of both locks/updates is represented in Xmax. This can save some Xmax's from becoming MultiXacts, which can be a significant optimization. This missed optimization opportunity was spotted by Andres Freund while investigating a bug reported by Oliver Seemann in message CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com and also directly as a performance regression reported by Dong Ye in message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com Reportedly, this patch fixes the performance regression. Since the missing optimization was reported as a significant performance regression from 9.2, backpatch to 9.3. Andres Freund, tweaked by Álvaro Herrera
2013-12-19 20:39:59 +01:00
old_status = MultiXactStatusNoKeyUpdate;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
Optimize updating a row that's locked by same xid Updating or locking a row that was already locked by the same transaction under the same Xid caused a MultiXact to be created; but this is unnecessary, because there's no usefulness in being able to differentiate two locks by the same transaction. In particular, if a transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't modify columns of the key, we would dutifully represent the resulting combination as a multixact -- even though a single key-update is sufficient. Optimize the case so that only the strongest of both locks/updates is represented in Xmax. This can save some Xmax's from becoming MultiXacts, which can be a significant optimization. This missed optimization opportunity was spotted by Andres Freund while investigating a bug reported by Oliver Seemann in message CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com and also directly as a performance regression reported by Dong Ye in message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com Reportedly, this patch fixes the performance regression. Since the missing optimization was reported as a significant performance regression from 9.2, backpatch to 9.3. Andres Freund, tweaked by Álvaro Herrera
2013-12-19 20:39:59 +01:00
old_mode = TUPLOCK_from_mxstatus(old_status);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
Optimize updating a row that's locked by same xid Updating or locking a row that was already locked by the same transaction under the same Xid caused a MultiXact to be created; but this is unnecessary, because there's no usefulness in being able to differentiate two locks by the same transaction. In particular, if a transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't modify columns of the key, we would dutifully represent the resulting combination as a multixact -- even though a single key-update is sufficient. Optimize the case so that only the strongest of both locks/updates is represented in Xmax. This can save some Xmax's from becoming MultiXacts, which can be a significant optimization. This missed optimization opportunity was spotted by Andres Freund while investigating a bug reported by Oliver Seemann in message CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com and also directly as a performance regression reported by Dong Ye in message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com Reportedly, this patch fixes the performance regression. Since the missing optimization was reported as a significant performance regression from 9.2, backpatch to 9.3. Andres Freund, tweaked by Álvaro Herrera
2013-12-19 20:39:59 +01:00
* If the lock to be acquired is for the same TransactionId as the
* existing lock, there's an optimization possible: consider only the
* strongest of both locks as the only one present, and restart.
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
*/
if (xmax == add_to_xmax)
{
/*
Optimize updating a row that's locked by same xid Updating or locking a row that was already locked by the same transaction under the same Xid caused a MultiXact to be created; but this is unnecessary, because there's no usefulness in being able to differentiate two locks by the same transaction. In particular, if a transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't modify columns of the key, we would dutifully represent the resulting combination as a multixact -- even though a single key-update is sufficient. Optimize the case so that only the strongest of both locks/updates is represented in Xmax. This can save some Xmax's from becoming MultiXacts, which can be a significant optimization. This missed optimization opportunity was spotted by Andres Freund while investigating a bug reported by Oliver Seemann in message CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com and also directly as a performance regression reported by Dong Ye in message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com Reportedly, this patch fixes the performance regression. Since the missing optimization was reported as a significant performance regression from 9.2, backpatch to 9.3. Andres Freund, tweaked by Álvaro Herrera
2013-12-19 20:39:59 +01:00
* Note that it's not possible for the original tuple to be
* updated: we wouldn't be here because the tuple would have been
* invisible and we wouldn't try to update it. As a subtlety,
* this code can also run when traversing an update chain to lock
* future versions of a tuple. But we wouldn't be here either,
* because the add_to_xmax would be different from the original
* updater.
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
*/
Optimize updating a row that's locked by same xid Updating or locking a row that was already locked by the same transaction under the same Xid caused a MultiXact to be created; but this is unnecessary, because there's no usefulness in being able to differentiate two locks by the same transaction. In particular, if a transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't modify columns of the key, we would dutifully represent the resulting combination as a multixact -- even though a single key-update is sufficient. Optimize the case so that only the strongest of both locks/updates is represented in Xmax. This can save some Xmax's from becoming MultiXacts, which can be a significant optimization. This missed optimization opportunity was spotted by Andres Freund while investigating a bug reported by Oliver Seemann in message CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com and also directly as a performance regression reported by Dong Ye in message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com Reportedly, this patch fixes the performance regression. Since the missing optimization was reported as a significant performance regression from 9.2, backpatch to 9.3. Andres Freund, tweaked by Álvaro Herrera
2013-12-19 20:39:59 +01:00
Assert(HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
/* acquire the strongest of both */
if (mode < old_mode)
mode = old_mode;
/* mustn't touch is_update */
old_infomask |= HEAP_XMAX_INVALID;
goto l5;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
Optimize updating a row that's locked by same xid Updating or locking a row that was already locked by the same transaction under the same Xid caused a MultiXact to be created; but this is unnecessary, because there's no usefulness in being able to differentiate two locks by the same transaction. In particular, if a transaction executed SELECT FOR UPDATE followed by an UPDATE that didn't modify columns of the key, we would dutifully represent the resulting combination as a multixact -- even though a single key-update is sufficient. Optimize the case so that only the strongest of both locks/updates is represented in Xmax. This can save some Xmax's from becoming MultiXacts, which can be a significant optimization. This missed optimization opportunity was spotted by Andres Freund while investigating a bug reported by Oliver Seemann in message CANCipfpfzoYnOz5jj=UZ70_R=CwDHv36dqWSpwsi27vpm1z5sA@mail.gmail.com and also directly as a performance regression reported by Dong Ye in message d54b8387.000012d8.00000010@YED-DEVD1.vmware.com Reportedly, this patch fixes the performance regression. Since the missing optimization was reported as a significant performance regression from 9.2, backpatch to 9.3. Andres Freund, tweaked by Álvaro Herrera
2013-12-19 20:39:59 +01:00
/* otherwise, just fall back to creating a new multixact */
new_status = get_mxact_status_for_lock(mode, is_update);
new_xmax = MultiXactIdCreate(xmax, old_status,
add_to_xmax, new_status);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
}
else if (!HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) &&
TransactionIdDidCommit(xmax))
{
/*
* It's a committed update, so we gotta preserve him as updater of the
* tuple.
*/
MultiXactStatus status;
MultiXactStatus new_status;
if (old_infomask2 & HEAP_KEYS_UPDATED)
status = MultiXactStatusUpdate;
else
status = MultiXactStatusNoKeyUpdate;
new_status = get_mxact_status_for_lock(mode, is_update);
pgindent run for release 9.3 This is the first run of the Perl-based pgindent script. Also update pgindent instructions.
2013-05-29 22:58:43 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* since it's not running, it's obviously impossible for the old
* updater to be identical to the current one, so we need not check
* for that case as we do in the block above.
*/
new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
}
else
{
/*
* Can get here iff the locking/updating transaction was running when
* the infomask was extracted from the tuple, but finished before
* TransactionIdIsInProgress got to run. Deal with it as if there was
* no locker at all in the first place.
*/
old_infomask |= HEAP_XMAX_INVALID;
goto l5;
}
*result_infomask = new_infomask;
*result_infomask2 = new_infomask2;
*result_xmax = new_xmax;
}
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
/*
* Subroutine for heap_lock_updated_tuple_rec.
*
Collection of typo fixes. Use "a" and "an" correctly, mostly in comments. Two error messages were also fixed (they were just elogs, so no translation work required). Two function comments in pg_proc.h were also fixed. Etsuro Fujita reported one of these, but I found a lot more with grep. Also fix a few other typos spotted while grepping for the a/an typos. For example, "consists out of ..." -> "consists of ...". Plus a "though"/ "through" mixup reported by Euler Taveira. Many of these typos were in old code, which would be nice to backpatch to make future backpatching easier. But much of the code was new, and I didn't feel like crafting separate patches for each branch. So no backpatching.
2015-05-20 15:18:11 +02:00
* Given a hypothetical multixact status held by the transaction identified
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
* with the given xid, does the current transaction need to wait, fail, or can
* it continue if it wanted to acquire a lock of the given mode? "needwait"
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
* is set to true if waiting is necessary; if it can continue, then TM_Ok is
* returned. If the lock is already held by the current transaction, return
* TM_SelfModified. In case of a conflict with another transaction, a
* different HeapTupleSatisfiesUpdate return code is returned.
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
*
* The held status is said to be hypothetical because it might correspond to a
* lock held by a single Xid, i.e. not a real MultiXactId; we express it this
* way for simplicity of API.
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
static TM_Result
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
test_lockmode_for_conflict(MultiXactStatus status, TransactionId xid,
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
LockTupleMode mode, HeapTuple tup,
bool *needwait)
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
{
MultiXactStatus wantedstatus;
*needwait = false;
wantedstatus = get_mxact_status_for_lock(mode, false);
/*
* Note: we *must* check TransactionIdIsInProgress before
Move remaining code from tqual.[ch] to heapam.h / heapam_visibility.c. Given these routines are heap specific, and that there will be more generic visibility support in via table AM, it makes sense to move the prototypes to heapam.h (routines like HeapTupleSatisfiesVacuum will not be exposed in a generic fashion, because they are too storage specific). Similarly, the code in tqual.c is specific to heap, so moving it into access/heap/ makes sense. Author: Andres Freund Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-01-22 02:03:15 +01:00
* TransactionIdDidAbort/Commit; see comment at top of heapam_visibility.c
* for an explanation.
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
*/
if (TransactionIdIsCurrentTransactionId(xid))
{
/*
Fix concurrent locking of tuple update chain If several sessions are concurrently locking a tuple update chain with nonconflicting lock modes using an old snapshot, and they all succeed, it may happen that some of them fail because of restarting the loop (due to a concurrent Xmax change) and getting an error in the subsequent pass while trying to obtain a tuple lock that they already have in some tuple version. This can only happen with very high concurrency (where a row is being both updated and FK-checked by multiple transactions concurrently), but it's been observed in the field and can have unpleasant consequences such as an FK check failing to see a tuple that definitely exists: ERROR: insert or update on table "child_table" violates foreign key constraint "fk_constraint_name" DETAIL: Key (keyid)=(123456) is not present in table "parent_table". (where the key is observably present in the table). Discussion: https://postgr.es/m/20170714210011.r25mrff4nxjhmf3g@alvherre.pgsql
2017-07-26 23:24:16 +02:00
* The tuple has already been locked by our own transaction. This is
* very rare but can happen if multiple transactions are trying to
* lock an ancient version of the same tuple.
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
return TM_SelfModified;
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
}
else if (TransactionIdIsInProgress(xid))
{
/*
* If the locking transaction is running, what we do depends on
* whether the lock modes conflict: if they do, then we must wait for
* it to finish; otherwise we can fall through to lock this tuple
* version without waiting.
*/
if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
LOCKMODE_from_mxstatus(wantedstatus)))
{
*needwait = true;
}
/*
* If we set needwait above, then this value doesn't matter;
* otherwise, this value signals to caller that it's okay to proceed.
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
return TM_Ok;
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
}
else if (TransactionIdDidAbort(xid))
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
return TM_Ok;
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
else if (TransactionIdDidCommit(xid))
{
/*
Fix improper abort during update chain locking In 247c76a98909, I added some code to do fine-grained checking of MultiXact status of locking/updating transactions when traversing an update chain. There was a thinko in that patch which would have the traversing abort, that is return HeapTupleUpdated, when the other transaction is a committed lock-only. In this case we should ignore it and return success instead. Of course, in the case where there is a committed update, HeapTupleUpdated is the correct return value. A user-visible symptom of this bug is that in REPEATABLE READ and SERIALIZABLE transaction isolation modes spurious serializability errors can occur: ERROR: could not serialize access due to concurrent update In order for this to happen, there needs to be a tuple that's key-share- locked and also updated, and the update must abort; a subsequent transaction trying to acquire a new lock on that tuple would abort with the above error. The reason is that the initial FOR KEY SHARE is seen as committed by the new locking transaction, which triggers this bug. (If the UPDATE commits, then the serialization error is correctly reported.) When running a query in READ COMMITTED mode, what happens is that the locking is aborted by the HeapTupleUpdated return value, then EvalPlanQual fetches the newest version of the tuple, which is then the only version that gets locked. (The second time the tuple is checked there is no misbehavior on the committed lock-only, because it's not checked by the code that traverses update chains; so no bug.) Only the newest version of the tuple is locked, not older ones, but this is harmless. The isolation test added by this commit illustrates the desired behavior, including the proper serialization errors that get thrown. Backpatch to 9.3.
2013-12-05 21:47:51 +01:00
* The other transaction committed. If it was only a locker, then the
* lock is completely gone now and we can return success; but if it
* was an update, then what we do depends on whether the two lock
* modes conflict. If they conflict, then we must report error to
* caller. But if they don't, we can fall through to allow the current
* transaction to lock the tuple.
*
* Note: the reason we worry about ISUPDATE here is because as soon as
* a transaction ends, all its locks are gone and meaningless, and
* thus we can ignore them; whereas its updates persist. In the
* TransactionIdIsInProgress case, above, we don't need to check
* because we know the lock is still "alive" and thus a conflict needs
* always be checked.
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
*/
Fix improper abort during update chain locking In 247c76a98909, I added some code to do fine-grained checking of MultiXact status of locking/updating transactions when traversing an update chain. There was a thinko in that patch which would have the traversing abort, that is return HeapTupleUpdated, when the other transaction is a committed lock-only. In this case we should ignore it and return success instead. Of course, in the case where there is a committed update, HeapTupleUpdated is the correct return value. A user-visible symptom of this bug is that in REPEATABLE READ and SERIALIZABLE transaction isolation modes spurious serializability errors can occur: ERROR: could not serialize access due to concurrent update In order for this to happen, there needs to be a tuple that's key-share- locked and also updated, and the update must abort; a subsequent transaction trying to acquire a new lock on that tuple would abort with the above error. The reason is that the initial FOR KEY SHARE is seen as committed by the new locking transaction, which triggers this bug. (If the UPDATE commits, then the serialization error is correctly reported.) When running a query in READ COMMITTED mode, what happens is that the locking is aborted by the HeapTupleUpdated return value, then EvalPlanQual fetches the newest version of the tuple, which is then the only version that gets locked. (The second time the tuple is checked there is no misbehavior on the committed lock-only, because it's not checked by the code that traverses update chains; so no bug.) Only the newest version of the tuple is locked, not older ones, but this is harmless. The isolation test added by this commit illustrates the desired behavior, including the proper serialization errors that get thrown. Backpatch to 9.3.
2013-12-05 21:47:51 +01:00
if (!ISUPDATE_from_mxstatus(status))
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
return TM_Ok;
Fix improper abort during update chain locking In 247c76a98909, I added some code to do fine-grained checking of MultiXact status of locking/updating transactions when traversing an update chain. There was a thinko in that patch which would have the traversing abort, that is return HeapTupleUpdated, when the other transaction is a committed lock-only. In this case we should ignore it and return success instead. Of course, in the case where there is a committed update, HeapTupleUpdated is the correct return value. A user-visible symptom of this bug is that in REPEATABLE READ and SERIALIZABLE transaction isolation modes spurious serializability errors can occur: ERROR: could not serialize access due to concurrent update In order for this to happen, there needs to be a tuple that's key-share- locked and also updated, and the update must abort; a subsequent transaction trying to acquire a new lock on that tuple would abort with the above error. The reason is that the initial FOR KEY SHARE is seen as committed by the new locking transaction, which triggers this bug. (If the UPDATE commits, then the serialization error is correctly reported.) When running a query in READ COMMITTED mode, what happens is that the locking is aborted by the HeapTupleUpdated return value, then EvalPlanQual fetches the newest version of the tuple, which is then the only version that gets locked. (The second time the tuple is checked there is no misbehavior on the committed lock-only, because it's not checked by the code that traverses update chains; so no bug.) Only the newest version of the tuple is locked, not older ones, but this is harmless. The isolation test added by this commit illustrates the desired behavior, including the proper serialization errors that get thrown. Backpatch to 9.3.
2013-12-05 21:47:51 +01:00
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
LOCKMODE_from_mxstatus(wantedstatus)))
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
{
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
/* bummer */
Remove pointless HeapTupleHeaderIndicatesMovedPartitions calls Pavan Deolasee recently noted that a few of the HeapTupleHeaderIndicatesMovedPartitions calls added by commit 5db6df0c0117 are useless, since they are done after comparing t_self with t_ctid. But because t_self can never be set to the magical values that indicate that the tuple moved partition, this can never succeed: if the first test fails (so we know t_self equals t_ctid), necessarily the second test will also fail. So these checks can be removed and no harm is done. There's no bug here, just a code legibility issue. Reported-by: Pavan Deolasee <pavan.deolasee@gmail.com> Discussion: https://postgr.es/m/20200929164411.GA15497@alvherre.pgsql
2021-02-22 20:51:44 +01:00
if (!ItemPointerEquals(&tup->t_self, &tup->t_data->t_ctid))
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
return TM_Updated;
else
return TM_Deleted;
}
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
return TM_Ok;
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
}
/* Not in progress, not aborted, not committed -- must have crashed */
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
return TM_Ok;
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* Recursive part of heap_lock_updated_tuple
*
* Fetch the tuple pointed to by tid in rel, and mark it as locked by the given
* xid with the given mode; if this tuple is updated, recurse to lock the new
* version as well.
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
static TM_Result
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
heap_lock_updated_tuple_rec(Relation rel, ItemPointer tid, TransactionId xid,
LockTupleMode mode)
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
TM_Result result;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
ItemPointerData tupid;
HeapTupleData mytup;
Buffer buf;
uint16 new_infomask,
new_infomask2,
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
old_infomask,
old_infomask2;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
TransactionId xmax,
new_xmax;
Compare Xmin to previous Xmax when locking an update chain Not doing so causes us to traverse an update chain that has been broken by concurrent page pruning. All other code that traverses update chains uses this check as one of the cases in which to stop iterating, so replicate it here too. Failure to do so leads to erroneous CLOG, subtrans or multixact lookups. Per discussion following the bug report by J Smith in CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com as diagnosed by Andres Freund.
2013-11-27 21:49:12 +01:00
TransactionId priorXmax = InvalidTransactionId;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
bool cleared_all_frozen = false;
Fix VM buffer pin management in heap_lock_updated_tuple_rec(). Sloppy coding in this function could lead to leaking a VM buffer pin, or to attempting to free the same pin twice. Repair. While at it, reduce the code's tendency to free and reacquire the same page pin. Back-patch to 9.6; before that, this routine did not concern itself with VM pages. Amit Kapila and Tom Lane Discussion: https://postgr.es/m/CAA4eK1KJKwhc=isgTQHjM76CAdVswzNeAuZkh_cx-6QgGkSEgA@mail.gmail.com
2018-03-02 23:40:48 +01:00
bool pinned_desired_page;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
Buffer vmbuffer = InvalidBuffer;
BlockNumber block;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
ItemPointerCopy(tid, &tupid);
for (;;)
{
new_infomask = 0;
new_xmax = InvalidTransactionId;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
block = ItemPointerGetBlockNumber(&tupid);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
ItemPointerCopy(&tupid, &(mytup.t_self));
Prevent access to no-longer-pinned buffer in heapam_tuple_lock(). heap_fetch() used to have a "keep_buf" parameter that told it to return ownership of the buffer pin to the caller after finding that the requested tuple TID exists but is invisible to the specified snapshot. This was thoughtlessly removed in commit 5db6df0c0, which broke heapam_tuple_lock() (formerly EvalPlanQualFetch) because that function needs to do more accesses to the tuple even if it's invisible. The net effect is that we would continue to touch the page for a microsecond or two after releasing pin on the buffer. Usually no harm would result; but if a different session decided to defragment the page concurrently, we could see garbage data and mistakenly conclude that there's no newer tuple version to chain up to. (It's hard to say whether this has happened in the field. The bug was actually found thanks to a later change that allowed valgrind to detect accesses to non-pinned buffers.) The most reasonable way to fix this is to reintroduce keep_buf, although I made it behave slightly differently: buffer ownership is passed back only if there is a valid tuple at the requested TID. In HEAD, we can just add the parameter back to heap_fetch(). To avoid an API break in the back branches, introduce an additional function heap_fetch_extended() in those branches. In HEAD there is an additional, less obvious API change: tuple->t_data will be set to NULL in all cases where buffer ownership is not returned, in particular when the tuple exists but fails the time qual (and !keep_buf). This is to defend against any other callers attempting to access non-pinned buffers. We concluded that making that change in back branches would be more likely to introduce problems than cure any. In passing, remove a comment about heap_fetch that was obsoleted by 9a8ee1dc6. Per bug #17462 from Daniil Anisimov. Back-patch to v12 where the bug was introduced. Discussion: https://postgr.es/m/17462-9c98a0f00df9bd36@postgresql.org
2022-04-13 19:35:02 +02:00
if (!heap_fetch(rel, SnapshotAny, &mytup, &buf, false))
Cope with heap_fetch failure while locking an update chain The reason for the fetch failure is that the tuple was removed because it was dead; so the failure is innocuous and can be ignored. Moreover, there's no need for further work and we can return success to the caller immediately. EvalPlanQualFetch is doing something very similar to this already. Report and test case from Andres Freund in 20131124000203.GA4403@alap2.anarazel.de
2013-11-27 21:45:25 +01:00
{
/*
* if we fail to find the updated version of the tuple, it's
* because it was vacuumed/pruned away after its creator
* transaction aborted. So behave as if we got to the end of the
* chain, and there's no further tuple to lock: return success to
* caller.
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Ok;
Fix VM buffer pin management in heap_lock_updated_tuple_rec(). Sloppy coding in this function could lead to leaking a VM buffer pin, or to attempting to free the same pin twice. Repair. While at it, reduce the code's tendency to free and reacquire the same page pin. Back-patch to 9.6; before that, this routine did not concern itself with VM pages. Amit Kapila and Tom Lane Discussion: https://postgr.es/m/CAA4eK1KJKwhc=isgTQHjM76CAdVswzNeAuZkh_cx-6QgGkSEgA@mail.gmail.com
2018-03-02 23:40:48 +01:00
goto out_unlocked;
Cope with heap_fetch failure while locking an update chain The reason for the fetch failure is that the tuple was removed because it was dead; so the failure is innocuous and can be ignored. Moreover, there's no need for further work and we can return success to the caller immediately. EvalPlanQualFetch is doing something very similar to this already. Report and test case from Andres Freund in 20131124000203.GA4403@alap2.anarazel.de
2013-11-27 21:45:25 +01:00
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
l4:
CHECK_FOR_INTERRUPTS();
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
/*
Fix hard to hit race condition in heapam's tuple locking code. As mentioned in its commit message, eca0f1db left open a race condition, where a page could be marked all-visible, after the code checked PageIsAllVisible() to pin the VM, but before the page is locked. Plug that hole. Reviewed-By: Robert Haas, Andres Freund Author: Amit Kapila Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-08-05 05:07:16 +02:00
* Before locking the buffer, pin the visibility map page if it
* appears to be necessary. Since we haven't got the lock yet,
* someone else might be in the middle of changing this, so we'll need
* to recheck after we have the lock.
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
*/
if (PageIsAllVisible(BufferGetPage(buf)))
Fix VM buffer pin management in heap_lock_updated_tuple_rec(). Sloppy coding in this function could lead to leaking a VM buffer pin, or to attempting to free the same pin twice. Repair. While at it, reduce the code's tendency to free and reacquire the same page pin. Back-patch to 9.6; before that, this routine did not concern itself with VM pages. Amit Kapila and Tom Lane Discussion: https://postgr.es/m/CAA4eK1KJKwhc=isgTQHjM76CAdVswzNeAuZkh_cx-6QgGkSEgA@mail.gmail.com
2018-03-02 23:40:48 +01:00
{
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
visibilitymap_pin(rel, block, &vmbuffer);
Fix VM buffer pin management in heap_lock_updated_tuple_rec(). Sloppy coding in this function could lead to leaking a VM buffer pin, or to attempting to free the same pin twice. Repair. While at it, reduce the code's tendency to free and reacquire the same page pin. Back-patch to 9.6; before that, this routine did not concern itself with VM pages. Amit Kapila and Tom Lane Discussion: https://postgr.es/m/CAA4eK1KJKwhc=isgTQHjM76CAdVswzNeAuZkh_cx-6QgGkSEgA@mail.gmail.com
2018-03-02 23:40:48 +01:00
pinned_desired_page = true;
}
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
else
Fix VM buffer pin management in heap_lock_updated_tuple_rec(). Sloppy coding in this function could lead to leaking a VM buffer pin, or to attempting to free the same pin twice. Repair. While at it, reduce the code's tendency to free and reacquire the same page pin. Back-patch to 9.6; before that, this routine did not concern itself with VM pages. Amit Kapila and Tom Lane Discussion: https://postgr.es/m/CAA4eK1KJKwhc=isgTQHjM76CAdVswzNeAuZkh_cx-6QgGkSEgA@mail.gmail.com
2018-03-02 23:40:48 +01:00
pinned_desired_page = false;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
Fix hard to hit race condition in heapam's tuple locking code. As mentioned in its commit message, eca0f1db left open a race condition, where a page could be marked all-visible, after the code checked PageIsAllVisible() to pin the VM, but before the page is locked. Plug that hole. Reviewed-By: Robert Haas, Andres Freund Author: Amit Kapila Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-08-05 05:07:16 +02:00
/*
* If we didn't pin the visibility map page and the page has become
* all visible while we were busy locking the buffer, we'll have to
* unlock and re-lock, to avoid holding the buffer lock across I/O.
* That's a bit unfortunate, but hopefully shouldn't happen often.
Fix VM buffer pin management in heap_lock_updated_tuple_rec(). Sloppy coding in this function could lead to leaking a VM buffer pin, or to attempting to free the same pin twice. Repair. While at it, reduce the code's tendency to free and reacquire the same page pin. Back-patch to 9.6; before that, this routine did not concern itself with VM pages. Amit Kapila and Tom Lane Discussion: https://postgr.es/m/CAA4eK1KJKwhc=isgTQHjM76CAdVswzNeAuZkh_cx-6QgGkSEgA@mail.gmail.com
2018-03-02 23:40:48 +01:00
*
* Note: in some paths through this function, we will reach here
* holding a pin on a vm page that may or may not be the one matching
* this page. If this page isn't all-visible, we won't use the vm
* page, but we hold onto such a pin till the end of the function.
Fix hard to hit race condition in heapam's tuple locking code. As mentioned in its commit message, eca0f1db left open a race condition, where a page could be marked all-visible, after the code checked PageIsAllVisible() to pin the VM, but before the page is locked. Plug that hole. Reviewed-By: Robert Haas, Andres Freund Author: Amit Kapila Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-08-05 05:07:16 +02:00
*/
Fix VM buffer pin management in heap_lock_updated_tuple_rec(). Sloppy coding in this function could lead to leaking a VM buffer pin, or to attempting to free the same pin twice. Repair. While at it, reduce the code's tendency to free and reacquire the same page pin. Back-patch to 9.6; before that, this routine did not concern itself with VM pages. Amit Kapila and Tom Lane Discussion: https://postgr.es/m/CAA4eK1KJKwhc=isgTQHjM76CAdVswzNeAuZkh_cx-6QgGkSEgA@mail.gmail.com
2018-03-02 23:40:48 +01:00
if (!pinned_desired_page && PageIsAllVisible(BufferGetPage(buf)))
Fix hard to hit race condition in heapam's tuple locking code. As mentioned in its commit message, eca0f1db left open a race condition, where a page could be marked all-visible, after the code checked PageIsAllVisible() to pin the VM, but before the page is locked. Plug that hole. Reviewed-By: Robert Haas, Andres Freund Author: Amit Kapila Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-08-05 05:07:16 +02:00
{
LockBuffer(buf, BUFFER_LOCK_UNLOCK);
visibilitymap_pin(rel, block, &vmbuffer);
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
}
Compare Xmin to previous Xmax when locking an update chain Not doing so causes us to traverse an update chain that has been broken by concurrent page pruning. All other code that traverses update chains uses this check as one of the cases in which to stop iterating, so replicate it here too. Failure to do so leads to erroneous CLOG, subtrans or multixact lookups. Per discussion following the bug report by J Smith in CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com as diagnosed by Andres Freund.
2013-11-27 21:49:12 +01:00
/*
* Check the tuple XMIN against prior XMAX, if any. If we reached the
* end of the chain, we're done, so return success.
*/
if (TransactionIdIsValid(priorXmax) &&
Revert bogus fixes of HOT-freezing bug It turns out we misdiagnosed what the real problem was. Revert the previous changes, because they may have worse consequences going forward. A better fix is forthcoming. The simplistic test case is kept, though disabled. Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de
2017-11-02 15:51:05 +01:00
!TransactionIdEquals(HeapTupleHeaderGetXmin(mytup.t_data),
priorXmax))
Compare Xmin to previous Xmax when locking an update chain Not doing so causes us to traverse an update chain that has been broken by concurrent page pruning. All other code that traverses update chains uses this check as one of the cases in which to stop iterating, so replicate it here too. Failure to do so leads to erroneous CLOG, subtrans or multixact lookups. Per discussion following the bug report by J Smith in CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com as diagnosed by Andres Freund.
2013-11-27 21:49:12 +01:00
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Ok;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
goto out_locked;
Compare Xmin to previous Xmax when locking an update chain Not doing so causes us to traverse an update chain that has been broken by concurrent page pruning. All other code that traverses update chains uses this check as one of the cases in which to stop iterating, so replicate it here too. Failure to do so leads to erroneous CLOG, subtrans or multixact lookups. Per discussion following the bug report by J Smith in CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com as diagnosed by Andres Freund.
2013-11-27 21:49:12 +01:00
}
Fix locking a tuple updated by an aborted (sub)transaction When heap_lock_tuple decides to follow the update chain, it tried to also lock any version of the tuple that was created by an update that was subsequently rolled back. This is pointless, since for all intents and purposes that tuple exists no more; and moreover it causes misbehavior, as reported independently by Marko Tiikkaja and Marti Raudsepp: some SELECT FOR UPDATE/SHARE queries may fail to return the tuples, and assertion-enabled builds crash. Fix by having heap_lock_updated_tuple test the xmin and return success immediately if the tuple was created by an aborted transaction. The condition where tuples become invisible occurs when an updated tuple chain is followed by heap_lock_updated_tuple, which reports the problem as HeapTupleSelfUpdated to its caller heap_lock_tuple, which in turn propagates that code outwards possibly leading the calling code (ExecLockRows) to believe that the tuple exists no longer. Backpatch to 9.3. Only on 9.5 and newer this leads to a visible failure, because of commit 27846f02c176; before that, heap_lock_tuple skips the whole dance when the tuple is already locked by the same transaction, because of the ancient HeapTupleSatisfiesUpdate behavior. Still, the buggy condition may also exist in more convoluted scenarios involving concurrent transactions, so it seems safer to fix the bug in the old branches too. Discussion: https://www.postgresql.org/message-id/CABRT9RC81YUf1=jsmWopcKJEro=VoeG2ou6sPwyOUTx_qteRsg@mail.gmail.com https://www.postgresql.org/message-id/48d3eade-98d3-8b9a-477e-1a8dc32a724d@joh.to
2016-09-09 20:54:29 +02:00
/*
* Also check Xmin: if this tuple was created by an aborted
* (sub)transaction, then we already locked the last live one in the
* chain, thus we're done, so return success.
*/
if (TransactionIdDidAbort(HeapTupleHeaderGetXmin(mytup.t_data)))
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Ok;
Fix VM buffer pin management in heap_lock_updated_tuple_rec(). Sloppy coding in this function could lead to leaking a VM buffer pin, or to attempting to free the same pin twice. Repair. While at it, reduce the code's tendency to free and reacquire the same page pin. Back-patch to 9.6; before that, this routine did not concern itself with VM pages. Amit Kapila and Tom Lane Discussion: https://postgr.es/m/CAA4eK1KJKwhc=isgTQHjM76CAdVswzNeAuZkh_cx-6QgGkSEgA@mail.gmail.com
2018-03-02 23:40:48 +01:00
goto out_locked;
Fix locking a tuple updated by an aborted (sub)transaction When heap_lock_tuple decides to follow the update chain, it tried to also lock any version of the tuple that was created by an update that was subsequently rolled back. This is pointless, since for all intents and purposes that tuple exists no more; and moreover it causes misbehavior, as reported independently by Marko Tiikkaja and Marti Raudsepp: some SELECT FOR UPDATE/SHARE queries may fail to return the tuples, and assertion-enabled builds crash. Fix by having heap_lock_updated_tuple test the xmin and return success immediately if the tuple was created by an aborted transaction. The condition where tuples become invisible occurs when an updated tuple chain is followed by heap_lock_updated_tuple, which reports the problem as HeapTupleSelfUpdated to its caller heap_lock_tuple, which in turn propagates that code outwards possibly leading the calling code (ExecLockRows) to believe that the tuple exists no longer. Backpatch to 9.3. Only on 9.5 and newer this leads to a visible failure, because of commit 27846f02c176; before that, heap_lock_tuple skips the whole dance when the tuple is already locked by the same transaction, because of the ancient HeapTupleSatisfiesUpdate behavior. Still, the buggy condition may also exist in more convoluted scenarios involving concurrent transactions, so it seems safer to fix the bug in the old branches too. Discussion: https://www.postgresql.org/message-id/CABRT9RC81YUf1=jsmWopcKJEro=VoeG2ou6sPwyOUTx_qteRsg@mail.gmail.com https://www.postgresql.org/message-id/48d3eade-98d3-8b9a-477e-1a8dc32a724d@joh.to
2016-09-09 20:54:29 +02:00
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
old_infomask = mytup.t_data->t_infomask;
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
old_infomask2 = mytup.t_data->t_infomask2;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
xmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
/*
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
* If this tuple version has been updated or locked by some concurrent
* transaction(s), what we do depends on whether our lock mode
* conflicts with what those other transactions hold, and also on the
* status of them.
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
*/
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
if (!(old_infomask & HEAP_XMAX_INVALID))
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
TransactionId rawxmax;
bool needwait;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
rawxmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
if (old_infomask & HEAP_XMAX_IS_MULTI)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
int nmembers;
int i;
MultiXactMember *members;
Fix handling of multixacts predating pg_upgrade After pg_upgrade, it is possible that some tuples' Xmax have multixacts corresponding to the old installation; such multixacts cannot have running members anymore. In many code sites we already know not to read them and clobber them silently, but at least when VACUUM tries to freeze a multixact or determine whether one needs freezing, there's an attempt to resolve it to its member transactions by calling GetMultiXactIdMembers, and if the multixact value is "in the future" with regards to the current valid multixact range, an error like this is raised: ERROR: MultiXactId 123 has not been created yet -- apparent wraparound and vacuuming fails. Per discussion with Andrew Gierth, it is completely bogus to try to resolve multixacts coming from before a pg_upgrade, regardless of where they stand with regards to the current valid multixact range. It's possible to get from under this problem by doing SELECT FOR UPDATE of the problem tuples, but if tables are large, this is slow and tedious, so a more thorough solution is desirable. To fix, we realize that multixacts in xmax created in 9.2 and previous have a specific bit pattern that is never used in 9.3 and later (we already knew this, per comments and infomask tests sprinkled in various places, but we weren't leveraging this knowledge appropriately). Whenever the infomask of the tuple matches that bit pattern, we just ignore the multixact completely as if Xmax wasn't set; or, in the case of tuple freezing, we act as if an unwanted value is set and clobber it without decoding. This guarantees that no errors will be raised, and that the values will be progressively removed until all tables are clean. Most callers of GetMultiXactIdMembers are patched to recognize directly that the value is a removable "empty" multixact and avoid calling GetMultiXactIdMembers altogether. To avoid changing the signature of GetMultiXactIdMembers() in back branches, we keep the "allow_old" boolean flag but rename it to "from_pgupgrade"; if the flag is true, we always return an empty set instead of looking up the multixact. (I suppose we could remove the argument in the master branch, but I chose not to do so in this commit). This was broken all along, but the error-facing message appeared first because of commit 8e9a16ab8f7f and was partially fixed in a25c2b7c4db3. This fix, backpatched all the way back to 9.3, goes approximately in the same direction as a25c2b7c4db3 but should cover all cases. Bug analysis by Andrew Gierth and Álvaro Herrera. A number of public reports match this bug: https://www.postgresql.org/message-id/20140330040029.GY4582@tamriel.snowman.net https://www.postgresql.org/message-id/538F3D70.6080902@publicrelay.com https://www.postgresql.org/message-id/556439CF.7070109@pscs.co.uk https://www.postgresql.org/message-id/SG2PR06MB0760098A111C88E31BD4D96FB3540@SG2PR06MB0760.apcprd06.prod.outlook.com https://www.postgresql.org/message-id/20160615203829.5798.4594@wrigleys.postgresql.org
2016-06-25 00:29:28 +02:00
/*
* We don't need a test for pg_upgrade'd tuples: this is only
* applied to tuples after the first in an update chain. Said
* first tuple in the chain may well be locked-in-9.2-and-
* pg_upgraded, but that one was already locked by our caller,
* not us; and any subsequent ones cannot be because our
* caller must necessarily have obtained a snapshot later than
* the pg_upgrade itself.
*/
Assert(!HEAP_LOCKED_UPGRADED(mytup.t_data->t_infomask));
Avoid uselessly looking up old LOCK_ONLY multixacts Commit 0ac5ad5134f2 removed an optimization in multixact.c that skipped fetching members of MultiXactId that were older than our OldestVisibleMXactId value. The reason this was removed is that it is possible for multixacts that contain updates to be older than that value. However, if the caller is certain that the multi does not contain an update (because the infomask bits say so), it can pass this info down to GetMultiXactIdMembers, enabling it to use the old optimization. Pointed out by Andres Freund in 20131121200517.GM7240@alap2.anarazel.de
2014-07-29 21:41:06 +02:00
nmembers = GetMultiXactIdMembers(rawxmax, &members, false,
HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
for (i = 0; i < nmembers; i++)
{
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
result = test_lockmode_for_conflict(members[i].status,
members[i].xid,
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
mode,
&mytup,
&needwait);
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
Fix concurrent locking of tuple update chain If several sessions are concurrently locking a tuple update chain with nonconflicting lock modes using an old snapshot, and they all succeed, it may happen that some of them fail because of restarting the loop (due to a concurrent Xmax change) and getting an error in the subsequent pass while trying to obtain a tuple lock that they already have in some tuple version. This can only happen with very high concurrency (where a row is being both updated and FK-checked by multiple transactions concurrently), but it's been observed in the field and can have unpleasant consequences such as an FK check failing to see a tuple that definitely exists: ERROR: insert or update on table "child_table" violates foreign key constraint "fk_constraint_name" DETAIL: Key (keyid)=(123456) is not present in table "parent_table". (where the key is observably present in the table). Discussion: https://postgr.es/m/20170714210011.r25mrff4nxjhmf3g@alvherre.pgsql
2017-07-26 23:24:16 +02:00
/*
* If the tuple was already locked by ourselves in a
* previous iteration of this (say heap_lock_tuple was
* forced to restart the locking loop because of a change
* in xmax), then we hold the lock already on this tuple
* version and we don't need to do anything; and this is
* not an error condition either. We just need to skip
* this tuple and continue locking the next version in the
* update chain.
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (result == TM_SelfModified)
Fix concurrent locking of tuple update chain If several sessions are concurrently locking a tuple update chain with nonconflicting lock modes using an old snapshot, and they all succeed, it may happen that some of them fail because of restarting the loop (due to a concurrent Xmax change) and getting an error in the subsequent pass while trying to obtain a tuple lock that they already have in some tuple version. This can only happen with very high concurrency (where a row is being both updated and FK-checked by multiple transactions concurrently), but it's been observed in the field and can have unpleasant consequences such as an FK check failing to see a tuple that definitely exists: ERROR: insert or update on table "child_table" violates foreign key constraint "fk_constraint_name" DETAIL: Key (keyid)=(123456) is not present in table "parent_table". (where the key is observably present in the table). Discussion: https://postgr.es/m/20170714210011.r25mrff4nxjhmf3g@alvherre.pgsql
2017-07-26 23:24:16 +02:00
{
pfree(members);
goto next;
}
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
if (needwait)
{
LockBuffer(buf, BUFFER_LOCK_UNLOCK);
Setup error context callback for transaction lock waits With this in place, a session blocking behind another one because of tuple locks will get a context line mentioning the relation name, tuple TID, and operation being done on tuple. For example: LOG: process 11367 still waiting for ShareLock on transaction 717 after 1000.108 ms DETAIL: Process holding the lock: 11366. Wait queue: 11367. CONTEXT: while updating tuple (0,2) in relation "foo" STATEMENT: UPDATE foo SET value = 3; Most usefully, the new line is displayed by log entries due to log_lock_waits, although of course it will be printed by any other log message as well. Author: Christian Kruse, some tweaks by Álvaro Herrera Reviewed-by: Amit Kapila, Andres Freund, Tom Lane, Robert Haas
2014-03-19 19:10:36 +01:00
XactLockTableWait(members[i].xid, rel,
Fix reference-after-free when waiting for another xact due to constraint. If an insertion or update had to wait for another transaction to finish, because there was another insertion with conflicting key in progress, we would pass a just-free'd item pointer to XactLockTableWait(). All calls to XactLockTableWait() and MultiXactIdWait() had similar issues. Some passed a pointer to a buffer in the buffer cache, after already releasing the lock. The call in EvalPlanQualFetch had already released the pin too. All but the call in execUtils.c would merely lead to reporting a bogus ctid, however (or an assertion failure, if enabled). All the callers that passed HeapTuple->t_data->t_ctid were slightly bogus anyway: if the tuple was updated (again) in the same transaction, its ctid field would point to the next tuple in the chain, not the tuple itself. Backpatch to 9.4, where the 'ctid' argument to XactLockTableWait was added (in commit f88d4cfc)
2015-02-04 15:00:34 +01:00
&mytup.t_self,
Setup error context callback for transaction lock waits With this in place, a session blocking behind another one because of tuple locks will get a context line mentioning the relation name, tuple TID, and operation being done on tuple. For example: LOG: process 11367 still waiting for ShareLock on transaction 717 after 1000.108 ms DETAIL: Process holding the lock: 11366. Wait queue: 11367. CONTEXT: while updating tuple (0,2) in relation "foo" STATEMENT: UPDATE foo SET value = 3; Most usefully, the new line is displayed by log entries due to log_lock_waits, although of course it will be printed by any other log message as well. Author: Christian Kruse, some tweaks by Álvaro Herrera Reviewed-by: Amit Kapila, Andres Freund, Tom Lane, Robert Haas
2014-03-19 19:10:36 +01:00
XLTW_LockUpdated);
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
pfree(members);
goto l4;
}
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (result != TM_Ok)
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
{
pfree(members);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
goto out_locked;
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
}
}
if (members)
pfree(members);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
else
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
MultiXactStatus status;
/*
* For a non-multi Xmax, we first need to compute the
* corresponding MultiXactStatus by using the infomask bits.
*/
if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
{
if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
status = MultiXactStatusForKeyShare;
else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
status = MultiXactStatusForShare;
else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
{
if (old_infomask2 & HEAP_KEYS_UPDATED)
status = MultiXactStatusForUpdate;
else
status = MultiXactStatusForNoKeyUpdate;
}
else
{
/*
* LOCK_ONLY present alone (a pg_upgraded tuple marked
* as share-locked in the old cluster) shouldn't be
* seen in the middle of an update chain.
*/
elog(ERROR, "invalid lock status in tuple");
}
}
else
{
/* it's an update, but which kind? */
if (old_infomask2 & HEAP_KEYS_UPDATED)
status = MultiXactStatusUpdate;
else
status = MultiXactStatusNoKeyUpdate;
}
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
result = test_lockmode_for_conflict(status, rawxmax, mode,
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
&mytup, &needwait);
Fix concurrent locking of tuple update chain If several sessions are concurrently locking a tuple update chain with nonconflicting lock modes using an old snapshot, and they all succeed, it may happen that some of them fail because of restarting the loop (due to a concurrent Xmax change) and getting an error in the subsequent pass while trying to obtain a tuple lock that they already have in some tuple version. This can only happen with very high concurrency (where a row is being both updated and FK-checked by multiple transactions concurrently), but it's been observed in the field and can have unpleasant consequences such as an FK check failing to see a tuple that definitely exists: ERROR: insert or update on table "child_table" violates foreign key constraint "fk_constraint_name" DETAIL: Key (keyid)=(123456) is not present in table "parent_table". (where the key is observably present in the table). Discussion: https://postgr.es/m/20170714210011.r25mrff4nxjhmf3g@alvherre.pgsql
2017-07-26 23:24:16 +02:00
/*
* If the tuple was already locked by ourselves in a previous
* iteration of this (say heap_lock_tuple was forced to
* restart the locking loop because of a change in xmax), then
* we hold the lock already on this tuple version and we don't
* need to do anything; and this is not an error condition
* either. We just need to skip this tuple and continue
* locking the next version in the update chain.
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (result == TM_SelfModified)
Fix concurrent locking of tuple update chain If several sessions are concurrently locking a tuple update chain with nonconflicting lock modes using an old snapshot, and they all succeed, it may happen that some of them fail because of restarting the loop (due to a concurrent Xmax change) and getting an error in the subsequent pass while trying to obtain a tuple lock that they already have in some tuple version. This can only happen with very high concurrency (where a row is being both updated and FK-checked by multiple transactions concurrently), but it's been observed in the field and can have unpleasant consequences such as an FK check failing to see a tuple that definitely exists: ERROR: insert or update on table "child_table" violates foreign key constraint "fk_constraint_name" DETAIL: Key (keyid)=(123456) is not present in table "parent_table". (where the key is observably present in the table). Discussion: https://postgr.es/m/20170714210011.r25mrff4nxjhmf3g@alvherre.pgsql
2017-07-26 23:24:16 +02:00
goto next;
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
if (needwait)
{
LockBuffer(buf, BUFFER_LOCK_UNLOCK);
Fix reference-after-free when waiting for another xact due to constraint. If an insertion or update had to wait for another transaction to finish, because there was another insertion with conflicting key in progress, we would pass a just-free'd item pointer to XactLockTableWait(). All calls to XactLockTableWait() and MultiXactIdWait() had similar issues. Some passed a pointer to a buffer in the buffer cache, after already releasing the lock. The call in EvalPlanQualFetch had already released the pin too. All but the call in execUtils.c would merely lead to reporting a bogus ctid, however (or an assertion failure, if enabled). All the callers that passed HeapTuple->t_data->t_ctid were slightly bogus anyway: if the tuple was updated (again) in the same transaction, its ctid field would point to the next tuple in the chain, not the tuple itself. Backpatch to 9.4, where the 'ctid' argument to XactLockTableWait was added (in commit f88d4cfc)
2015-02-04 15:00:34 +01:00
XactLockTableWait(rawxmax, rel, &mytup.t_self,
Setup error context callback for transaction lock waits With this in place, a session blocking behind another one because of tuple locks will get a context line mentioning the relation name, tuple TID, and operation being done on tuple. For example: LOG: process 11367 still waiting for ShareLock on transaction 717 after 1000.108 ms DETAIL: Process holding the lock: 11366. Wait queue: 11367. CONTEXT: while updating tuple (0,2) in relation "foo" STATEMENT: UPDATE foo SET value = 3; Most usefully, the new line is displayed by log entries due to log_lock_waits, although of course it will be printed by any other log message as well. Author: Christian Kruse, some tweaks by Álvaro Herrera Reviewed-by: Amit Kapila, Andres Freund, Tom Lane, Robert Haas
2014-03-19 19:10:36 +01:00
XLTW_LockUpdated);
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
goto l4;
}
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
if (result != TM_Ok)
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
{
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
goto out_locked;
Use a more granular approach to follow update chains Instead of simply checking the KEYS_UPDATED bit, we need to check whether each lock held on the future version of the tuple conflicts with the lock we're trying to acquire. Per bug report #8434 by Tomonari Katsumata
2013-11-27 21:50:33 +01:00
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
}
/* compute the new Xmax and infomask values for the tuple ... */
compute_new_xmax_infomask(xmax, old_infomask, mytup.t_data->t_infomask2,
xid, mode, false,
&new_xmax, &new_infomask, &new_infomask2);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
if (PageIsAllVisible(BufferGetPage(buf)) &&
visibilitymap_clear(rel, block, vmbuffer,
VISIBILITYMAP_ALL_FROZEN))
cleared_all_frozen = true;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
START_CRIT_SECTION();
/* ... and set them */
HeapTupleHeaderSetXmax(mytup.t_data, new_xmax);
mytup.t_data->t_infomask &= ~HEAP_XMAX_BITS;
mytup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
mytup.t_data->t_infomask |= new_infomask;
mytup.t_data->t_infomask2 |= new_infomask2;
MarkBufferDirty(buf);
/* XLOG stuff */
if (RelationNeedsWAL(rel))
{
xl_heap_lock_updated xlrec;
XLogRecPtr recptr;
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
Page page = BufferGetPage(buf);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogBeginInsert();
XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
xlrec.offnum = ItemPointerGetOffsetNumber(&mytup.t_self);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
xlrec.xmax = new_xmax;
xlrec.infobits_set = compute_infobits(new_infomask, new_infomask2);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
xlrec.flags =
cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterData((char *) &xlrec, SizeOfHeapLockUpdated);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_LOCK_UPDATED);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
PageSetLSN(page, recptr);
}
END_CRIT_SECTION();
Fix concurrent locking of tuple update chain If several sessions are concurrently locking a tuple update chain with nonconflicting lock modes using an old snapshot, and they all succeed, it may happen that some of them fail because of restarting the loop (due to a concurrent Xmax change) and getting an error in the subsequent pass while trying to obtain a tuple lock that they already have in some tuple version. This can only happen with very high concurrency (where a row is being both updated and FK-checked by multiple transactions concurrently), but it's been observed in the field and can have unpleasant consequences such as an FK check failing to see a tuple that definitely exists: ERROR: insert or update on table "child_table" violates foreign key constraint "fk_constraint_name" DETAIL: Key (keyid)=(123456) is not present in table "parent_table". (where the key is observably present in the table). Discussion: https://postgr.es/m/20170714210011.r25mrff4nxjhmf3g@alvherre.pgsql
2017-07-26 23:24:16 +02:00
next:
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/* if we find the end of update chain, we're done. */
if (mytup.t_data->t_infomask & HEAP_XMAX_INVALID ||
Raise error when affecting tuple moved into different partition. When an update moves a row between partitions (supported since 2f178441044b), our normal logic for following update chains in READ COMMITTED mode doesn't work anymore. Cross partition updates are modeled as an delete from the old and insert into the new partition. No ctid chain exists across partitions, and there's no convenient space to introduce that link. Not throwing an error in a partitioned context when one would have been thrown without partitioning is obviously problematic. This commit introduces infrastructure to detect when a tuple has been moved, not just plainly deleted. That allows to throw an error when encountering a deletion that's actually a move, while attempting to following a ctid chain. The row deleted as part of a cross partition update is marked by pointing it's t_ctid to an invalid block, instead of self as a normal update would. That was deemed to be the least invasive and most future proof way to represent the knowledge, given how few infomask bits are there to be recycled (there's also some locking issues with using infomask bits). External code following ctid chains should be updated to check for moved tuples. The most likely consequence of not doing so is a missed error. Author: Amul Sul, editorialized by me Reviewed-By: Amit Kapila, Pavan Deolasee, Andres Freund, Robert Haas Discussion: http://postgr.es/m/CAAJ_b95PkwojoYfz0bzXU8OokcTVGzN6vYGCNVUukeUDrnF3dw@mail.gmail.com
2018-04-07 22:24:10 +02:00
HeapTupleHeaderIndicatesMovedPartitions(mytup.t_data) ||
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
ItemPointerEquals(&mytup.t_self, &mytup.t_data->t_ctid) ||
HeapTupleHeaderIsOnlyLocked(mytup.t_data))
{
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Ok;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
goto out_locked;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
Use the standard lock manager to establish priority order when there is contention for a tuple-level lock. This solves the problem of a would-be exclusive locker being starved out by an indefinite succession of share-lockers. Per recent discussion with Alvaro.
2005-04-30 21:03:33 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/* tail recursion */
Compare Xmin to previous Xmax when locking an update chain Not doing so causes us to traverse an update chain that has been broken by concurrent page pruning. All other code that traverses update chains uses this check as one of the cases in which to stop iterating, so replicate it here too. Failure to do so leads to erroneous CLOG, subtrans or multixact lookups. Per discussion following the bug report by J Smith in CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com as diagnosed by Andres Freund.
2013-11-27 21:49:12 +01:00
priorXmax = HeapTupleHeaderGetUpdateXid(mytup.t_data);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
ItemPointerCopy(&(mytup.t_data->t_ctid), &tupid);
UnlockReleaseBuffer(buf);
}
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
result = TM_Ok;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
out_locked:
UnlockReleaseBuffer(buf);
Fix VM buffer pin management in heap_lock_updated_tuple_rec(). Sloppy coding in this function could lead to leaking a VM buffer pin, or to attempting to free the same pin twice. Repair. While at it, reduce the code's tendency to free and reacquire the same page pin. Back-patch to 9.6; before that, this routine did not concern itself with VM pages. Amit Kapila and Tom Lane Discussion: https://postgr.es/m/CAA4eK1KJKwhc=isgTQHjM76CAdVswzNeAuZkh_cx-6QgGkSEgA@mail.gmail.com
2018-03-02 23:40:48 +01:00
out_unlocked:
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
if (vmbuffer != InvalidBuffer)
ReleaseBuffer(vmbuffer);
return result;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
/*
* heap_lock_updated_tuple
* Follow update chain when locking an updated tuple, acquiring locks (row
* marks) on the updated versions.
*
* The initial tuple is assumed to be already locked.
*
Fix misc typos. Oskari Saarenmaa. Backpatch to stable branches where applicable.
2015-09-05 10:35:49 +02:00
* This function doesn't check visibility, it just unconditionally marks the
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* tuple(s) as locked. If any tuple in the updated chain is being deleted
* concurrently (or updated with the key being modified), sleep until the
* transaction doing it is finished.
*
* Note that we don't acquire heavyweight tuple locks on the tuples we walk
* when we have to wait for other transactions to release them, as opposed to
* what heap_lock_tuple does. The reason is that having more than one
* transaction walking the chain is probably uncommon enough that risk of
* starvation is not likely: one of the preconditions for being here is that
* the snapshot in use predates the update that created this tuple (because we
* started at an earlier version of the tuple), but at the same time such a
* transaction cannot be using repeatable read or serializable isolation
* levels, because that would lead to a serializability failure.
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
static TM_Result
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
heap_lock_updated_tuple(Relation rel, HeapTuple tuple, ItemPointer ctid,
TransactionId xid, LockTupleMode mode)
{
Raise error when affecting tuple moved into different partition. When an update moves a row between partitions (supported since 2f178441044b), our normal logic for following update chains in READ COMMITTED mode doesn't work anymore. Cross partition updates are modeled as an delete from the old and insert into the new partition. No ctid chain exists across partitions, and there's no convenient space to introduce that link. Not throwing an error in a partitioned context when one would have been thrown without partitioning is obviously problematic. This commit introduces infrastructure to detect when a tuple has been moved, not just plainly deleted. That allows to throw an error when encountering a deletion that's actually a move, while attempting to following a ctid chain. The row deleted as part of a cross partition update is marked by pointing it's t_ctid to an invalid block, instead of self as a normal update would. That was deemed to be the least invasive and most future proof way to represent the knowledge, given how few infomask bits are there to be recycled (there's also some locking issues with using infomask bits). External code following ctid chains should be updated to check for moved tuples. The most likely consequence of not doing so is a missed error. Author: Amul Sul, editorialized by me Reviewed-By: Amit Kapila, Pavan Deolasee, Andres Freund, Robert Haas Discussion: http://postgr.es/m/CAAJ_b95PkwojoYfz0bzXU8OokcTVGzN6vYGCNVUukeUDrnF3dw@mail.gmail.com
2018-04-07 22:24:10 +02:00
/*
* If the tuple has not been updated, or has moved into another partition
* (effectively a delete) stop here.
*/
if (!HeapTupleHeaderIndicatesMovedPartitions(tuple->t_data) &&
!ItemPointerEquals(&tuple->t_self, ctid))
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
/*
* If this is the first possibly-multixact-able operation in the
* current transaction, set my per-backend OldestMemberMXactId
* setting. We can be certain that the transaction will never become a
* member of any older MultiXactIds than that. (We have to do this
* even if we end up just using our own TransactionId below, since
* some other backend could incorporate our XID into a MultiXact
* immediately afterwards.)
*/
MultiXactIdSetOldestMember();
return heap_lock_updated_tuple_rec(rel, ctid, xid, mode);
}
/* nothing to lock */
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
return TM_Ok;
Postgres95 1.01 Distribution - Virgin Sources
1996-07-09 08:22:35 +02:00
}
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
/*
* heap_finish_speculative - mark speculative insertion as successful
*
* To successfully finish a speculative insertion we have to clear speculative
* token from tuple. To do so the t_ctid field, which will contain a
* speculative token value, is modified in place to point to the tuple itself,
* which is characteristic of a newly inserted ordinary tuple.
*
* NB: It is not ok to commit without either finishing or aborting a
* speculative insertion. We could treat speculative tuples of committed
* transactions implicitly as completed, but then we would have to be prepared
* to deal with speculative tokens on committed tuples. That wouldn't be
* difficult - no-one looks at the ctid field of a tuple with invalid xmax -
* but clearing the token at completion isn't very expensive either.
* An explicit confirmation WAL record also makes logical decoding simpler.
*/
void
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
heap_finish_speculative(Relation relation, ItemPointer tid)
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
{
Buffer buffer;
Page page;
OffsetNumber offnum;
ItemId lp = NULL;
HeapTupleHeader htup;
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = (Page) BufferGetPage(buffer);
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
offnum = ItemPointerGetOffsetNumber(tid);
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
if (PageGetMaxOffsetNumber(page) >= offnum)
lp = PageGetItemId(page, offnum);
if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(ERROR, "invalid lp");
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
htup = (HeapTupleHeader) PageGetItem(page, lp);
/* NO EREPORT(ERROR) from here till changes are logged */
START_CRIT_SECTION();
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
Assert(HeapTupleHeaderIsSpeculative(htup));
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
MarkBufferDirty(buffer);
/*
* Replace the speculative insertion token with a real t_ctid, pointing to
* itself like it does on regular tuples.
*/
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
htup->t_ctid = *tid;
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
/* XLOG stuff */
if (RelationNeedsWAL(relation))
{
xl_heap_confirm xlrec;
XLogRecPtr recptr;
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
xlrec.offnum = ItemPointerGetOffsetNumber(tid);
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
XLogBeginInsert();
/* We want the same filtering on this as on a plain insert */
Skip checkpoints, archiving on idle systems. Some background activity (like checkpoints, archive timeout, standby snapshots) is not supposed to happen on an idle system. Unfortunately so far it was not easy to determine when a system is idle, which defeated some of the attempts to avoid redundant activity on an idle system. To make that easier, allow to make individual WAL insertions as not being "important". By checking whether any important activity happened since the last time an activity was performed, it now is easy to check whether some action needs to be repeated. Use the new facility for checkpoints, archive timeout and standby snapshots. The lack of a facility causes some issues in older releases, but in my opinion the consequences (superflous checkpoints / archived segments) aren't grave enough to warrant backpatching. Author: Michael Paquier, editorialized by Andres Freund Reviewed-By: Andres Freund, David Steele, Amit Kapila, Kyotaro HORIGUCHI Bug: #13685 Discussion: https://www.postgresql.org/message-id/20151016203031.3019.72930@wrigleys.postgresql.org https://www.postgresql.org/message-id/CAB7nPqQcPqxEM3S735Bd2RzApNqSNJVietAC=6kfkYv_45dKwA@mail.gmail.com Backpatch: -
2016-12-22 20:31:50 +01:00
XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
XLogRegisterData((char *) &xlrec, SizeOfHeapConfirm);
XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_CONFIRM);
PageSetLSN(page, recptr);
}
END_CRIT_SECTION();
UnlockReleaseBuffer(buffer);
}
/*
* heap_abort_speculative - kill a speculatively inserted tuple
*
* Marks a tuple that was speculatively inserted in the same command as dead,
* by setting its xmin as invalid. That makes it immediately appear as dead
* to all transactions, including our own. In particular, it makes
* HeapTupleSatisfiesDirty() regard the tuple as dead, so that another backend
* inserting a duplicate key value won't unnecessarily wait for our whole
* transaction to finish (it'll just wait for our speculative insertion to
* finish).
*
* Killing the tuple prevents "unprincipled deadlocks", which are deadlocks
* that arise due to a mutual dependency that is not user visible. By
* definition, unprincipled deadlocks cannot be prevented by the user
* reordering lock acquisition in client code, because the implementation level
* lock acquisitions are not under the user's direct control. If speculative
* inserters did not take this precaution, then under high concurrency they
* could deadlock with each other, which would not be acceptable.
*
* This is somewhat redundant with heap_delete, but we prefer to have a
Fix deletion of speculatively inserted TOAST on conflict INSERT .. ON CONFLICT runs a pre-check of the possible conflicting constraints before performing the actual speculative insertion. In case the inserted tuple included TOASTed columns the ON CONFLICT condition would be handled correctly in case the conflict was caught by the pre-check, but if two transactions entered the speculative insertion phase at the same time, one would have to re-try, and the code for aborting a speculative insertion did not handle deleting the speculatively inserted TOAST datums correctly. TOAST deletion would fail with "ERROR: attempted to delete invisible tuple" as we attempted to remove the TOAST tuples using simple_heap_delete which reasoned that the given tuples should not be visible to the command that wrote them. This commit updates the heap_abort_speculative() function which aborts the conflicting tuple to use itself, via toast_delete, for deleting associated TOAST datums. Like before, the inserted toast rows are not marked as being speculative. This commit also adds a isolationtester spec test, exercising the relevant code path. Unfortunately 9.5 cannot handle two waiting sessions, and thus cannot execute this test. Reported-By: Viren Negi, Oskari Saarenmaa Author: Oskari Saarenmaa, edited a bit by me Bug: #14150 Discussion: <20160519123338.12513.20271@wrigleys.postgresql.org> Backpatch: 9.5, where ON CONFLICT was introduced
2016-08-18 02:03:36 +02:00
* dedicated routine with stripped down requirements. Note that this is also
* used to delete the TOAST tuples created during speculative insertion.
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
*
* This routine does not affect logical decoding as it only looks at
* confirmation records.
*/
void
tableam: Add tuple_{insert, delete, update, lock} and use. This adds new, required, table AM callbacks for insert/delete/update and lock_tuple. To be able to reasonably use those, the EvalPlanQual mechanism had to be adapted, moving more logic into the AM. Previously both delete/update/lock call-sites and the EPQ mechanism had to have awareness of the specific tuple format to be able to fetch the latest version of a tuple. Obviously that needs to be abstracted away. To do so, move the logic that find the latest row version into the AM. lock_tuple has a new flag argument, TUPLE_LOCK_FLAG_FIND_LAST_VERSION, that forces it to lock the last version, rather than the current one. It'd have been possible to do so via a separate callback as well, but finding the last version usually also necessitates locking the newest version, making it sensible to combine the two. This replaces the previous use of EvalPlanQualFetch(). Additionally HeapTupleUpdated, which previously signaled either a concurrent update or delete, is now split into two, to avoid callers needing AM specific knowledge to differentiate. The move of finding the latest row version into tuple_lock means that encountering a row concurrently moved into another partition will now raise an error about "tuple to be locked" rather than "tuple to be updated/deleted" - which is accurate, as that always happens when locking rows. While possible slightly less helpful for users, it seems like an acceptable trade-off. As part of this commit HTSU_Result has been renamed to TM_Result, and its members been expanded to differentiated between updating and deleting. HeapUpdateFailureData has been renamed to TM_FailureData. The interface to speculative insertion is changed so nodeModifyTable.c does not have to set the speculative token itself anymore. Instead there's a version of tuple_insert, tuple_insert_speculative, that performs the speculative insertion (without requiring a flag to signal that fact), and the speculative insertion is either made permanent with table_complete_speculative(succeeded = true) or aborted with succeeded = false). Note that multi_insert is not yet routed through tableam, nor is COPY. Changing multi_insert requires changes to copy.c that are large enough to better be done separately. Similarly, although simpler, CREATE TABLE AS and CREATE MATERIALIZED VIEW are also only going to be adjusted in a later commit. Author: Andres Freund and Haribabu Kommi Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de https://postgr.es/m/20190313003903.nwvrxi7rw3ywhdel@alap3.anarazel.de https://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
2019-03-24 03:55:57 +01:00
heap_abort_speculative(Relation relation, ItemPointer tid)
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
{
TransactionId xid = GetCurrentTransactionId();
ItemId lp;
HeapTupleData tp;
Page page;
BlockNumber block;
Buffer buffer;
Use TransactionXmin instead of RecentGlobalXmin in heap_abort_speculative(). There's a very low risk that RecentGlobalXmin could be far enough in the past to be older than relfrozenxid, or even wrapped around. Luckily the consequences of that having happened wouldn't be too bad - the page wouldn't be pruned for a while. Avoid that risk by using TransactionXmin instead. As that's announced via MyPgXact->xmin, it is protected against wrapping around (see code comments for details around relfrozenxid). Author: Andres Freund Discussion: https://postgr.es/m/20200328213023.s4eyijhdosuc4vcj@alap3.anarazel.de Backpatch: 9.5-
2020-04-06 02:47:30 +02:00
TransactionId prune_xid;
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
Assert(ItemPointerIsValid(tid));
block = ItemPointerGetBlockNumber(tid);
buffer = ReadBuffer(relation, block);
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(buffer);
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
/*
* Page can't be all visible, we just inserted into it, and are still
* running.
*/
Assert(!PageIsAllVisible(page));
lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
Assert(ItemIdIsNormal(lp));
tp.t_tableOid = RelationGetRelid(relation);
tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
tp.t_len = ItemIdGetLength(lp);
tp.t_self = *tid;
/*
* Sanity check that the tuple really is a speculatively inserted tuple,
* inserted by us.
*/
if (tp.t_data->t_choice.t_heap.t_xmin != xid)
elog(ERROR, "attempted to kill a tuple inserted by another transaction");
Fix deletion of speculatively inserted TOAST on conflict INSERT .. ON CONFLICT runs a pre-check of the possible conflicting constraints before performing the actual speculative insertion. In case the inserted tuple included TOASTed columns the ON CONFLICT condition would be handled correctly in case the conflict was caught by the pre-check, but if two transactions entered the speculative insertion phase at the same time, one would have to re-try, and the code for aborting a speculative insertion did not handle deleting the speculatively inserted TOAST datums correctly. TOAST deletion would fail with "ERROR: attempted to delete invisible tuple" as we attempted to remove the TOAST tuples using simple_heap_delete which reasoned that the given tuples should not be visible to the command that wrote them. This commit updates the heap_abort_speculative() function which aborts the conflicting tuple to use itself, via toast_delete, for deleting associated TOAST datums. Like before, the inserted toast rows are not marked as being speculative. This commit also adds a isolationtester spec test, exercising the relevant code path. Unfortunately 9.5 cannot handle two waiting sessions, and thus cannot execute this test. Reported-By: Viren Negi, Oskari Saarenmaa Author: Oskari Saarenmaa, edited a bit by me Bug: #14150 Discussion: <20160519123338.12513.20271@wrigleys.postgresql.org> Backpatch: 9.5, where ON CONFLICT was introduced
2016-08-18 02:03:36 +02:00
if (!(IsToastRelation(relation) || HeapTupleHeaderIsSpeculative(tp.t_data)))
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
elog(ERROR, "attempted to kill a non-speculative tuple");
Assert(!HeapTupleHeaderIsHeapOnly(tp.t_data));
/*
* No need to check for serializable conflicts here. There is never a
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
* need for a combo CID, either. No need to extract replica identity, or
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
* do anything special with infomask bits.
*/
START_CRIT_SECTION();
/*
Use TransactionXmin instead of RecentGlobalXmin in heap_abort_speculative(). There's a very low risk that RecentGlobalXmin could be far enough in the past to be older than relfrozenxid, or even wrapped around. Luckily the consequences of that having happened wouldn't be too bad - the page wouldn't be pruned for a while. Avoid that risk by using TransactionXmin instead. As that's announced via MyPgXact->xmin, it is protected against wrapping around (see code comments for details around relfrozenxid). Author: Andres Freund Discussion: https://postgr.es/m/20200328213023.s4eyijhdosuc4vcj@alap3.anarazel.de Backpatch: 9.5-
2020-04-06 02:47:30 +02:00
* The tuple will become DEAD immediately. Flag that this page is a
* candidate for pruning by setting xmin to TransactionXmin. While not
* immediately prunable, it is the oldest xid we can cheaply determine
* that's safe against wraparound / being older than the table's
* relfrozenxid. To defend against the unlikely case of a new relation
* having a newer relfrozenxid than our TransactionXmin, use relfrozenxid
* if so (vacuum can't subsequently move relfrozenxid to beyond
* TransactionXmin, so there's no race here).
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
*/
Use TransactionXmin instead of RecentGlobalXmin in heap_abort_speculative(). There's a very low risk that RecentGlobalXmin could be far enough in the past to be older than relfrozenxid, or even wrapped around. Luckily the consequences of that having happened wouldn't be too bad - the page wouldn't be pruned for a while. Avoid that risk by using TransactionXmin instead. As that's announced via MyPgXact->xmin, it is protected against wrapping around (see code comments for details around relfrozenxid). Author: Andres Freund Discussion: https://postgr.es/m/20200328213023.s4eyijhdosuc4vcj@alap3.anarazel.de Backpatch: 9.5-
2020-04-06 02:47:30 +02:00
Assert(TransactionIdIsValid(TransactionXmin));
if (TransactionIdPrecedes(TransactionXmin, relation->rd_rel->relfrozenxid))
prune_xid = relation->rd_rel->relfrozenxid;
else
prune_xid = TransactionXmin;
PageSetPrunable(page, prune_xid);
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
/* store transaction information of xact deleting the tuple */
tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
/*
* Set the tuple header xmin to InvalidTransactionId. This makes the
* tuple immediately invisible everyone. (In particular, to any
* transactions waiting on the speculative token, woken up later.)
*/
HeapTupleHeaderSetXmin(tp.t_data, InvalidTransactionId);
/* Clear the speculative insertion token too */
tp.t_data->t_ctid = tp.t_self;
MarkBufferDirty(buffer);
/*
* XLOG stuff
*
* The WAL records generated here match heap_delete(). The same recovery
* routines are used.
*/
if (RelationNeedsWAL(relation))
{
xl_heap_delete xlrec;
XLogRecPtr recptr;
xlrec.flags = XLH_DELETE_IS_SUPER;
xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
tp.t_data->t_infomask2);
xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
xlrec.xmax = xid;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHeapDelete);
XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
/* No replica identity & replication origin logged */
recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
PageSetLSN(page, recptr);
}
END_CRIT_SECTION();
LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
if (HeapTupleHasExternal(&tp))
Fix deletion of speculatively inserted TOAST on conflict INSERT .. ON CONFLICT runs a pre-check of the possible conflicting constraints before performing the actual speculative insertion. In case the inserted tuple included TOASTed columns the ON CONFLICT condition would be handled correctly in case the conflict was caught by the pre-check, but if two transactions entered the speculative insertion phase at the same time, one would have to re-try, and the code for aborting a speculative insertion did not handle deleting the speculatively inserted TOAST datums correctly. TOAST deletion would fail with "ERROR: attempted to delete invisible tuple" as we attempted to remove the TOAST tuples using simple_heap_delete which reasoned that the given tuples should not be visible to the command that wrote them. This commit updates the heap_abort_speculative() function which aborts the conflicting tuple to use itself, via toast_delete, for deleting associated TOAST datums. Like before, the inserted toast rows are not marked as being speculative. This commit also adds a isolationtester spec test, exercising the relevant code path. Unfortunately 9.5 cannot handle two waiting sessions, and thus cannot execute this test. Reported-By: Viren Negi, Oskari Saarenmaa Author: Oskari Saarenmaa, edited a bit by me Bug: #14150 Discussion: <20160519123338.12513.20271@wrigleys.postgresql.org> Backpatch: 9.5, where ON CONFLICT was introduced
2016-08-18 02:03:36 +02:00
{
Assert(!IsToastRelation(relation));
Rename some toasting functions based on whether they are heap-specific. The old names for the attribute-detoasting functions names included the word "heap," which seems outdated now that the heap is only one of potentially many table access methods. On the other hand, toast_insert_or_update and toast_delete are heap-specific, so rename them by adding "heap_" as a prefix. Not all of the work of making the TOAST system fully accessible to AMs other than the heap is done yet, but there seems to be little harm in getting this renaming out of the way now. Commit 8b94dab06617ef80a0901ab103ebd8754427ef5a already divided up the functions among various files partially according to whether it was intended that they should be heap-specific or AM-agnostic, so this is just clarifying the division contemplated by that commit. Patch by me, reviewed and tested by Prabhat Sabu, Thomas Munro, Andres Freund, and Álvaro Herrera. Discussion: http://postgr.es/m/CA+TgmoZv-=2iWM4jcw5ZhJeL18HF96+W1yJeYrnGMYdkFFnEpQ@mail.gmail.com
2019-10-04 20:24:46 +02:00
heap_toast_delete(relation, &tp, true);
Fix deletion of speculatively inserted TOAST on conflict INSERT .. ON CONFLICT runs a pre-check of the possible conflicting constraints before performing the actual speculative insertion. In case the inserted tuple included TOASTed columns the ON CONFLICT condition would be handled correctly in case the conflict was caught by the pre-check, but if two transactions entered the speculative insertion phase at the same time, one would have to re-try, and the code for aborting a speculative insertion did not handle deleting the speculatively inserted TOAST datums correctly. TOAST deletion would fail with "ERROR: attempted to delete invisible tuple" as we attempted to remove the TOAST tuples using simple_heap_delete which reasoned that the given tuples should not be visible to the command that wrote them. This commit updates the heap_abort_speculative() function which aborts the conflicting tuple to use itself, via toast_delete, for deleting associated TOAST datums. Like before, the inserted toast rows are not marked as being speculative. This commit also adds a isolationtester spec test, exercising the relevant code path. Unfortunately 9.5 cannot handle two waiting sessions, and thus cannot execute this test. Reported-By: Viren Negi, Oskari Saarenmaa Author: Oskari Saarenmaa, edited a bit by me Bug: #14150 Discussion: <20160519123338.12513.20271@wrigleys.postgresql.org> Backpatch: 9.5, where ON CONFLICT was introduced
2016-08-18 02:03:36 +02:00
}
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
/*
* Never need to mark tuple for invalidation, since catalogs don't support
* speculative insertion
*/
/* Now we can release the buffer */
ReleaseBuffer(buffer);
/* count deletion, as we counted the insertion too */
pgstat_count_heap_delete(relation);
}
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
/*
* heap_inplace_update - update a tuple "in place" (ie, overwrite it)
*
* Overwriting violates both MVCC and transactional safety, so the uses
* of this function in Postgres are extremely limited. Nonetheless we
* find some places to use it.
*
* The tuple cannot change size, and therefore it's reasonable to assume
* that its null bitmap (if any) doesn't change either. So we just
* overwrite the data portion of the tuple without touching the null
* bitmap or any of the header fields.
*
* tuple is an in-memory tuple structure containing the data to be written
* over the target tuple. Also, tuple->t_self identifies the target tuple.
Avoid using tuple from syscache for update of pg_database.datfrozenxid pg_database.datfrozenxid gets updated using an in-place update at the end of vacuum or autovacuum. Since 96cdeae, as pg_database has a toast relation, it is possible for a pg_database tuple to have toast values if there is a large set of ACLs in place. In such a case, the in-place update would fail because of the flattening of the toast values done for the catcache entry fetched. Instead of using a copy from the catcache, this changes the logic to fetch the copy of the tuple by directly scanning pg_database. Per the lack of complaints on the matter, no backpatch is done. Note that before 96cdeae, attempting to insert such a tuple to pg_database would cause a "row is too big" error, so the end-of-vacuum problem was not reachable. Author: Ashwin Agrawal, Junfeng Yang Discussion: https://postgr.es/m/DM5PR0501MB38800D9E4605BCA72DD35557CCE10@DM5PR0501MB3880.namprd05.prod.outlook.com
2020-12-08 04:13:19 +01:00
*
* Note that the tuple updated here had better not come directly from the
* syscache if the relation has a toast relation as this tuple could
* include toast values that have been expanded, causing a failure here.
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
*/
void
heap_inplace_update(Relation relation, HeapTuple tuple)
{
Buffer buffer;
Page page;
OffsetNumber offnum;
ItemId lp = NULL;
HeapTupleHeader htup;
uint32 oldlen;
uint32 newlen;
Create an infrastructure for parallel computation in PostgreSQL. This does four basic things. First, it provides convenience routines to coordinate the startup and shutdown of parallel workers. Second, it synchronizes various pieces of state (e.g. GUCs, combo CID mappings, transaction snapshot) from the parallel group leader to the worker processes. Third, it prohibits various operations that would result in unsafe changes to that state while parallelism is active. Finally, it propagates events that would result in an ErrorResponse, NoticeResponse, or NotifyResponse message being sent to the client from the parallel workers back to the master, from which they can then be sent on to the client. Robert Haas, Amit Kapila, Noah Misch, Rushabh Lathia, Jeevan Chalke. Suggestions and review from Andres Freund, Heikki Linnakangas, Noah Misch, Simon Riggs, Euler Taveira, and Jim Nasby.
2015-04-30 21:02:14 +02:00
/*
Fix comments in heapam.c. After commits 85f6b49c2c and 3ba59ccc89, we can allow parallel inserts which was earlier not possible as parallel group members won't conflict for relation extension and page lock. In those commits, we forgot to update comments at few places. Author: Amit Kapila Reviewed-by: Robert Haas and Dilip Kumar Backpatch-through: 13 Discussion: https://postgr.es/m/CAFiTN-tMrQh5FFMPx5aWJ+1gi1H6JxktEhq5mDwCHgnEO5oBkA@mail.gmail.com
2020-09-18 06:10:04 +02:00
* For now, we don't allow parallel updates. Unlike a regular update,
* this should never create a combo CID, so it might be possible to relax
* this restriction, but not without more thought and testing. It's not
* clear that it would be useful, anyway.
Create an infrastructure for parallel computation in PostgreSQL. This does four basic things. First, it provides convenience routines to coordinate the startup and shutdown of parallel workers. Second, it synchronizes various pieces of state (e.g. GUCs, combo CID mappings, transaction snapshot) from the parallel group leader to the worker processes. Third, it prohibits various operations that would result in unsafe changes to that state while parallelism is active. Finally, it propagates events that would result in an ErrorResponse, NoticeResponse, or NotifyResponse message being sent to the client from the parallel workers back to the master, from which they can then be sent on to the client. Robert Haas, Amit Kapila, Noah Misch, Rushabh Lathia, Jeevan Chalke. Suggestions and review from Andres Freund, Heikki Linnakangas, Noah Misch, Simon Riggs, Euler Taveira, and Jim Nasby.
2015-04-30 21:02:14 +02:00
*/
if (IsInParallelMode())
ereport(ERROR,
(errcode(ERRCODE_INVALID_TRANSACTION_STATE),
errmsg("cannot update tuples during a parallel operation")));
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&(tuple->t_self)));
LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = (Page) BufferGetPage(buffer);
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
offnum = ItemPointerGetOffsetNumber(&(tuple->t_self));
if (PageGetMaxOffsetNumber(page) >= offnum)
lp = PageGetItemId(page, offnum);
Redefine the lp_flags field of item pointers as having four states, rather than two independent bits (one of which was never used in heap pages anyway, or at least hadn't been in a very long time). This gives us flexibility to add the HOT notions of redirected and dead item pointers without requiring anything so klugy as magic values of lp_off and lp_len. The state values are chosen so that for the states currently in use (pre-HOT) there is no change in the physical representation.
2007-09-13 00:10:26 +02:00
if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(ERROR, "invalid lp");
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
htup = (HeapTupleHeader) PageGetItem(page, lp);
oldlen = ItemIdGetLength(lp) - htup->t_hoff;
newlen = tuple->t_len - tuple->t_data->t_hoff;
if (oldlen != newlen || htup->t_hoff != tuple->t_data->t_hoff)
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(ERROR, "wrong tuple length");
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
/* NO EREPORT(ERROR) from here till changes are logged */
START_CRIT_SECTION();
memcpy((char *) htup + htup->t_hoff,
(char *) tuple->t_data + tuple->t_data->t_hoff,
newlen);
MarkBufferDirty(buffer);
/* XLOG stuff */
Generalize concept of temporary relations to "relation persistence". This commit replaces pg_class.relistemp with pg_class.relpersistence; and also modifies the RangeVar node type to carry relpersistence rather than istemp. It also removes removes rd_istemp from RelationData and instead performs the correct computation based on relpersistence. For clarity, we add three new macros: RelationNeedsWAL(), RelationUsesLocalBuffers(), and RelationUsesTempNamespace(), so that we can clarify the purpose of each check that previous depended on rd_istemp. This is intended as infrastructure for the upcoming unlogged tables patch, as well as for future possible work on global temporary tables.
2010-12-13 18:34:26 +01:00
if (RelationNeedsWAL(relation))
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
{
xl_heap_inplace xlrec;
XLogRecPtr recptr;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHeapInplace);
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
XLogRegisterBufData(0, (char *) htup + htup->t_hoff, newlen);
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
Introduce replication progress tracking infrastructure. When implementing a replication solution ontop of logical decoding, two related problems exist: * How to safely keep track of replication progress * How to change replication behavior, based on the origin of a row; e.g. to avoid loops in bi-directional replication setups The solution to these problems, as implemented here, consist out of three parts: 1) 'replication origins', which identify nodes in a replication setup. 2) 'replication progress tracking', which remembers, for each replication origin, how far replay has progressed in a efficient and crash safe manner. 3) The ability to filter out changes performed on the behest of a replication origin during logical decoding; this allows complex replication topologies. E.g. by filtering all replayed changes out. Most of this could also be implemented in "userspace", e.g. by inserting additional rows contain origin information, but that ends up being much less efficient and more complicated. We don't want to require various replication solutions to reimplement logic for this independently. The infrastructure is intended to be generic enough to be reusable. This infrastructure also replaces the 'nodeid' infrastructure of commit timestamps. It is intended to provide all the former capabilities, except that there's only 2^16 different origins; but now they integrate with logical decoding. Additionally more functionality is accessible via SQL. Since the commit timestamp infrastructure has also been introduced in 9.5 (commit 73c986add) changing the API is not a problem. For now the number of origins for which the replication progress can be tracked simultaneously is determined by the max_replication_slots GUC. That GUC is not a perfect match to configure this, but there doesn't seem to be sufficient reason to introduce a separate new one. Bumps both catversion and wal page magic. Author: Andres Freund, with contributions from Petr Jelinek and Craig Ringer Reviewed-By: Heikki Linnakangas, Petr Jelinek, Robert Haas, Steve Singer Discussion: 20150216002155.GI15326@awork2.anarazel.de, 20140923182422.GA15776@alap3.anarazel.de, 20131114172632.GE7522@alap2.anarazel.de
2015-04-29 19:30:53 +02:00
/* inplace updates aren't decoded atm, don't log the origin */
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_INPLACE);
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
PageSetLSN(page, recptr);
}
END_CRIT_SECTION();
UnlockReleaseBuffer(buffer);
Revise sinval code to remove no-longer-used tuple TID from inval messages. This requires adjusting the API for syscache callback functions: they now get a hash value, not a TID, to identify the target tuple. Most of them weren't paying any attention to that argument anyway, but plancache did require a small amount of fixing. Also, improve performance a trifle by avoiding sending duplicate inval messages when a heap_update isn't changing the catcache lookup columns.
2011-08-17 01:27:46 +02:00
/*
* Send out shared cache inval if necessary. Note that because we only
* pass the new version of the tuple, this mustn't be used for any
* operations that could change catcache lookup keys. But we aren't
* bothering with index updates either, so that's true a fortiori.
*/
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
if (!IsBootstrapProcessingMode())
Revise sinval code to remove no-longer-used tuple TID from inval messages. This requires adjusting the API for syscache callback functions: they now get a hash value, not a TID, to identify the target tuple. Most of them weren't paying any attention to that argument anyway, but plancache did require a small amount of fixing. Also, improve performance a trifle by avoiding sending duplicate inval messages when a heap_update isn't changing the catcache lookup columns.
2011-08-17 01:27:46 +02:00
CacheInvalidateHeapTuple(relation, tuple, NULL);
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
}
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
#define FRM_NOOP 0x0001
#define FRM_INVALIDATE_XMAX 0x0002
#define FRM_RETURN_IS_XID 0x0004
#define FRM_RETURN_IS_MULTI 0x0008
#define FRM_MARK_COMMITTED 0x0010
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
/*
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
* FreezeMultiXactId
* Determine what to do during freezing when a tuple is marked by a
* MultiXactId.
*
* "flags" is an output value; it's used to tell caller what to do on return.
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* "pagefrz" is an input/output value, used to manage page level freezing.
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
*
* Possible values that we can set in "flags":
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
* FRM_NOOP
* don't do anything -- keep existing Xmax
* FRM_INVALIDATE_XMAX
* mark Xmax as InvalidTransactionId and set XMAX_INVALID flag.
* FRM_RETURN_IS_XID
* The Xid return value is a single update Xid to set as xmax.
* FRM_MARK_COMMITTED
* Xmax can be marked as HEAP_XMAX_COMMITTED
* FRM_RETURN_IS_MULTI
* The return value is a new MultiXactId to set as new Xmax.
* (caller must obtain proper infomask bits using GetMultiXactIdHintBits)
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
*
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* Caller delegates control of page freezing to us. In practice we always
* force freezing of caller's page unless FRM_NOOP processing is indicated.
* We help caller ensure that XIDs < FreezeLimit and MXIDs < MultiXactCutoff
* can never be left behind. We freely choose when and how to process each
* Multi, without ever violating the cutoff postconditions for freezing.
*
* It's useful to remove Multis on a proactive timeline (relative to freezing
* XIDs) to keep MultiXact member SLRU buffer misses to a minimum. It can also
* be cheaper in the short run, for us, since we too can avoid SLRU buffer
* misses through eager processing.
*
* NB: Creates a _new_ MultiXactId when FRM_RETURN_IS_MULTI is set, though only
* when FreezeLimit and/or MultiXactCutoff cutoffs leave us with no choice.
* This can usually be put off, which is usually enough to avoid it altogether.
* Allocating new multis during VACUUM should be avoided on general principle;
* only VACUUM can advance relminmxid, so allocating new Multis here comes with
* its own special risks.
*
* NB: Caller must maintain "no freeze" NewRelfrozenXid/NewRelminMxid trackers
* using heap_tuple_should_freeze when we haven't forced page-level freezing.
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
*
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* NB: Caller should avoid needlessly calling heap_tuple_should_freeze when we
* have already forced page-level freezing, since that might incur the same
* SLRU buffer misses that we specifically intended to avoid by freezing.
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
*/
static TransactionId
FreezeMultiXactId(MultiXactId multi, uint16 t_infomask,
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
const struct VacuumCutoffs *cutoffs, uint16 *flags,
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
HeapPageFreeze *pagefrz)
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
{
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
TransactionId newxmax;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
MultiXactMember *members;
int nmembers;
bool need_replace;
int nnewmembers;
MultiXactMember *newmembers;
bool has_lockers;
TransactionId update_xid;
bool update_committed;
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
TransactionId FreezePageRelfrozenXid;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
*flags = 0;
/* We should only be called in Multis */
Assert(t_infomask & HEAP_XMAX_IS_MULTI);
Fix handling of multixacts predating pg_upgrade After pg_upgrade, it is possible that some tuples' Xmax have multixacts corresponding to the old installation; such multixacts cannot have running members anymore. In many code sites we already know not to read them and clobber them silently, but at least when VACUUM tries to freeze a multixact or determine whether one needs freezing, there's an attempt to resolve it to its member transactions by calling GetMultiXactIdMembers, and if the multixact value is "in the future" with regards to the current valid multixact range, an error like this is raised: ERROR: MultiXactId 123 has not been created yet -- apparent wraparound and vacuuming fails. Per discussion with Andrew Gierth, it is completely bogus to try to resolve multixacts coming from before a pg_upgrade, regardless of where they stand with regards to the current valid multixact range. It's possible to get from under this problem by doing SELECT FOR UPDATE of the problem tuples, but if tables are large, this is slow and tedious, so a more thorough solution is desirable. To fix, we realize that multixacts in xmax created in 9.2 and previous have a specific bit pattern that is never used in 9.3 and later (we already knew this, per comments and infomask tests sprinkled in various places, but we weren't leveraging this knowledge appropriately). Whenever the infomask of the tuple matches that bit pattern, we just ignore the multixact completely as if Xmax wasn't set; or, in the case of tuple freezing, we act as if an unwanted value is set and clobber it without decoding. This guarantees that no errors will be raised, and that the values will be progressively removed until all tables are clean. Most callers of GetMultiXactIdMembers are patched to recognize directly that the value is a removable "empty" multixact and avoid calling GetMultiXactIdMembers altogether. To avoid changing the signature of GetMultiXactIdMembers() in back branches, we keep the "allow_old" boolean flag but rename it to "from_pgupgrade"; if the flag is true, we always return an empty set instead of looking up the multixact. (I suppose we could remove the argument in the master branch, but I chose not to do so in this commit). This was broken all along, but the error-facing message appeared first because of commit 8e9a16ab8f7f and was partially fixed in a25c2b7c4db3. This fix, backpatched all the way back to 9.3, goes approximately in the same direction as a25c2b7c4db3 but should cover all cases. Bug analysis by Andrew Gierth and Álvaro Herrera. A number of public reports match this bug: https://www.postgresql.org/message-id/20140330040029.GY4582@tamriel.snowman.net https://www.postgresql.org/message-id/538F3D70.6080902@publicrelay.com https://www.postgresql.org/message-id/556439CF.7070109@pscs.co.uk https://www.postgresql.org/message-id/SG2PR06MB0760098A111C88E31BD4D96FB3540@SG2PR06MB0760.apcprd06.prod.outlook.com https://www.postgresql.org/message-id/20160615203829.5798.4594@wrigleys.postgresql.org
2016-06-25 00:29:28 +02:00
if (!MultiXactIdIsValid(multi) ||
HEAP_LOCKED_UPGRADED(t_infomask))
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
{
*flags |= FRM_INVALIDATE_XMAX;
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
pagefrz->freeze_required = true;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
return InvalidTransactionId;
}
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
else if (MultiXactIdPrecedes(multi, cutoffs->relminmxid))
Perform a lot more sanity checks when freezing tuples. The previous commit has shown that the sanity checks around freezing aren't strong enough. Strengthening them seems especially important because the existance of the bug has caused corruption that we don't want to make even worse during future vacuum cycles. The errors are emitted with ereport rather than elog, despite being "should never happen" messages, so a proper error code is emitted. To avoid superflous translations, mark messages as internal. Author: Andres Freund and Alvaro Herrera Reviewed-By: Alvaro Herrera, Michael Paquier Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de Backpatch: 9.3-
2017-11-14 03:45:47 +01:00
ereport(ERROR,
(errcode(ERRCODE_DATA_CORRUPTED),
errmsg_internal("found multixact %u from before relminmxid %u",
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
multi, cutoffs->relminmxid)));
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
else if (MultiXactIdPrecedes(multi, cutoffs->OldestMxact))
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
{
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
TransactionId update_xact;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
/*
Perform a lot more sanity checks when freezing tuples. The previous commit has shown that the sanity checks around freezing aren't strong enough. Strengthening them seems especially important because the existance of the bug has caused corruption that we don't want to make even worse during future vacuum cycles. The errors are emitted with ereport rather than elog, despite being "should never happen" messages, so a proper error code is emitted. To avoid superflous translations, mark messages as internal. Author: Andres Freund and Alvaro Herrera Reviewed-By: Alvaro Herrera, Michael Paquier Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de Backpatch: 9.3-
2017-11-14 03:45:47 +01:00
* This old multi cannot possibly have members still running, but
* verify just in case. If it was a locker only, it can be removed
* without any further consideration; but if it contained an update,
* we might need to preserve it.
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
*/
Perform a lot more sanity checks when freezing tuples. The previous commit has shown that the sanity checks around freezing aren't strong enough. Strengthening them seems especially important because the existance of the bug has caused corruption that we don't want to make even worse during future vacuum cycles. The errors are emitted with ereport rather than elog, despite being "should never happen" messages, so a proper error code is emitted. To avoid superflous translations, mark messages as internal. Author: Andres Freund and Alvaro Herrera Reviewed-By: Alvaro Herrera, Michael Paquier Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de Backpatch: 9.3-
2017-11-14 03:45:47 +01:00
if (MultiXactIdIsRunning(multi,
HEAP_XMAX_IS_LOCKED_ONLY(t_infomask)))
ereport(ERROR,
(errcode(ERRCODE_DATA_CORRUPTED),
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
errmsg_internal("multixact %u from before multi freeze cutoff %u found to be still running",
multi, cutoffs->OldestMxact)));
Perform a lot more sanity checks when freezing tuples. The previous commit has shown that the sanity checks around freezing aren't strong enough. Strengthening them seems especially important because the existance of the bug has caused corruption that we don't want to make even worse during future vacuum cycles. The errors are emitted with ereport rather than elog, despite being "should never happen" messages, so a proper error code is emitted. To avoid superflous translations, mark messages as internal. Author: Andres Freund and Alvaro Herrera Reviewed-By: Alvaro Herrera, Michael Paquier Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de Backpatch: 9.3-
2017-11-14 03:45:47 +01:00
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
if (HEAP_XMAX_IS_LOCKED_ONLY(t_infomask))
{
*flags |= FRM_INVALIDATE_XMAX;
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
pagefrz->freeze_required = true;
return InvalidTransactionId;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
}
Perform a lot more sanity checks when freezing tuples. The previous commit has shown that the sanity checks around freezing aren't strong enough. Strengthening them seems especially important because the existance of the bug has caused corruption that we don't want to make even worse during future vacuum cycles. The errors are emitted with ereport rather than elog, despite being "should never happen" messages, so a proper error code is emitted. To avoid superflous translations, mark messages as internal. Author: Andres Freund and Alvaro Herrera Reviewed-By: Alvaro Herrera, Michael Paquier Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de Backpatch: 9.3-
2017-11-14 03:45:47 +01:00
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
/* replace multi with single XID for its updater? */
update_xact = MultiXactIdGetUpdateXid(multi, t_infomask);
if (TransactionIdPrecedes(update_xact, cutoffs->relfrozenxid))
ereport(ERROR,
(errcode(ERRCODE_DATA_CORRUPTED),
errmsg_internal("multixact %u contains update XID %u from before relfrozenxid %u",
multi, update_xact,
cutoffs->relfrozenxid)));
else if (TransactionIdPrecedes(update_xact, cutoffs->OldestXmin))
{
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
/*
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* Updater XID has to have aborted (otherwise the tuple would have
* been pruned away instead, since updater XID is < OldestXmin).
* Just remove xmax.
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
*/
Check that xmax didn't commit in freeze check. We cannot rely on TransactionIdDidAbort here, since in general it may report transactions that were in-progress at the time of an earlier hard crash as not aborted, effectively behaving as if they were still in progress even after crash recovery completes. Go back to defensively verifying that xmax didn't commit instead. Oversight in commit 79d4bf4e. Author: Peter Geoghegan <pg@bowt.ie> Reported-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/20230104035636.hy5djyr2as4gbc4q@awork3.anarazel.de
2023-01-04 06:48:27 +01:00
if (TransactionIdDidCommit(update_xact))
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
ereport(ERROR,
(errcode(ERRCODE_DATA_CORRUPTED),
Check that xmax didn't commit in freeze check. We cannot rely on TransactionIdDidAbort here, since in general it may report transactions that were in-progress at the time of an earlier hard crash as not aborted, effectively behaving as if they were still in progress even after crash recovery completes. Go back to defensively verifying that xmax didn't commit instead. Oversight in commit 79d4bf4e. Author: Peter Geoghegan <pg@bowt.ie> Reported-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/20230104035636.hy5djyr2as4gbc4q@awork3.anarazel.de
2023-01-04 06:48:27 +01:00
errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
multi, update_xact,
cutoffs->OldestXmin)));
*flags |= FRM_INVALIDATE_XMAX;
pagefrz->freeze_required = true;
return InvalidTransactionId;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
}
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
/* Have to keep updater XID as new xmax */
*flags |= FRM_RETURN_IS_XID;
pagefrz->freeze_required = true;
return update_xact;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
}
/*
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
* Some member(s) of this Multi may be below FreezeLimit xid cutoff, so we
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
* need to walk the whole members array to figure out what to do, if
* anything.
*/
Avoid uselessly looking up old LOCK_ONLY multixacts Commit 0ac5ad5134f2 removed an optimization in multixact.c that skipped fetching members of MultiXactId that were older than our OldestVisibleMXactId value. The reason this was removed is that it is possible for multixacts that contain updates to be older than that value. However, if the caller is certain that the multi does not contain an update (because the infomask bits say so), it can pass this info down to GetMultiXactIdMembers, enabling it to use the old optimization. Pointed out by Andres Freund in 20131121200517.GM7240@alap2.anarazel.de
2014-07-29 21:41:06 +02:00
nmembers =
Fix handling of multixacts predating pg_upgrade After pg_upgrade, it is possible that some tuples' Xmax have multixacts corresponding to the old installation; such multixacts cannot have running members anymore. In many code sites we already know not to read them and clobber them silently, but at least when VACUUM tries to freeze a multixact or determine whether one needs freezing, there's an attempt to resolve it to its member transactions by calling GetMultiXactIdMembers, and if the multixact value is "in the future" with regards to the current valid multixact range, an error like this is raised: ERROR: MultiXactId 123 has not been created yet -- apparent wraparound and vacuuming fails. Per discussion with Andrew Gierth, it is completely bogus to try to resolve multixacts coming from before a pg_upgrade, regardless of where they stand with regards to the current valid multixact range. It's possible to get from under this problem by doing SELECT FOR UPDATE of the problem tuples, but if tables are large, this is slow and tedious, so a more thorough solution is desirable. To fix, we realize that multixacts in xmax created in 9.2 and previous have a specific bit pattern that is never used in 9.3 and later (we already knew this, per comments and infomask tests sprinkled in various places, but we weren't leveraging this knowledge appropriately). Whenever the infomask of the tuple matches that bit pattern, we just ignore the multixact completely as if Xmax wasn't set; or, in the case of tuple freezing, we act as if an unwanted value is set and clobber it without decoding. This guarantees that no errors will be raised, and that the values will be progressively removed until all tables are clean. Most callers of GetMultiXactIdMembers are patched to recognize directly that the value is a removable "empty" multixact and avoid calling GetMultiXactIdMembers altogether. To avoid changing the signature of GetMultiXactIdMembers() in back branches, we keep the "allow_old" boolean flag but rename it to "from_pgupgrade"; if the flag is true, we always return an empty set instead of looking up the multixact. (I suppose we could remove the argument in the master branch, but I chose not to do so in this commit). This was broken all along, but the error-facing message appeared first because of commit 8e9a16ab8f7f and was partially fixed in a25c2b7c4db3. This fix, backpatched all the way back to 9.3, goes approximately in the same direction as a25c2b7c4db3 but should cover all cases. Bug analysis by Andrew Gierth and Álvaro Herrera. A number of public reports match this bug: https://www.postgresql.org/message-id/20140330040029.GY4582@tamriel.snowman.net https://www.postgresql.org/message-id/538F3D70.6080902@publicrelay.com https://www.postgresql.org/message-id/556439CF.7070109@pscs.co.uk https://www.postgresql.org/message-id/SG2PR06MB0760098A111C88E31BD4D96FB3540@SG2PR06MB0760.apcprd06.prod.outlook.com https://www.postgresql.org/message-id/20160615203829.5798.4594@wrigleys.postgresql.org
2016-06-25 00:29:28 +02:00
GetMultiXactIdMembers(multi, &members, false,
Avoid uselessly looking up old LOCK_ONLY multixacts Commit 0ac5ad5134f2 removed an optimization in multixact.c that skipped fetching members of MultiXactId that were older than our OldestVisibleMXactId value. The reason this was removed is that it is possible for multixacts that contain updates to be older than that value. However, if the caller is certain that the multi does not contain an update (because the infomask bits say so), it can pass this info down to GetMultiXactIdMembers, enabling it to use the old optimization. Pointed out by Andres Freund in 20131121200517.GM7240@alap2.anarazel.de
2014-07-29 21:41:06 +02:00
HEAP_XMAX_IS_LOCKED_ONLY(t_infomask));
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
if (nmembers <= 0)
{
/* Nothing worth keeping */
*flags |= FRM_INVALIDATE_XMAX;
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
pagefrz->freeze_required = true;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
return InvalidTransactionId;
}
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
/*
* The FRM_NOOP case is the only case where we might need to ratchet back
* FreezePageRelfrozenXid or FreezePageRelminMxid. It is also the only
* case where our caller might ratchet back its NoFreezePageRelfrozenXid
* or NoFreezePageRelminMxid "no freeze" trackers to deal with a multi.
* FRM_NOOP handling should result in the NewRelfrozenXid/NewRelminMxid
* trackers managed by VACUUM being ratcheting back by xmax to the degree
* required to make it safe to leave xmax undisturbed, independent of
* whether or not page freezing is triggered somewhere else.
*
* Our policy is to force freezing in every case other than FRM_NOOP,
* which obviates the need to maintain either set of trackers, anywhere.
* Every other case will reliably execute a freeze plan for xmax that
* either replaces xmax with an XID/MXID >= OldestXmin/OldestMxact, or
* sets xmax to an InvalidTransactionId XID, rendering xmax fully frozen.
* (VACUUM's NewRelfrozenXid/NewRelminMxid trackers are initialized with
* OldestXmin/OldestMxact, so later values never need to be tracked here.)
*/
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
need_replace = false;
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
FreezePageRelfrozenXid = pagefrz->FreezePageRelfrozenXid;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
for (int i = 0; i < nmembers; i++)
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
{
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
TransactionId xid = members[i].xid;
Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
{
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
/* Can't violate the FreezeLimit postcondition */
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
need_replace = true;
break;
}
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
if (TransactionIdPrecedes(xid, FreezePageRelfrozenXid))
FreezePageRelfrozenXid = xid;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
}
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
/* Can't violate the MultiXactCutoff postcondition, either */
if (!need_replace)
need_replace = MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff);
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
if (!need_replace)
{
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
/*
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* vacuumlazy.c might ratchet back NewRelminMxid, NewRelfrozenXid, or
* both together to make it safe to retain this particular multi after
* freezing its page
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
*/
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
*flags |= FRM_NOOP;
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
pagefrz->FreezePageRelfrozenXid = FreezePageRelfrozenXid;
if (MultiXactIdPrecedes(multi, pagefrz->FreezePageRelminMxid))
pagefrz->FreezePageRelminMxid = multi;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
pfree(members);
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
return multi;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
}
/*
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
* Do a more thorough second pass over the multi to figure out which
* member XIDs actually need to be kept. Checking the precise status of
* individual members might even show that we don't need to keep anything.
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* That is quite possible even though the Multi must be >= OldestMxact,
* since our second pass only keeps member XIDs when it's truly necessary;
* even member XIDs >= OldestXmin often won't be kept by second pass.
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
*/
nnewmembers = 0;
newmembers = palloc(sizeof(MultiXactMember) * nmembers);
has_lockers = false;
update_xid = InvalidTransactionId;
update_committed = false;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
/*
* Determine whether to keep each member xid, or to ignore it instead
*/
for (int i = 0; i < nmembers; i++)
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
{
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
TransactionId xid = members[i].xid;
MultiXactStatus mstatus = members[i].status;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
Perform a lot more sanity checks when freezing tuples. The previous commit has shown that the sanity checks around freezing aren't strong enough. Strengthening them seems especially important because the existance of the bug has caused corruption that we don't want to make even worse during future vacuum cycles. The errors are emitted with ereport rather than elog, despite being "should never happen" messages, so a proper error code is emitted. To avoid superflous translations, mark messages as internal. Author: Andres Freund and Alvaro Herrera Reviewed-By: Alvaro Herrera, Michael Paquier Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de Backpatch: 9.3-
2017-11-14 03:45:47 +01:00
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
if (!ISUPDATE_from_mxstatus(mstatus))
{
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
/*
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
* Locker XID (not updater XID). We only keep lockers that are
* still running.
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
*/
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
if (TransactionIdIsCurrentTransactionId(xid) ||
TransactionIdIsInProgress(xid))
Perform a lot more sanity checks when freezing tuples. The previous commit has shown that the sanity checks around freezing aren't strong enough. Strengthening them seems especially important because the existance of the bug has caused corruption that we don't want to make even worse during future vacuum cycles. The errors are emitted with ereport rather than elog, despite being "should never happen" messages, so a proper error code is emitted. To avoid superflous translations, mark messages as internal. Author: Andres Freund and Alvaro Herrera Reviewed-By: Alvaro Herrera, Michael Paquier Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de Backpatch: 9.3-
2017-11-14 03:45:47 +01:00
{
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
ereport(ERROR,
(errcode(ERRCODE_DATA_CORRUPTED),
errmsg_internal("multixact %u contains running locker XID %u from before removable cutoff %u",
multi, xid,
cutoffs->OldestXmin)));
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
newmembers[nnewmembers++] = members[i];
has_lockers = true;
Perform a lot more sanity checks when freezing tuples. The previous commit has shown that the sanity checks around freezing aren't strong enough. Strengthening them seems especially important because the existance of the bug has caused corruption that we don't want to make even worse during future vacuum cycles. The errors are emitted with ereport rather than elog, despite being "should never happen" messages, so a proper error code is emitted. To avoid superflous translations, mark messages as internal. Author: Andres Freund and Alvaro Herrera Reviewed-By: Alvaro Herrera, Michael Paquier Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de Backpatch: 9.3-
2017-11-14 03:45:47 +01:00
}
Simplify multixact freezing a bit Testing for abortedness of a multixact member that's being frozen is unnecessary: we only need to know whether the transaction is still in progress or committed to determine whether it must be kept or not. This let us simplify the code a bit and avoid a useless TransactionIdDidAbort test. Suggested by Andres Freund awhile back.
2014-07-29 21:40:55 +02:00
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
continue;
}
/*
* Updater XID (not locker XID). Should we keep it?
*
* Since the tuple wasn't totally removed when vacuum pruned, the
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* update Xid cannot possibly be older than OldestXmin cutoff unless
* the updater XID aborted. If the updater transaction is known
* aborted or crashed then it's okay to ignore it, otherwise not.
*
* In any case the Multi should never contain two updaters, whatever
* their individual commit status. Check for that first, in passing.
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
*/
if (TransactionIdIsValid(update_xid))
ereport(ERROR,
(errcode(ERRCODE_DATA_CORRUPTED),
errmsg_internal("multixact %u has two or more updating members",
multi),
errdetail_internal("First updater XID=%u second updater XID=%u.",
update_xid, xid)));
Revert bogus fixes of HOT-freezing bug It turns out we misdiagnosed what the real problem was. Revert the previous changes, because they may have worse consequences going forward. A better fix is forthcoming. The simplistic test case is kept, though disabled. Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de
2017-11-02 15:51:05 +01:00
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
/*
* As with all tuple visibility routines, it's critical to test
* TransactionIdIsInProgress before TransactionIdDidCommit, because of
* race conditions explained in detail in heapam_visibility.c.
*/
if (TransactionIdIsCurrentTransactionId(xid) ||
TransactionIdIsInProgress(xid))
update_xid = xid;
else if (TransactionIdDidCommit(xid))
{
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
/*
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
* The transaction committed, so we can tell caller to set
* HEAP_XMAX_COMMITTED. (We can only do this because we know the
* transaction is not running.)
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
*/
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
update_committed = true;
update_xid = xid;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
}
else
{
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
/*
* Not in progress, not committed -- must be aborted or crashed;
* we can ignore it.
*/
continue;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
}
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
/*
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* We determined that updater must be kept -- add it to pending new
* members list
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
*/
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
ereport(ERROR,
(errcode(ERRCODE_DATA_CORRUPTED),
errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
multi, xid, cutoffs->OldestXmin)));
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
newmembers[nnewmembers++] = members[i];
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
}
pfree(members);
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
/*
* Determine what to do with caller's multi based on information gathered
* during our second pass
*/
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
if (nnewmembers == 0)
{
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
/* Nothing worth keeping */
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
*flags |= FRM_INVALIDATE_XMAX;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
newxmax = InvalidTransactionId;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
}
else if (TransactionIdIsValid(update_xid) && !has_lockers)
{
/*
* If there's a single member and it's an update, pass it back alone
* without creating a new Multi. (XXX we could do this when there's a
* single remaining locker, too, but that would complicate the API too
* much; moreover, the case with the single updater is more
* interesting, because those are longer-lived.)
*/
Assert(nnewmembers == 1);
*flags |= FRM_RETURN_IS_XID;
if (update_committed)
*flags |= FRM_MARK_COMMITTED;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
newxmax = update_xid;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
}
else
{
/*
* Create a new multixact with the surviving members of the previous
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* one, to set as new Xmax in the tuple
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
*/
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
newxmax = MultiXactIdCreateFromMembers(nnewmembers, newmembers);
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
*flags |= FRM_RETURN_IS_MULTI;
}
pfree(newmembers);
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
pagefrz->freeze_required = true;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
return newxmax;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
}
/*
* heap_prepare_freeze_tuple
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
*
* Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* are older than the OldestXmin and/or OldestMxact freeze cutoffs. If so,
* setup enough state (in the *frz output argument) to enable caller to
* process this tuple as part of freezing its page, and return true. Return
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
* false if nothing can be changed about the tuple right now.
*
* Also sets *totally_frozen to true if the tuple will be totally frozen once
* caller executes returned freeze plan (or if the tuple was already totally
* frozen by an earlier VACUUM). This indicates that there are no remaining
* XIDs or MultiXactIds that will need to be processed by a future VACUUM.
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
*
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* VACUUM caller must assemble HeapTupleFreeze freeze plan entries for every
* tuple that we returned true for, and call heap_freeze_execute_prepared to
* execute freezing. Caller must initialize pagefrz fields for page as a
* whole before first call here for each heap page.
*
* VACUUM caller decides on whether or not to freeze the page as a whole.
* We'll often prepare freeze plans for a page that caller just discards.
* However, VACUUM doesn't always get to make a choice; it must freeze when
* pagefrz.freeze_required is set, to ensure that any XIDs < FreezeLimit (and
* MXIDs < MultiXactCutoff) can never be left behind. We help to make sure
* that VACUUM always follows that rule.
*
* We sometimes force freezing of xmax MultiXactId values long before it is
* strictly necessary to do so just to ensure the FreezeLimit postcondition.
* It's worth processing MultiXactIds proactively when it is cheap to do so,
* and it's convenient to make that happen by piggy-backing it on the "force
* freezing" mechanism. Conversely, we sometimes delay freezing MultiXactIds
* because it is expensive right now (though only when it's still possible to
* do so without violating the FreezeLimit/MultiXactCutoff postcondition).
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
*
* It is assumed that the caller has checked the tuple with
* HeapTupleSatisfiesVacuum() and determined that it is not HEAPTUPLE_DEAD
Revert bogus fixes of HOT-freezing bug It turns out we misdiagnosed what the real problem was. Revert the previous changes, because they may have worse consequences going forward. A better fix is forthcoming. The simplistic test case is kept, though disabled. Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de
2017-11-02 15:51:05 +01:00
* (else we should be removing the tuple, not freezing it).
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
*
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
* NB: This function has side effects: it might allocate a new MultiXactId.
* It will be set as tuple's new xmax when our *frz output is processed within
* heap_execute_freeze_tuple later on. If the tuple is in a shared buffer
* then caller had better have an exclusive lock on it already.
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
*/
bool
Perform a lot more sanity checks when freezing tuples. The previous commit has shown that the sanity checks around freezing aren't strong enough. Strengthening them seems especially important because the existance of the bug has caused corruption that we don't want to make even worse during future vacuum cycles. The errors are emitted with ereport rather than elog, despite being "should never happen" messages, so a proper error code is emitted. To avoid superflous translations, mark messages as internal. Author: Andres Freund and Alvaro Herrera Reviewed-By: Alvaro Herrera, Michael Paquier Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de Backpatch: 9.3-
2017-11-14 03:45:47 +01:00
heap_prepare_freeze_tuple(HeapTupleHeader tuple,
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
const struct VacuumCutoffs *cutoffs,
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
HeapPageFreeze *pagefrz,
HeapTupleFreeze *frz, bool *totally_frozen)
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
{
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
bool xmin_already_frozen = false,
xmax_already_frozen = false;
bool freeze_xmin = false,
replace_xvac = false,
replace_xmax = false,
freeze_xmax = false;
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
TransactionId xid;
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
frz->xmax = HeapTupleHeaderGetRawXmax(tuple);
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
frz->t_infomask2 = tuple->t_infomask2;
frz->t_infomask = tuple->t_infomask;
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
frz->frzflags = 0;
frz->checkflags = 0;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
heap_prepare_freeze_tuple: Simplify coding Commit d2599ecfcc74 introduced some contorted, confused code around: readers would think that it's possible for HeapTupleHeaderGetXmin return a non-frozen value for some frozen tuples, which would be disastrous. There's no actual bug, but it seems better to make it clearer. Per gripe from Tom Lane and Andres Freund. Discussion: https://postgr.es/m/30116.1555430496@sss.pgh.pa.us
2019-05-02 22:13:48 +02:00
/*
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
* Process xmin, while keeping track of whether it's already frozen, or
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* will become frozen iff our freeze plan is executed by caller (could be
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
* neither).
heap_prepare_freeze_tuple: Simplify coding Commit d2599ecfcc74 introduced some contorted, confused code around: readers would think that it's possible for HeapTupleHeaderGetXmin return a non-frozen value for some frozen tuples, which would be disastrous. There's no actual bug, but it seems better to make it clearer. Per gripe from Tom Lane and Andres Freund. Discussion: https://postgr.es/m/30116.1555430496@sss.pgh.pa.us
2019-05-02 22:13:48 +02:00
*/
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
xid = HeapTupleHeaderGetXmin(tuple);
heap_prepare_freeze_tuple: Simplify coding Commit d2599ecfcc74 introduced some contorted, confused code around: readers would think that it's possible for HeapTupleHeaderGetXmin return a non-frozen value for some frozen tuples, which would be disastrous. There's no actual bug, but it seems better to make it clearer. Per gripe from Tom Lane and Andres Freund. Discussion: https://postgr.es/m/30116.1555430496@sss.pgh.pa.us
2019-05-02 22:13:48 +02:00
if (!TransactionIdIsNormal(xid))
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
xmin_already_frozen = true;
heap_prepare_freeze_tuple: Simplify coding Commit d2599ecfcc74 introduced some contorted, confused code around: readers would think that it's possible for HeapTupleHeaderGetXmin return a non-frozen value for some frozen tuples, which would be disastrous. There's no actual bug, but it seems better to make it clearer. Per gripe from Tom Lane and Andres Freund. Discussion: https://postgr.es/m/30116.1555430496@sss.pgh.pa.us
2019-05-02 22:13:48 +02:00
else
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
{
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
Perform a lot more sanity checks when freezing tuples. The previous commit has shown that the sanity checks around freezing aren't strong enough. Strengthening them seems especially important because the existance of the bug has caused corruption that we don't want to make even worse during future vacuum cycles. The errors are emitted with ereport rather than elog, despite being "should never happen" messages, so a proper error code is emitted. To avoid superflous translations, mark messages as internal. Author: Andres Freund and Alvaro Herrera Reviewed-By: Alvaro Herrera, Michael Paquier Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de Backpatch: 9.3-
2017-11-14 03:45:47 +01:00
ereport(ERROR,
(errcode(ERRCODE_DATA_CORRUPTED),
errmsg_internal("found xmin %u from before relfrozenxid %u",
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
xid, cutoffs->relfrozenxid)));
Perform a lot more sanity checks when freezing tuples. The previous commit has shown that the sanity checks around freezing aren't strong enough. Strengthening them seems especially important because the existance of the bug has caused corruption that we don't want to make even worse during future vacuum cycles. The errors are emitted with ereport rather than elog, despite being "should never happen" messages, so a proper error code is emitted. To avoid superflous translations, mark messages as internal. Author: Andres Freund and Alvaro Herrera Reviewed-By: Alvaro Herrera, Michael Paquier Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de Backpatch: 9.3-
2017-11-14 03:45:47 +01:00
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
/* Will set freeze_xmin flags in freeze plan below */
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
freeze_xmin = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
/* Verify that xmin committed if and when freeze plan is executed */
if (freeze_xmin)
frz->checkflags |= HEAP_FREEZE_CHECK_XMIN_COMMITTED;
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
}
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
/*
* Old-style VACUUM FULL is gone, but we have to process xvac for as long
* as we support having MOVED_OFF/MOVED_IN tuples in the database
*/
xid = HeapTupleHeaderGetXvac(tuple);
if (TransactionIdIsNormal(xid))
{
Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
Assert(TransactionIdPrecedes(xid, cutoffs->OldestXmin));
/*
* For Xvac, we always freeze proactively. This allows totally_frozen
* tracking to ignore xvac.
*/
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
replace_xvac = pagefrz->freeze_required = true;
/* Will set replace_xvac flags in freeze plan below */
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
}
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
/* Now process xmax */
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
xid = frz->xmax;
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
{
Don't test HEAP_XMAX_INVALID when freezing xmax. We shouldn't ever need to rely on whether HEAP_XMAX_INVALID is set in t_infomask when considering whether or not an xmax should be deemed already frozen, since that status flag is just a hint. The only acceptable representation for an "xmax_already_frozen" raw xmax field is the transaction ID value zero (also known as InvalidTransactionId). Adjust code that superficially appeared to rely on HEAP_XMAX_INVALID to make the rule about xmax_already_frozen clear. Also avoid needlessly rereading the tuple's raw xmax. Oversight in bugfix commit d2599ecf. There is no evidence that this ever led to incorrect behavior, so no backpatch. The worst consequence of this bug was that VACUUM could hypothetically fail to notice and report on certain kinds of corruption, which seems fairly benign. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wzkh3DMCDRPfhZxj9xCq9v3WmzvmbiCpf1dNKUBPadhCbQ@mail.gmail.com
2022-11-23 19:49:39 +01:00
/* Raw xmax is a MultiXactId */
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
TransactionId newxmax;
uint16 flags;
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
/*
* We will either remove xmax completely (in the "freeze_xmax" path),
* process xmax by replacing it (in the "replace_xmax" path), or
* perform no-op xmax processing. The only constraint is that the
* FreezeLimit/MultiXactCutoff postcondition must never be violated.
*/
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
newxmax = FreezeMultiXactId(xid, tuple->t_infomask, cutoffs,
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
&flags, pagefrz);
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
if (flags & FRM_NOOP)
{
/*
* xmax is a MultiXactId, and nothing about it changes for now.
* This is the only case where 'freeze_required' won't have been
* set for us by FreezeMultiXactId, as well as the only case where
* neither freeze_xmax nor replace_xmax are set (given a multi).
*
* This is a no-op, but the call to FreezeMultiXactId might have
* ratcheted back NewRelfrozenXid and/or NewRelminMxid trackers
* for us (the "freeze page" variants, specifically). That'll
* make it safe for our caller to freeze the page later on, while
* leaving this particular xmax undisturbed.
*
* FreezeMultiXactId is _not_ responsible for the "no freeze"
* NewRelfrozenXid/NewRelminMxid trackers, though -- that's our
* job. A call to heap_tuple_should_freeze for this same tuple
* will take place below if 'freeze_required' isn't set already.
* (This repeats work from FreezeMultiXactId, but allows "no
* freeze" tracker maintenance to happen in only one place.)
*/
Assert(!MultiXactIdPrecedes(newxmax, cutoffs->MultiXactCutoff));
Assert(MultiXactIdIsValid(newxmax) && xid == newxmax);
}
else if (flags & FRM_RETURN_IS_XID)
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
{
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
/*
* xmax will become an updater Xid (original MultiXact's updater
* member Xid will be carried forward as a simple Xid in Xmax).
*/
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
Assert(!TransactionIdPrecedes(newxmax, cutoffs->OldestXmin));
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
/*
* NB -- some of these transformations are only valid because we
* know the return Xid is a tuple updater (i.e. not merely a
Fix misc typos. Oskari Saarenmaa. Backpatch to stable branches where applicable.
2015-09-05 10:35:49 +02:00
* locker.) Also note that the only reason we don't explicitly
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
* worry about HEAP_KEYS_UPDATED is because it lives in
* t_infomask2 rather than t_infomask.
*/
frz->t_infomask &= ~HEAP_XMAX_BITS;
frz->xmax = newxmax;
if (flags & FRM_MARK_COMMITTED)
Fix potential data corruption during freeze Fix oversight in 3b97e6823b94 bug fix. Bitwise AND is used instead of OR and it cleans all bits in t_infomask heap tuple field. Backpatch to 9.3
2017-07-06 16:18:55 +02:00
frz->t_infomask |= HEAP_XMAX_COMMITTED;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
replace_xmax = true;
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
}
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
else if (flags & FRM_RETURN_IS_MULTI)
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
{
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
uint16 newbits;
uint16 newbits2;
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
/*
* xmax is an old MultiXactId that we have to replace with a new
* MultiXactId, to carry forward two or more original member XIDs.
*/
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
Assert(!MultiXactIdPrecedes(newxmax, cutoffs->OldestMxact));
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
/*
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
* We can't use GetMultiXactIdHintBits directly on the new multi
* here; that routine initializes the masks to all zeroes, which
* would lose other bits we need. Doing it this way ensures all
* unrelated bits remain untouched.
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
*/
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
frz->t_infomask &= ~HEAP_XMAX_BITS;
frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
GetMultiXactIdHintBits(newxmax, &newbits, &newbits2);
frz->t_infomask |= newbits;
frz->t_infomask2 |= newbits2;
frz->xmax = newxmax;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
replace_xmax = true;
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
}
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
else
{
/*
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
* Freeze plan for tuple "freezes xmax" in the strictest sense:
* it'll leave nothing in xmax (neither an Xid nor a MultiXactId).
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
*/
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
Assert(flags & FRM_INVALIDATE_XMAX);
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
Assert(!TransactionIdIsValid(newxmax));
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
/* Will set freeze_xmax flags in freeze plan below */
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
freeze_xmax = true;
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
}
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
/* MultiXactId processing forces freezing (barring FRM_NOOP case) */
Assert(pagefrz->freeze_required || (!freeze_xmax && !replace_xmax));
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
}
Fix lazy_scan_heap so that it won't mark pages all-frozen too soon. Commit a892234f830e832110f63fc0a2afce2fb21d1584 added a new bit per page to the visibility map fork indicating whether the page is all-frozen, but incorrectly assumed that if lazy_scan_heap chose to freeze a tuple then that tuple would not need to later be frozen again. This turns out to be false, because xmin and xmax (and conceivably xvac, if dealing with tuples from very old releases) could be frozen at separate times. Thanks to Andres Freund for help in uncovering and tracking down this issue.
2016-06-15 20:23:39 +02:00
else if (TransactionIdIsNormal(xid))
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
{
Don't test HEAP_XMAX_INVALID when freezing xmax. We shouldn't ever need to rely on whether HEAP_XMAX_INVALID is set in t_infomask when considering whether or not an xmax should be deemed already frozen, since that status flag is just a hint. The only acceptable representation for an "xmax_already_frozen" raw xmax field is the transaction ID value zero (also known as InvalidTransactionId). Adjust code that superficially appeared to rely on HEAP_XMAX_INVALID to make the rule about xmax_already_frozen clear. Also avoid needlessly rereading the tuple's raw xmax. Oversight in bugfix commit d2599ecf. There is no evidence that this ever led to incorrect behavior, so no backpatch. The worst consequence of this bug was that VACUUM could hypothetically fail to notice and report on certain kinds of corruption, which seems fairly benign. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wzkh3DMCDRPfhZxj9xCq9v3WmzvmbiCpf1dNKUBPadhCbQ@mail.gmail.com
2022-11-23 19:49:39 +01:00
/* Raw xmax is normal XID */
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
Perform a lot more sanity checks when freezing tuples. The previous commit has shown that the sanity checks around freezing aren't strong enough. Strengthening them seems especially important because the existance of the bug has caused corruption that we don't want to make even worse during future vacuum cycles. The errors are emitted with ereport rather than elog, despite being "should never happen" messages, so a proper error code is emitted. To avoid superflous translations, mark messages as internal. Author: Andres Freund and Alvaro Herrera Reviewed-By: Alvaro Herrera, Michael Paquier Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de Backpatch: 9.3-
2017-11-14 03:45:47 +01:00
ereport(ERROR,
(errcode(ERRCODE_DATA_CORRUPTED),
errmsg_internal("found xmax %u from before relfrozenxid %u",
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
xid, cutoffs->relfrozenxid)));
Perform a lot more sanity checks when freezing tuples. The previous commit has shown that the sanity checks around freezing aren't strong enough. Strengthening them seems especially important because the existance of the bug has caused corruption that we don't want to make even worse during future vacuum cycles. The errors are emitted with ereport rather than elog, despite being "should never happen" messages, so a proper error code is emitted. To avoid superflous translations, mark messages as internal. Author: Andres Freund and Alvaro Herrera Reviewed-By: Alvaro Herrera, Michael Paquier Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de Backpatch: 9.3-
2017-11-14 03:45:47 +01:00
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
/* Will set freeze_xmax flags in freeze plan below */
freeze_xmax = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
/*
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
* Verify that xmax aborted if and when freeze plan is executed,
* provided it's from an update. (A lock-only xmax can be removed
* independent of this, since the lock is released at xact end.)
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
*/
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
if (freeze_xmax && !HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask))
frz->checkflags |= HEAP_FREEZE_CHECK_XMAX_ABORTED;
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
}
Don't test HEAP_XMAX_INVALID when freezing xmax. We shouldn't ever need to rely on whether HEAP_XMAX_INVALID is set in t_infomask when considering whether or not an xmax should be deemed already frozen, since that status flag is just a hint. The only acceptable representation for an "xmax_already_frozen" raw xmax field is the transaction ID value zero (also known as InvalidTransactionId). Adjust code that superficially appeared to rely on HEAP_XMAX_INVALID to make the rule about xmax_already_frozen clear. Also avoid needlessly rereading the tuple's raw xmax. Oversight in bugfix commit d2599ecf. There is no evidence that this ever led to incorrect behavior, so no backpatch. The worst consequence of this bug was that VACUUM could hypothetically fail to notice and report on certain kinds of corruption, which seems fairly benign. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wzkh3DMCDRPfhZxj9xCq9v3WmzvmbiCpf1dNKUBPadhCbQ@mail.gmail.com
2022-11-23 19:49:39 +01:00
else if (!TransactionIdIsValid(xid))
Don't mark pages all-visible spuriously Dan Wood diagnosed a long-standing problem that pages containing tuples that are locked by multixacts containing live lockers may spuriously end up as candidates for getting their all-visible flag set. This has the long-term effect that multixacts remain unfrozen; this may previously pass undetected, but since commit XYZ it would be reported as "ERROR: found multixact 134100944 from before relminmxid 192042633" because when a later vacuum tries to freeze the page it detects that a multixact that should have gotten frozen, wasn't. Dan proposed a (correct) patch that simply sets a variable to its correct value, after a bogus initialization. But, per discussion, it seems better coding to avoid the bogus initializations altogether, since they could give rise to more bugs later. Therefore this fix rewrites the logic a little bit to avoid depending on the bogus initializations. This bug was part of a family introduced in 9.6 by commit a892234f830e; later, commit 38e9f90a227d fixed most of them, but this one was unnoticed. Authors: Dan Wood, Pavan Deolasee, Álvaro Herrera Reviewed-by: Masahiko Sawada, Pavan Deolasee, Álvaro Herrera Discussion: https://postgr.es/m/84EBAC55-F06D-4FBE-A3F3-8BDA093CE3E3@amazon.com
2018-05-04 20:24:44 +02:00
{
Don't test HEAP_XMAX_INVALID when freezing xmax. We shouldn't ever need to rely on whether HEAP_XMAX_INVALID is set in t_infomask when considering whether or not an xmax should be deemed already frozen, since that status flag is just a hint. The only acceptable representation for an "xmax_already_frozen" raw xmax field is the transaction ID value zero (also known as InvalidTransactionId). Adjust code that superficially appeared to rely on HEAP_XMAX_INVALID to make the rule about xmax_already_frozen clear. Also avoid needlessly rereading the tuple's raw xmax. Oversight in bugfix commit d2599ecf. There is no evidence that this ever led to incorrect behavior, so no backpatch. The worst consequence of this bug was that VACUUM could hypothetically fail to notice and report on certain kinds of corruption, which seems fairly benign. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wzkh3DMCDRPfhZxj9xCq9v3WmzvmbiCpf1dNKUBPadhCbQ@mail.gmail.com
2022-11-23 19:49:39 +01:00
/* Raw xmax is InvalidTransactionId XID */
Assert((tuple->t_infomask & HEAP_XMAX_IS_MULTI) == 0);
Don't mark pages all-visible spuriously Dan Wood diagnosed a long-standing problem that pages containing tuples that are locked by multixacts containing live lockers may spuriously end up as candidates for getting their all-visible flag set. This has the long-term effect that multixacts remain unfrozen; this may previously pass undetected, but since commit XYZ it would be reported as "ERROR: found multixact 134100944 from before relminmxid 192042633" because when a later vacuum tries to freeze the page it detects that a multixact that should have gotten frozen, wasn't. Dan proposed a (correct) patch that simply sets a variable to its correct value, after a bogus initialization. But, per discussion, it seems better coding to avoid the bogus initializations altogether, since they could give rise to more bugs later. Therefore this fix rewrites the logic a little bit to avoid depending on the bogus initializations. This bug was part of a family introduced in 9.6 by commit a892234f830e; later, commit 38e9f90a227d fixed most of them, but this one was unnoticed. Authors: Dan Wood, Pavan Deolasee, Álvaro Herrera Reviewed-by: Masahiko Sawada, Pavan Deolasee, Álvaro Herrera Discussion: https://postgr.es/m/84EBAC55-F06D-4FBE-A3F3-8BDA093CE3E3@amazon.com
2018-05-04 20:24:44 +02:00
xmax_already_frozen = true;
}
else
ereport(ERROR,
(errcode(ERRCODE_DATA_CORRUPTED),
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
errmsg_internal("found raw xmax %u (infomask 0x%04x) not invalid and not multi",
Don't mark pages all-visible spuriously Dan Wood diagnosed a long-standing problem that pages containing tuples that are locked by multixacts containing live lockers may spuriously end up as candidates for getting their all-visible flag set. This has the long-term effect that multixacts remain unfrozen; this may previously pass undetected, but since commit XYZ it would be reported as "ERROR: found multixact 134100944 from before relminmxid 192042633" because when a later vacuum tries to freeze the page it detects that a multixact that should have gotten frozen, wasn't. Dan proposed a (correct) patch that simply sets a variable to its correct value, after a bogus initialization. But, per discussion, it seems better coding to avoid the bogus initializations altogether, since they could give rise to more bugs later. Therefore this fix rewrites the logic a little bit to avoid depending on the bogus initializations. This bug was part of a family introduced in 9.6 by commit a892234f830e; later, commit 38e9f90a227d fixed most of them, but this one was unnoticed. Authors: Dan Wood, Pavan Deolasee, Álvaro Herrera Reviewed-by: Masahiko Sawada, Pavan Deolasee, Álvaro Herrera Discussion: https://postgr.es/m/84EBAC55-F06D-4FBE-A3F3-8BDA093CE3E3@amazon.com
2018-05-04 20:24:44 +02:00
xid, tuple->t_infomask)));
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
if (freeze_xmin)
{
Assert(!xmin_already_frozen);
frz->t_infomask |= HEAP_XMIN_FROZEN;
}
if (replace_xvac)
{
/*
* If a MOVED_OFF tuple is not dead, the xvac transaction must have
* failed; whereas a non-dead MOVED_IN tuple must mean the xvac
* transaction succeeded.
*/
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
Assert(pagefrz->freeze_required);
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
if (tuple->t_infomask & HEAP_MOVED_OFF)
frz->frzflags |= XLH_INVALID_XVAC;
else
frz->frzflags |= XLH_FREEZE_XVAC;
}
if (replace_xmax)
{
Assert(!xmax_already_frozen && !freeze_xmax);
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
Assert(pagefrz->freeze_required);
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
/* Already set replace_xmax flags in freeze plan earlier */
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
}
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
if (freeze_xmax)
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
{
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
Assert(!xmax_already_frozen && !replace_xmax);
Don't mark pages all-visible spuriously Dan Wood diagnosed a long-standing problem that pages containing tuples that are locked by multixacts containing live lockers may spuriously end up as candidates for getting their all-visible flag set. This has the long-term effect that multixacts remain unfrozen; this may previously pass undetected, but since commit XYZ it would be reported as "ERROR: found multixact 134100944 from before relminmxid 192042633" because when a later vacuum tries to freeze the page it detects that a multixact that should have gotten frozen, wasn't. Dan proposed a (correct) patch that simply sets a variable to its correct value, after a bogus initialization. But, per discussion, it seems better coding to avoid the bogus initializations altogether, since they could give rise to more bugs later. Therefore this fix rewrites the logic a little bit to avoid depending on the bogus initializations. This bug was part of a family introduced in 9.6 by commit a892234f830e; later, commit 38e9f90a227d fixed most of them, but this one was unnoticed. Authors: Dan Wood, Pavan Deolasee, Álvaro Herrera Reviewed-by: Masahiko Sawada, Pavan Deolasee, Álvaro Herrera Discussion: https://postgr.es/m/84EBAC55-F06D-4FBE-A3F3-8BDA093CE3E3@amazon.com
2018-05-04 20:24:44 +02:00
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
frz->xmax = InvalidTransactionId;
pgindent run for 8.3.
2007-11-15 22:14:46 +01:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* The tuple might be marked either XMAX_INVALID or XMAX_COMMITTED +
* LOCKED. Normalize to INVALID just to be sure no one gets confused.
* Also get rid of the HEAP_KEYS_UPDATED bit.
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
*/
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
frz->t_infomask &= ~HEAP_XMAX_BITS;
frz->t_infomask |= HEAP_XMAX_INVALID;
frz->t_infomask2 &= ~HEAP_HOT_UPDATED;
frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
}
/*
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
* Determine if this tuple is already totally frozen, or will become
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* totally frozen (provided caller executes freeze plans for the page)
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
*/
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
*totally_frozen = ((freeze_xmin || xmin_already_frozen) &&
(freeze_xmax || xmax_already_frozen));
pgindent run for 8.3.
2007-11-15 22:14:46 +01:00
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
if (!pagefrz->freeze_required && !(xmin_already_frozen &&
xmax_already_frozen))
{
/*
* So far no previous tuple from the page made freezing mandatory.
* Does this tuple force caller to freeze the entire page?
*/
pagefrz->freeze_required =
heap_tuple_should_freeze(tuple, cutoffs,
&pagefrz->NoFreezePageRelfrozenXid,
&pagefrz->NoFreezePageRelminMxid);
}
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
/* Tell caller if this tuple has a usable freeze plan set in *frz */
return freeze_xmin || replace_xvac || replace_xmax || freeze_xmax;
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
}
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
/*
* heap_execute_freeze_tuple
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
* Execute the prepared freezing of a tuple with caller's freeze plan.
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
*
* Caller is responsible for ensuring that no other backend can access the
* storage underlying this tuple, either by holding an exclusive lock on the
Fix typo in comment. Amit Langote
2015-12-18 18:03:15 +01:00
* buffer containing it (which is what lazy VACUUM does), or by having it be
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
* in private storage (which is what CLUSTER and friends do).
*/
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
static inline void
heap_execute_freeze_tuple(HeapTupleHeader tuple, HeapTupleFreeze *frz)
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
{
HeapTupleHeaderSetXmax(tuple, frz->xmax);
if (frz->frzflags & XLH_FREEZE_XVAC)
HeapTupleHeaderSetXvac(tuple, FrozenTransactionId);
if (frz->frzflags & XLH_INVALID_XVAC)
HeapTupleHeaderSetXvac(tuple, InvalidTransactionId);
tuple->t_infomask = frz->t_infomask;
tuple->t_infomask2 = frz->t_infomask2;
}
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
/*
* heap_freeze_execute_prepared
*
* Executes freezing of one or more heap tuples on a page on behalf of caller.
* Caller passes an array of tuple plans from heap_prepare_freeze_tuple.
* Caller must set 'offset' in each plan for us. Note that we destructively
* sort caller's tuples array in-place, so caller had better be done with it.
*
* WAL-logs the changes so that VACUUM can advance the rel's relfrozenxid
* later on without any risk of unsafe pg_xact lookups, even following a hard
* crash (or when querying from a standby). We represent freezing by setting
* infomask bits in tuple headers, but this shouldn't be thought of as a hint.
* See section on buffer access rules in src/backend/storage/buffer/README.
*/
void
heap_freeze_execute_prepared(Relation rel, Buffer buffer,
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
TransactionId snapshotConflictHorizon,
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
HeapTupleFreeze *tuples, int ntuples)
{
Page page = BufferGetPage(buffer);
Assert(ntuples > 0);
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
/*
* Perform xmin/xmax XID status sanity checks before critical section.
*
* heap_prepare_freeze_tuple doesn't perform these checks directly because
* pg_xact lookups are relatively expensive. They shouldn't be repeated
* by successive VACUUMs that each decide against freezing the same page.
*/
for (int i = 0; i < ntuples; i++)
{
HeapTupleFreeze *frz = tuples + i;
ItemId itemid = PageGetItemId(page, frz->offset);
HeapTupleHeader htup;
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
htup = (HeapTupleHeader) PageGetItem(page, itemid);
/* Deliberately avoid relying on tuple hint bits here */
if (frz->checkflags & HEAP_FREEZE_CHECK_XMIN_COMMITTED)
{
TransactionId xmin = HeapTupleHeaderGetRawXmin(htup);
Assert(!HeapTupleHeaderXminFrozen(htup));
if (unlikely(!TransactionIdDidCommit(xmin)))
ereport(ERROR,
(errcode(ERRCODE_DATA_CORRUPTED),
errmsg_internal("uncommitted xmin %u needs to be frozen",
xmin)));
}
Check that xmax didn't commit in freeze check. We cannot rely on TransactionIdDidAbort here, since in general it may report transactions that were in-progress at the time of an earlier hard crash as not aborted, effectively behaving as if they were still in progress even after crash recovery completes. Go back to defensively verifying that xmax didn't commit instead. Oversight in commit 79d4bf4e. Author: Peter Geoghegan <pg@bowt.ie> Reported-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/20230104035636.hy5djyr2as4gbc4q@awork3.anarazel.de
2023-01-04 06:48:27 +01:00
/*
* TransactionIdDidAbort won't work reliably in the presence of XIDs
* left behind by transactions that were in progress during a crash,
* so we can only check that xmax didn't commit
*/
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
if (frz->checkflags & HEAP_FREEZE_CHECK_XMAX_ABORTED)
{
TransactionId xmax = HeapTupleHeaderGetRawXmax(htup);
Assert(TransactionIdIsNormal(xmax));
Check that xmax didn't commit in freeze check. We cannot rely on TransactionIdDidAbort here, since in general it may report transactions that were in-progress at the time of an earlier hard crash as not aborted, effectively behaving as if they were still in progress even after crash recovery completes. Go back to defensively verifying that xmax didn't commit instead. Oversight in commit 79d4bf4e. Author: Peter Geoghegan <pg@bowt.ie> Reported-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/20230104035636.hy5djyr2as4gbc4q@awork3.anarazel.de
2023-01-04 06:48:27 +01:00
if (unlikely(TransactionIdDidCommit(xmax)))
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
ereport(ERROR,
(errcode(ERRCODE_DATA_CORRUPTED),
Check that xmax didn't commit in freeze check. We cannot rely on TransactionIdDidAbort here, since in general it may report transactions that were in-progress at the time of an earlier hard crash as not aborted, effectively behaving as if they were still in progress even after crash recovery completes. Go back to defensively verifying that xmax didn't commit instead. Oversight in commit 79d4bf4e. Author: Peter Geoghegan <pg@bowt.ie> Reported-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/20230104035636.hy5djyr2as4gbc4q@awork3.anarazel.de
2023-01-04 06:48:27 +01:00
errmsg_internal("cannot freeze committed xmax %u",
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
xmax)));
}
}
START_CRIT_SECTION();
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
for (int i = 0; i < ntuples; i++)
{
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
HeapTupleFreeze *frz = tuples + i;
ItemId itemid = PageGetItemId(page, frz->offset);
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
HeapTupleHeader htup;
htup = (HeapTupleHeader) PageGetItem(page, itemid);
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
heap_execute_freeze_tuple(htup, frz);
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
}
Delay commit status checks until freezing executes. pg_xact lookups are relatively expensive. Move the xmin/xmax commit status checks from the point that freeze plans are prepared to the point that they're actually executed. Otherwise we'll repeat many commit status checks whenever multiple successive VACUUM operations scan the same pages and decide against freezing each time, which is a waste of cycles. Oversight in commit 1de58df4, which added page-level freezing. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkZpe4K6qMfEt8H4qYJCKc2R7TPvKsBva7jc9w7iGXQSw@mail.gmail.com
2023-01-03 20:22:36 +01:00
MarkBufferDirty(buffer);
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
/* Now WAL-log freezing if necessary */
if (RelationNeedsWAL(rel))
{
xl_heap_freeze_plan plans[MaxHeapTuplesPerPage];
OffsetNumber offsets[MaxHeapTuplesPerPage];
int nplans;
xl_heap_freeze_page xlrec;
XLogRecPtr recptr;
/* Prepare deduplicated representation for use in WAL record */
Rename and relocate freeze plan dedup routines. Rename the heapam.c freeze plan deduplication routines added by commit 9e540599 to names that follow conventions for functions in heapam.c. Also relocate the functions so that they're next to their caller, which runs during original execution, when FREEZE_PAGE WAL records are built. The routines were initially placed next to (and followed the naming conventions of) conceptually related REDO routine code, but that scheme turned out to be kind of jarring when considered in a wider context. Author: Peter Geoghegan <pg@bowt.ie> Reported-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/20230109214308.icz26oqvt3k2274c@awork3.anarazel.de
2023-01-12 02:30:42 +01:00
nplans = heap_log_freeze_plan(tuples, ntuples, plans, offsets);
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
xlrec.snapshotConflictHorizon = snapshotConflictHorizon;
Add info in WAL records in preparation for logical slot conflict handling This commit only implements one prerequisite part for allowing logical decoding. The commit message contains an explanation of the overall design, which later commits will refer back to. Overall design: 1. We want to enable logical decoding on standbys, but replay of WAL from the primary might remove data that is needed by logical decoding, causing error(s) on the standby. To prevent those errors, a new replication conflict scenario needs to be addressed (as much as hot standby does). 2. Our chosen strategy for dealing with this type of replication slot is to invalidate logical slots for which needed data has been removed. 3. To do this we need the latestRemovedXid for each change, just as we do for physical replication conflicts, but we also need to know whether any particular change was to data that logical replication might access. That way, during WAL replay, we know when there is a risk of conflict and, if so, if there is a conflict. 4. We can't rely on the standby's relcache entries for this purpose in any way, because the startup process can't access catalog contents. 5. Therefore every WAL record that potentially removes data from the index or heap must carry a flag indicating whether or not it is one that might be accessed during logical decoding. Why do we need this for logical decoding on standby? First, let's forget about logical decoding on standby and recall that on a primary database, any catalog rows that may be needed by a logical decoding replication slot are not removed. This is done thanks to the catalog_xmin associated with the logical replication slot. But, with logical decoding on standby, in the following cases: - hot_standby_feedback is off - hot_standby_feedback is on but there is no a physical slot between the primary and the standby. Then, hot_standby_feedback will work, but only while the connection is alive (for example a node restart would break it) Then, the primary may delete system catalog rows that could be needed by the logical decoding on the standby (as it does not know about the catalog_xmin on the standby). So, it’s mandatory to identify those rows and invalidate the slots that may need them if any. Identifying those rows is the purpose of this commit. Implementation: When a WAL replay on standby indicates that a catalog table tuple is to be deleted by an xid that is greater than a logical slot's catalog_xmin, then that means the slot's catalog_xmin conflicts with the xid, and we need to handle the conflict. While subsequent commits will do the actual conflict handling, this commit adds a new field isCatalogRel in such WAL records (and a new bit set in the xl_heap_visible flags field), that is true for catalog tables, so as to arrange for conflict handling. The affected WAL records are the ones that already contain the snapshotConflictHorizon field, namely: - gistxlogDelete - gistxlogPageReuse - xl_hash_vacuum_one_page - xl_heap_prune - xl_heap_freeze_page - xl_heap_visible - xl_btree_reuse_page - xl_btree_delete - spgxlogVacuumRedirect Due to this new field being added, xl_hash_vacuum_one_page and gistxlogDelete do now contain the offsets to be deleted as a FLEXIBLE_ARRAY_MEMBER. This is needed to ensure correct alignment. It's not needed on the others struct where isCatalogRel has been added. This commit just introduces the WAL format changes mentioned above. Handling the actual conflicts will follow in future commits. Bumps XLOG_PAGE_MAGIC as the several WAL records are changed. Author: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Author: Andres Freund <andres@anarazel.de> (in an older version) Author: Amit Khandekar <amitdkhan.pg@gmail.com> (in an older version) Reviewed-by: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Reviewed-by: Andres Freund <andres@anarazel.de> Reviewed-by: Robert Haas <robertmhaas@gmail.com> Reviewed-by: Fabrízio de Royes Mello <fabriziomello@gmail.com> Reviewed-by: Melanie Plageman <melanieplageman@gmail.com>
2023-04-02 21:32:19 +02:00
xlrec.isCatalogRel = RelationIsAccessibleInLogicalDecoding(rel);
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
xlrec.nplans = nplans;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHeapFreezePage);
/*
* The freeze plan array and offset array are not actually in the
* buffer, but pretend that they are. When XLogInsert stores the
* whole buffer, the arrays need not be stored too.
*/
XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
XLogRegisterBufData(0, (char *) plans,
nplans * sizeof(xl_heap_freeze_plan));
XLogRegisterBufData(0, (char *) offsets,
ntuples * sizeof(OffsetNumber));
recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_FREEZE_PAGE);
PageSetLSN(page, recptr);
}
END_CRIT_SECTION();
}
Rename and relocate freeze plan dedup routines. Rename the heapam.c freeze plan deduplication routines added by commit 9e540599 to names that follow conventions for functions in heapam.c. Also relocate the functions so that they're next to their caller, which runs during original execution, when FREEZE_PAGE WAL records are built. The routines were initially placed next to (and followed the naming conventions of) conceptually related REDO routine code, but that scheme turned out to be kind of jarring when considered in a wider context. Author: Peter Geoghegan <pg@bowt.ie> Reported-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/20230109214308.icz26oqvt3k2274c@awork3.anarazel.de
2023-01-12 02:30:42 +01:00
/*
* Comparator used to deduplicate XLOG_HEAP2_FREEZE_PAGE freeze plans
*/
static int
heap_log_freeze_cmp(const void *arg1, const void *arg2)
{
HeapTupleFreeze *frz1 = (HeapTupleFreeze *) arg1;
HeapTupleFreeze *frz2 = (HeapTupleFreeze *) arg2;
if (frz1->xmax < frz2->xmax)
return -1;
else if (frz1->xmax > frz2->xmax)
return 1;
if (frz1->t_infomask2 < frz2->t_infomask2)
return -1;
else if (frz1->t_infomask2 > frz2->t_infomask2)
return 1;
if (frz1->t_infomask < frz2->t_infomask)
return -1;
else if (frz1->t_infomask > frz2->t_infomask)
return 1;
if (frz1->frzflags < frz2->frzflags)
return -1;
else if (frz1->frzflags > frz2->frzflags)
return 1;
/*
* heap_log_freeze_eq would consider these tuple-wise plans to be equal.
* (So the tuples will share a single canonical freeze plan.)
*
* We tiebreak on page offset number to keep each freeze plan's page
* offset number array individually sorted. (Unnecessary, but be tidy.)
*/
if (frz1->offset < frz2->offset)
return -1;
else if (frz1->offset > frz2->offset)
return 1;
Assert(false);
return 0;
}
/*
* Compare fields that describe actions required to freeze tuple with caller's
* open plan. If everything matches then the frz tuple plan is equivalent to
* caller's plan.
*/
static inline bool
heap_log_freeze_eq(xl_heap_freeze_plan *plan, HeapTupleFreeze *frz)
{
if (plan->xmax == frz->xmax &&
plan->t_infomask2 == frz->t_infomask2 &&
plan->t_infomask == frz->t_infomask &&
plan->frzflags == frz->frzflags)
return true;
/* Caller must call heap_log_freeze_new_plan again for frz */
return false;
}
/*
* Start new plan initialized using tuple-level actions. At least one tuple
* will have steps required to freeze described by caller's plan during REDO.
*/
static inline void
heap_log_freeze_new_plan(xl_heap_freeze_plan *plan, HeapTupleFreeze *frz)
{
plan->xmax = frz->xmax;
plan->t_infomask2 = frz->t_infomask2;
plan->t_infomask = frz->t_infomask;
plan->frzflags = frz->frzflags;
plan->ntuples = 1; /* for now */
}
/*
* Deduplicate tuple-based freeze plans so that each distinct set of
* processing steps is only stored once in XLOG_HEAP2_FREEZE_PAGE records.
* Called during original execution of freezing (for logged relations).
*
* Return value is number of plans set in *plans_out for caller. Also writes
* an array of offset numbers into *offsets_out output argument for caller
* (actually there is one array per freeze plan, but that's not of immediate
* concern to our caller).
*/
static int
heap_log_freeze_plan(HeapTupleFreeze *tuples, int ntuples,
xl_heap_freeze_plan *plans_out,
OffsetNumber *offsets_out)
{
int nplans = 0;
/* Sort tuple-based freeze plans in the order required to deduplicate */
qsort(tuples, ntuples, sizeof(HeapTupleFreeze), heap_log_freeze_cmp);
for (int i = 0; i < ntuples; i++)
{
HeapTupleFreeze *frz = tuples + i;
if (i == 0)
{
/* New canonical freeze plan starting with first tup */
heap_log_freeze_new_plan(plans_out, frz);
nplans++;
}
else if (heap_log_freeze_eq(plans_out, frz))
{
/* tup matches open canonical plan -- include tup in it */
Assert(offsets_out[i - 1] < frz->offset);
plans_out->ntuples++;
}
else
{
/* Tup doesn't match current plan -- done with it now */
plans_out++;
/* New canonical freeze plan starting with this tup */
heap_log_freeze_new_plan(plans_out, frz);
nplans++;
}
/*
* Save page offset number in dedicated buffer in passing.
*
* REDO routine relies on the record's offset numbers array grouping
* offset numbers by freeze plan. The sort order within each grouping
* is ascending offset number order, just to keep things tidy.
*/
offsets_out[i] = frz->offset;
}
Assert(nplans > 0 && nplans <= ntuples);
return nplans;
}
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
/*
* heap_freeze_tuple
* Freeze tuple in place, without WAL logging.
*
* Useful for callers like CLUSTER that perform their own WAL logging.
*/
bool
Perform a lot more sanity checks when freezing tuples. The previous commit has shown that the sanity checks around freezing aren't strong enough. Strengthening them seems especially important because the existance of the bug has caused corruption that we don't want to make even worse during future vacuum cycles. The errors are emitted with ereport rather than elog, despite being "should never happen" messages, so a proper error code is emitted. To avoid superflous translations, mark messages as internal. Author: Andres Freund and Alvaro Herrera Reviewed-By: Alvaro Herrera, Michael Paquier Discussion: https://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de Backpatch: 9.3-
2017-11-14 03:45:47 +01:00
heap_freeze_tuple(HeapTupleHeader tuple,
TransactionId relfrozenxid, TransactionId relminmxid,
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
TransactionId FreezeLimit, TransactionId MultiXactCutoff)
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
{
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
HeapTupleFreeze frz;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
bool do_freeze;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
bool totally_frozen;
struct VacuumCutoffs cutoffs;
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
HeapPageFreeze pagefrz;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
cutoffs.relfrozenxid = relfrozenxid;
cutoffs.relminmxid = relminmxid;
cutoffs.OldestXmin = FreezeLimit;
cutoffs.OldestMxact = MultiXactCutoff;
cutoffs.FreezeLimit = FreezeLimit;
cutoffs.MultiXactCutoff = MultiXactCutoff;
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
pagefrz.freeze_required = true;
pagefrz.FreezePageRelfrozenXid = FreezeLimit;
pagefrz.FreezePageRelminMxid = MultiXactCutoff;
pagefrz.NoFreezePageRelfrozenXid = FreezeLimit;
pagefrz.NoFreezePageRelminMxid = MultiXactCutoff;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
do_freeze = heap_prepare_freeze_tuple(tuple, &cutoffs,
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
&pagefrz, &frz, &totally_frozen);
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
/*
* Note that because this is not a WAL-logged operation, we don't need to
* fill in the offset in the freeze record.
*/
if (do_freeze)
heap_execute_freeze_tuple(tuple, &frz);
return do_freeze;
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* For a given MultiXactId, return the hint bits that should be set in the
* tuple's infomask.
*
* Normally this should be called for a multixact that was just created, and
* so is on our local cache, so the GetMembers call is fast.
*/
static void
GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
uint16 *new_infomask2)
{
int nmembers;
MultiXactMember *members;
int i;
uint16 bits = HEAP_XMAX_IS_MULTI;
uint16 bits2 = 0;
bool has_update = false;
Restrict infomask bits to set on multixacts We must only set the bit(s) for the strongest lock held in the tuple; otherwise, a multixact containing members with exclusive lock and key-share lock will behave as though only a share lock is held. This bug was introduced in commit 0ac5ad5134, somewhere along development, when we allowed a singleton FOR SHARE lock to be implemented without a MultiXact by using a multi-bit pattern. I overlooked that GetMultiXactIdHintBits() needed to be tweaked as well. Previously, we could have the bits for FOR KEY SHARE and FOR UPDATE simultaneously set and it wouldn't cause a problem. Per report from digoal@126.com
2013-01-31 23:12:35 +01:00
LockTupleMode strongest = LockTupleKeyShare;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* We only use this in multis we just created, so they cannot be values
* pre-pg_upgrade.
*/
Avoid uselessly looking up old LOCK_ONLY multixacts Commit 0ac5ad5134f2 removed an optimization in multixact.c that skipped fetching members of MultiXactId that were older than our OldestVisibleMXactId value. The reason this was removed is that it is possible for multixacts that contain updates to be older than that value. However, if the caller is certain that the multi does not contain an update (because the infomask bits say so), it can pass this info down to GetMultiXactIdMembers, enabling it to use the old optimization. Pointed out by Andres Freund in 20131121200517.GM7240@alap2.anarazel.de
2014-07-29 21:41:06 +02:00
nmembers = GetMultiXactIdMembers(multi, &members, false, false);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
for (i = 0; i < nmembers; i++)
{
Restrict infomask bits to set on multixacts We must only set the bit(s) for the strongest lock held in the tuple; otherwise, a multixact containing members with exclusive lock and key-share lock will behave as though only a share lock is held. This bug was introduced in commit 0ac5ad5134, somewhere along development, when we allowed a singleton FOR SHARE lock to be implemented without a MultiXact by using a multi-bit pattern. I overlooked that GetMultiXactIdHintBits() needed to be tweaked as well. Previously, we could have the bits for FOR KEY SHARE and FOR UPDATE simultaneously set and it wouldn't cause a problem. Per report from digoal@126.com
2013-01-31 23:12:35 +01:00
LockTupleMode mode;
/*
* Remember the strongest lock mode held by any member of the
* multixact.
*/
mode = TUPLOCK_from_mxstatus(members[i].status);
if (mode > strongest)
strongest = mode;
/* See what other bits we need */
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
switch (members[i].status)
{
case MultiXactStatusForKeyShare:
case MultiXactStatusForShare:
case MultiXactStatusForNoKeyUpdate:
break;
Restrict infomask bits to set on multixacts We must only set the bit(s) for the strongest lock held in the tuple; otherwise, a multixact containing members with exclusive lock and key-share lock will behave as though only a share lock is held. This bug was introduced in commit 0ac5ad5134, somewhere along development, when we allowed a singleton FOR SHARE lock to be implemented without a MultiXact by using a multi-bit pattern. I overlooked that GetMultiXactIdHintBits() needed to be tweaked as well. Previously, we could have the bits for FOR KEY SHARE and FOR UPDATE simultaneously set and it wouldn't cause a problem. Per report from digoal@126.com
2013-01-31 23:12:35 +01:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
case MultiXactStatusForUpdate:
bits2 |= HEAP_KEYS_UPDATED;
break;
Restrict infomask bits to set on multixacts We must only set the bit(s) for the strongest lock held in the tuple; otherwise, a multixact containing members with exclusive lock and key-share lock will behave as though only a share lock is held. This bug was introduced in commit 0ac5ad5134, somewhere along development, when we allowed a singleton FOR SHARE lock to be implemented without a MultiXact by using a multi-bit pattern. I overlooked that GetMultiXactIdHintBits() needed to be tweaked as well. Previously, we could have the bits for FOR KEY SHARE and FOR UPDATE simultaneously set and it wouldn't cause a problem. Per report from digoal@126.com
2013-01-31 23:12:35 +01:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
case MultiXactStatusNoKeyUpdate:
has_update = true;
break;
Restrict infomask bits to set on multixacts We must only set the bit(s) for the strongest lock held in the tuple; otherwise, a multixact containing members with exclusive lock and key-share lock will behave as though only a share lock is held. This bug was introduced in commit 0ac5ad5134, somewhere along development, when we allowed a singleton FOR SHARE lock to be implemented without a MultiXact by using a multi-bit pattern. I overlooked that GetMultiXactIdHintBits() needed to be tweaked as well. Previously, we could have the bits for FOR KEY SHARE and FOR UPDATE simultaneously set and it wouldn't cause a problem. Per report from digoal@126.com
2013-01-31 23:12:35 +01:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
case MultiXactStatusUpdate:
bits2 |= HEAP_KEYS_UPDATED;
has_update = true;
break;
}
}
Restrict infomask bits to set on multixacts We must only set the bit(s) for the strongest lock held in the tuple; otherwise, a multixact containing members with exclusive lock and key-share lock will behave as though only a share lock is held. This bug was introduced in commit 0ac5ad5134, somewhere along development, when we allowed a singleton FOR SHARE lock to be implemented without a MultiXact by using a multi-bit pattern. I overlooked that GetMultiXactIdHintBits() needed to be tweaked as well. Previously, we could have the bits for FOR KEY SHARE and FOR UPDATE simultaneously set and it wouldn't cause a problem. Per report from digoal@126.com
2013-01-31 23:12:35 +01:00
if (strongest == LockTupleExclusive ||
strongest == LockTupleNoKeyExclusive)
bits |= HEAP_XMAX_EXCL_LOCK;
else if (strongest == LockTupleShare)
bits |= HEAP_XMAX_SHR_LOCK;
else if (strongest == LockTupleKeyShare)
bits |= HEAP_XMAX_KEYSHR_LOCK;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if (!has_update)
bits |= HEAP_XMAX_LOCK_ONLY;
if (nmembers > 0)
pfree(members);
*new_infomask = bits;
*new_infomask2 = bits2;
}
/*
* MultiXactIdGetUpdateXid
*
* Given a multixact Xmax and corresponding infomask, which does not have the
* HEAP_XMAX_LOCK_ONLY bit set, obtain and return the Xid of the updating
* transaction.
Don't TransactionIdDidAbort in HeapTupleGetUpdateXid It is dangerous to do so, because some code expects to be able to see what's the true Xmax even if it is aborted (particularly while traversing HOT chains). So don't do it, and instead rely on the callers to verify for abortedness, if necessary. Several race conditions and bugs fixed in the process. One isolation test changes the expected output due to these. This also reverts commit c235a6a589b, which is no longer necessary. Backpatch to 9.3, where this function was introduced. Andres Freund
2013-11-29 20:08:06 +01:00
*
* Caller is expected to check the status of the updating transaction, if
* necessary.
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
*/
static TransactionId
MultiXactIdGetUpdateXid(TransactionId xmax, uint16 t_infomask)
{
TransactionId update_xact = InvalidTransactionId;
MultiXactMember *members;
int nmembers;
Assert(!(t_infomask & HEAP_XMAX_LOCK_ONLY));
Assert(t_infomask & HEAP_XMAX_IS_MULTI);
/*
* Since we know the LOCK_ONLY bit is not set, this cannot be a multi from
* pre-pg_upgrade.
*/
Avoid uselessly looking up old LOCK_ONLY multixacts Commit 0ac5ad5134f2 removed an optimization in multixact.c that skipped fetching members of MultiXactId that were older than our OldestVisibleMXactId value. The reason this was removed is that it is possible for multixacts that contain updates to be older than that value. However, if the caller is certain that the multi does not contain an update (because the infomask bits say so), it can pass this info down to GetMultiXactIdMembers, enabling it to use the old optimization. Pointed out by Andres Freund in 20131121200517.GM7240@alap2.anarazel.de
2014-07-29 21:41:06 +02:00
nmembers = GetMultiXactIdMembers(xmax, &members, false, false);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if (nmembers > 0)
{
int i;
for (i = 0; i < nmembers; i++)
{
/* Ignore lockers */
Don't TransactionIdDidAbort in HeapTupleGetUpdateXid It is dangerous to do so, because some code expects to be able to see what's the true Xmax even if it is aborted (particularly while traversing HOT chains). So don't do it, and instead rely on the callers to verify for abortedness, if necessary. Several race conditions and bugs fixed in the process. One isolation test changes the expected output due to these. This also reverts commit c235a6a589b, which is no longer necessary. Backpatch to 9.3, where this function was introduced. Andres Freund
2013-11-29 20:08:06 +01:00
if (!ISUPDATE_from_mxstatus(members[i].status))
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
continue;
Don't TransactionIdDidAbort in HeapTupleGetUpdateXid It is dangerous to do so, because some code expects to be able to see what's the true Xmax even if it is aborted (particularly while traversing HOT chains). So don't do it, and instead rely on the callers to verify for abortedness, if necessary. Several race conditions and bugs fixed in the process. One isolation test changes the expected output due to these. This also reverts commit c235a6a589b, which is no longer necessary. Backpatch to 9.3, where this function was introduced. Andres Freund
2013-11-29 20:08:06 +01:00
/* there can be at most one updater */
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Assert(update_xact == InvalidTransactionId);
update_xact = members[i].xid;
#ifndef USE_ASSERT_CHECKING
pgindent run for release 9.3 This is the first run of the Perl-based pgindent script. Also update pgindent instructions.
2013-05-29 22:58:43 +02:00
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* in an assert-enabled build, walk the whole array to ensure
* there's no other updater.
*/
break;
#endif
}
pfree(members);
}
return update_xact;
}
/*
* HeapTupleGetUpdateXid
* As above, but use a HeapTupleHeader
*
* See also HeapTupleHeaderGetUpdateXid, which can be used without previously
* checking the hint bits.
*/
TransactionId
HeapTupleGetUpdateXid(HeapTupleHeader tuple)
{
return MultiXactIdGetUpdateXid(HeapTupleHeaderGetRawXmax(tuple),
tuple->t_infomask);
}
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
/*
* Does the given multixact conflict with the current transaction grabbing a
* tuple lock of the given strength?
*
* The passed infomask pairs up with the given multixact in the tuple header.
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
*
* If current_is_member is not NULL, it is set to 'true' if the current
* transaction is a member of the given multixact.
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
*/
static bool
DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
LockTupleMode lockmode, bool *current_is_member)
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
{
int nmembers;
MultiXactMember *members;
bool result = false;
Fix thinko in lock mode enum Commit 0e5680f4737a9c6aa94aa9e77543e5de60411322 contained a thinko mixing LOCKMODE with LockTupleMode. This caused misbehavior in the case where a tuple is marked with a multixact with at most a FOR SHARE lock, and another transaction tries to acquire a FOR NO KEY EXCLUSIVE lock; this case should block but doesn't. Include a new isolation tester spec file to explicitely try all the tuple lock combinations; without the fix it shows the problem: starting permutation: s1_begin s1_lcksvpt s1_tuplock2 s2_tuplock3 s1_commit step s1_begin: BEGIN; step s1_lcksvpt: SELECT * FROM multixact_conflict FOR KEY SHARE; SAVEPOINT foo; a 1 step s1_tuplock2: SELECT * FROM multixact_conflict FOR SHARE; a 1 step s2_tuplock3: SELECT * FROM multixact_conflict FOR NO KEY UPDATE; a 1 step s1_commit: COMMIT; With the fixed code, step s2_tuplock3 blocks until session 1 commits, which is the correct behavior. All other cases behave correctly. Backpatch to 9.3, like the commit that introduced the problem.
2015-01-04 19:48:29 +01:00
LOCKMODE wanted = tupleLockExtraInfo[lockmode].hwlock;
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
Fix handling of multixacts predating pg_upgrade After pg_upgrade, it is possible that some tuples' Xmax have multixacts corresponding to the old installation; such multixacts cannot have running members anymore. In many code sites we already know not to read them and clobber them silently, but at least when VACUUM tries to freeze a multixact or determine whether one needs freezing, there's an attempt to resolve it to its member transactions by calling GetMultiXactIdMembers, and if the multixact value is "in the future" with regards to the current valid multixact range, an error like this is raised: ERROR: MultiXactId 123 has not been created yet -- apparent wraparound and vacuuming fails. Per discussion with Andrew Gierth, it is completely bogus to try to resolve multixacts coming from before a pg_upgrade, regardless of where they stand with regards to the current valid multixact range. It's possible to get from under this problem by doing SELECT FOR UPDATE of the problem tuples, but if tables are large, this is slow and tedious, so a more thorough solution is desirable. To fix, we realize that multixacts in xmax created in 9.2 and previous have a specific bit pattern that is never used in 9.3 and later (we already knew this, per comments and infomask tests sprinkled in various places, but we weren't leveraging this knowledge appropriately). Whenever the infomask of the tuple matches that bit pattern, we just ignore the multixact completely as if Xmax wasn't set; or, in the case of tuple freezing, we act as if an unwanted value is set and clobber it without decoding. This guarantees that no errors will be raised, and that the values will be progressively removed until all tables are clean. Most callers of GetMultiXactIdMembers are patched to recognize directly that the value is a removable "empty" multixact and avoid calling GetMultiXactIdMembers altogether. To avoid changing the signature of GetMultiXactIdMembers() in back branches, we keep the "allow_old" boolean flag but rename it to "from_pgupgrade"; if the flag is true, we always return an empty set instead of looking up the multixact. (I suppose we could remove the argument in the master branch, but I chose not to do so in this commit). This was broken all along, but the error-facing message appeared first because of commit 8e9a16ab8f7f and was partially fixed in a25c2b7c4db3. This fix, backpatched all the way back to 9.3, goes approximately in the same direction as a25c2b7c4db3 but should cover all cases. Bug analysis by Andrew Gierth and Álvaro Herrera. A number of public reports match this bug: https://www.postgresql.org/message-id/20140330040029.GY4582@tamriel.snowman.net https://www.postgresql.org/message-id/538F3D70.6080902@publicrelay.com https://www.postgresql.org/message-id/556439CF.7070109@pscs.co.uk https://www.postgresql.org/message-id/SG2PR06MB0760098A111C88E31BD4D96FB3540@SG2PR06MB0760.apcprd06.prod.outlook.com https://www.postgresql.org/message-id/20160615203829.5798.4594@wrigleys.postgresql.org
2016-06-25 00:29:28 +02:00
if (HEAP_LOCKED_UPGRADED(infomask))
return false;
nmembers = GetMultiXactIdMembers(multi, &members, false,
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
HEAP_XMAX_IS_LOCKED_ONLY(infomask));
if (nmembers >= 0)
{
int i;
for (i = 0; i < nmembers; i++)
{
TransactionId memxid;
Fix thinko in lock mode enum Commit 0e5680f4737a9c6aa94aa9e77543e5de60411322 contained a thinko mixing LOCKMODE with LockTupleMode. This caused misbehavior in the case where a tuple is marked with a multixact with at most a FOR SHARE lock, and another transaction tries to acquire a FOR NO KEY EXCLUSIVE lock; this case should block but doesn't. Include a new isolation tester spec file to explicitely try all the tuple lock combinations; without the fix it shows the problem: starting permutation: s1_begin s1_lcksvpt s1_tuplock2 s2_tuplock3 s1_commit step s1_begin: BEGIN; step s1_lcksvpt: SELECT * FROM multixact_conflict FOR KEY SHARE; SAVEPOINT foo; a 1 step s1_tuplock2: SELECT * FROM multixact_conflict FOR SHARE; a 1 step s2_tuplock3: SELECT * FROM multixact_conflict FOR NO KEY UPDATE; a 1 step s1_commit: COMMIT; With the fixed code, step s2_tuplock3 blocks until session 1 commits, which is the correct behavior. All other cases behave correctly. Backpatch to 9.3, like the commit that introduced the problem.
2015-01-04 19:48:29 +01:00
LOCKMODE memlockmode;
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
if (result && (current_is_member == NULL || *current_is_member))
break;
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
memlockmode = LOCKMODE_from_mxstatus(members[i].status);
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
/* ignore members from current xact (but track their presence) */
Revert "Avoid spurious deadlocks when upgrading a tuple lock" This reverts commits 3da73d6839dc and de87a084c0a5. This code has some tricky corner cases that I'm not sure are correct and not properly tested anyway, so I'm reverting the whole thing for next week's releases (reintroducing the deadlock bug that we set to fix). I'll try again afterwards. Discussion: https://postgr.es/m/E1hbXKQ-0003g1-0C@gemulon.postgresql.org
2019-06-17 04:24:21 +02:00
memxid = members[i].xid;
if (TransactionIdIsCurrentTransactionId(memxid))
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
{
if (current_is_member != NULL)
*current_is_member = true;
continue;
}
else if (result)
continue;
/* ignore members that don't conflict with the lock we want */
if (!DoLockModesConflict(memlockmode, wanted))
Revert "Avoid spurious deadlocks when upgrading a tuple lock" This reverts commits 3da73d6839dc and de87a084c0a5. This code has some tricky corner cases that I'm not sure are correct and not properly tested anyway, so I'm reverting the whole thing for next week's releases (reintroducing the deadlock bug that we set to fix). I'll try again afterwards. Discussion: https://postgr.es/m/E1hbXKQ-0003g1-0C@gemulon.postgresql.org
2019-06-17 04:24:21 +02:00
continue;
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
if (ISUPDATE_from_mxstatus(members[i].status))
{
/* ignore aborted updaters */
if (TransactionIdDidAbort(memxid))
continue;
}
else
{
/* ignore lockers-only that are no longer in progress */
if (!TransactionIdIsInProgress(memxid))
continue;
}
/*
* Whatever remains are either live lockers that conflict with our
* wanted lock, and updaters that are not aborted. Those conflict
Avoid spurious deadlocks when upgrading a tuple lock This puts back reverted commit de87a084c0a5, with some bug fixes. When two (or more) transactions are waiting for transaction T1 to release a tuple-level lock, and transaction T1 upgrades its lock to a higher level, a spurious deadlock can be reported among the waiting transactions when T1 finishes. The simplest example case seems to be: T1: select id from job where name = 'a' for key share; Y: select id from job where name = 'a' for update; -- starts waiting for T1 Z: select id from job where name = 'a' for key share; T1: update job set name = 'b' where id = 1; Z: update job set name = 'c' where id = 1; -- starts waiting for T1 T1: rollback; At this point, transaction Y is rolled back on account of a deadlock: Y holds the heavyweight tuple lock and is waiting for the Xmax to be released, while Z holds part of the multixact and tries to acquire the heavyweight lock (per protocol) and goes to sleep; once T1 releases its part of the multixact, Z is awakened only to be put back to sleep on the heavyweight lock that Y is holding while sleeping. Kaboom. This can be avoided by having Z skip the heavyweight lock acquisition. As far as I can see, the biggest downside is that if there are multiple Z transactions, the order in which they resume after T1 finishes is not guaranteed. Backpatch to 9.6. The patch applies cleanly on 9.5, but the new tests don't work there (because isolationtester is not smart enough), so I'm not going to risk it. Author: Oleksii Kliukin Discussion: https://postgr.es/m/B9C9D7CD-EB94-4635-91B6-E558ACEC0EC3@hintbits.com Discussion: https://postgr.es/m/2815.1560521451@sss.pgh.pa.us
2019-06-19 00:23:16 +02:00
* with what we want. Set up to return true, but keep going to
* look for the current transaction among the multixact members,
* if needed.
Grab heavyweight tuple lock only before sleeping We were trying to acquire the lock even when we were subsequently not sleeping in some other transaction, which opens us up unnecessarily to deadlocks. In particular, this is troublesome if an update tries to lock an updated version of a tuple and finds itself doing EvalPlanQual update chain walking; more than two sessions doing this concurrently will find themselves sleeping on each other because the HW tuple lock acquisition in heap_lock_tuple called from EvalPlanQualFetch races with the same tuple lock being acquired in heap_update -- one of these sessions sleeps on the other one to finish while holding the tuple lock, and the other one sleeps on the tuple lock. Per trouble report from Andrew Sackville-West in http://www.postgresql.org/message-id/20140731233051.GN17765@andrew-ThinkPad-X230 His scenario can be simplified down to a relatively simple isolationtester spec file which I don't include in this commit; the reason is that the current isolationtester is not able to deal with more than one blocked session concurrently and it blocks instead of raising the expected deadlock. In the future, if we improve isolationtester, it would be good to include the spec file in the isolation schedule. I posted it in http://www.postgresql.org/message-id/20141212205254.GC1768@alvh.no-ip.org Hat tip to Mark Kirkwood, who helped diagnose the trouble.
2014-12-26 17:52:27 +01:00
*/
result = true;
}
pfree(members);
}
return result;
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* Do_MultiXactIdWait
* Actual implementation for the two functions below.
*
Setup error context callback for transaction lock waits With this in place, a session blocking behind another one because of tuple locks will get a context line mentioning the relation name, tuple TID, and operation being done on tuple. For example: LOG: process 11367 still waiting for ShareLock on transaction 717 after 1000.108 ms DETAIL: Process holding the lock: 11366. Wait queue: 11367. CONTEXT: while updating tuple (0,2) in relation "foo" STATEMENT: UPDATE foo SET value = 3; Most usefully, the new line is displayed by log entries due to log_lock_waits, although of course it will be printed by any other log message as well. Author: Christian Kruse, some tweaks by Álvaro Herrera Reviewed-by: Amit Kapila, Andres Freund, Tom Lane, Robert Haas
2014-03-19 19:10:36 +01:00
* 'multi', 'status' and 'infomask' indicate what to sleep on (the status is
* needed to ensure we only sleep on conflicting members, and the infomask is
* used to optimize multixact access in case it's a lock-only multi); 'nowait'
* indicates whether to use conditional lock acquisition, to allow callers to
* fail if lock is unavailable. 'rel', 'ctid' and 'oper' are used to set up
* context information for error messages. 'remaining', if not NULL, receives
* the number of members that are still running, including any (non-aborted)
* subtransactions of our own transaction.
*
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
* We do this by sleeping on each member using XactLockTableWait. Any
* members that belong to the current backend are *not* waited for, however;
* this would not merely be useless but would lead to Assert failure inside
* XactLockTableWait. By the time this returns, it is certain that all
* transactions *of other backends* that were members of the MultiXactId
* that conflict with the requested status are dead (and no new ones can have
* been added, since it is not legal to add members to an existing
* MultiXactId).
*
* But by the time we finish sleeping, someone else may have changed the Xmax
* of the containing tuple, so the caller needs to iterate on us somehow.
*
* Note that in case we return false, the number of remaining members is
* not to be trusted.
*/
static bool
Do_MultiXactIdWait(MultiXactId multi, MultiXactStatus status,
Setup error context callback for transaction lock waits With this in place, a session blocking behind another one because of tuple locks will get a context line mentioning the relation name, tuple TID, and operation being done on tuple. For example: LOG: process 11367 still waiting for ShareLock on transaction 717 after 1000.108 ms DETAIL: Process holding the lock: 11366. Wait queue: 11367. CONTEXT: while updating tuple (0,2) in relation "foo" STATEMENT: UPDATE foo SET value = 3; Most usefully, the new line is displayed by log entries due to log_lock_waits, although of course it will be printed by any other log message as well. Author: Christian Kruse, some tweaks by Álvaro Herrera Reviewed-by: Amit Kapila, Andres Freund, Tom Lane, Robert Haas
2014-03-19 19:10:36 +01:00
uint16 infomask, bool nowait,
Relation rel, ItemPointer ctid, XLTW_Oper oper,
int *remaining)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
bool result = true;
MultiXactMember *members;
int nmembers;
int remain = 0;
Fix handling of multixacts predating pg_upgrade After pg_upgrade, it is possible that some tuples' Xmax have multixacts corresponding to the old installation; such multixacts cannot have running members anymore. In many code sites we already know not to read them and clobber them silently, but at least when VACUUM tries to freeze a multixact or determine whether one needs freezing, there's an attempt to resolve it to its member transactions by calling GetMultiXactIdMembers, and if the multixact value is "in the future" with regards to the current valid multixact range, an error like this is raised: ERROR: MultiXactId 123 has not been created yet -- apparent wraparound and vacuuming fails. Per discussion with Andrew Gierth, it is completely bogus to try to resolve multixacts coming from before a pg_upgrade, regardless of where they stand with regards to the current valid multixact range. It's possible to get from under this problem by doing SELECT FOR UPDATE of the problem tuples, but if tables are large, this is slow and tedious, so a more thorough solution is desirable. To fix, we realize that multixacts in xmax created in 9.2 and previous have a specific bit pattern that is never used in 9.3 and later (we already knew this, per comments and infomask tests sprinkled in various places, but we weren't leveraging this knowledge appropriately). Whenever the infomask of the tuple matches that bit pattern, we just ignore the multixact completely as if Xmax wasn't set; or, in the case of tuple freezing, we act as if an unwanted value is set and clobber it without decoding. This guarantees that no errors will be raised, and that the values will be progressively removed until all tables are clean. Most callers of GetMultiXactIdMembers are patched to recognize directly that the value is a removable "empty" multixact and avoid calling GetMultiXactIdMembers altogether. To avoid changing the signature of GetMultiXactIdMembers() in back branches, we keep the "allow_old" boolean flag but rename it to "from_pgupgrade"; if the flag is true, we always return an empty set instead of looking up the multixact. (I suppose we could remove the argument in the master branch, but I chose not to do so in this commit). This was broken all along, but the error-facing message appeared first because of commit 8e9a16ab8f7f and was partially fixed in a25c2b7c4db3. This fix, backpatched all the way back to 9.3, goes approximately in the same direction as a25c2b7c4db3 but should cover all cases. Bug analysis by Andrew Gierth and Álvaro Herrera. A number of public reports match this bug: https://www.postgresql.org/message-id/20140330040029.GY4582@tamriel.snowman.net https://www.postgresql.org/message-id/538F3D70.6080902@publicrelay.com https://www.postgresql.org/message-id/556439CF.7070109@pscs.co.uk https://www.postgresql.org/message-id/SG2PR06MB0760098A111C88E31BD4D96FB3540@SG2PR06MB0760.apcprd06.prod.outlook.com https://www.postgresql.org/message-id/20160615203829.5798.4594@wrigleys.postgresql.org
2016-06-25 00:29:28 +02:00
/* for pre-pg_upgrade tuples, no need to sleep at all */
nmembers = HEAP_LOCKED_UPGRADED(infomask) ? -1 :
GetMultiXactIdMembers(multi, &members, false,
HEAP_XMAX_IS_LOCKED_ONLY(infomask));
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
if (nmembers >= 0)
{
int i;
for (i = 0; i < nmembers; i++)
{
TransactionId memxid = members[i].xid;
MultiXactStatus memstatus = members[i].status;
if (TransactionIdIsCurrentTransactionId(memxid))
{
remain++;
continue;
}
if (!DoLockModesConflict(LOCKMODE_from_mxstatus(memstatus),
LOCKMODE_from_mxstatus(status)))
{
if (remaining && TransactionIdIsInProgress(memxid))
remain++;
continue;
}
/*
* This member conflicts with our multi, so we have to sleep (or
* return failure, if asked to avoid waiting.)
Setup error context callback for transaction lock waits With this in place, a session blocking behind another one because of tuple locks will get a context line mentioning the relation name, tuple TID, and operation being done on tuple. For example: LOG: process 11367 still waiting for ShareLock on transaction 717 after 1000.108 ms DETAIL: Process holding the lock: 11366. Wait queue: 11367. CONTEXT: while updating tuple (0,2) in relation "foo" STATEMENT: UPDATE foo SET value = 3; Most usefully, the new line is displayed by log entries due to log_lock_waits, although of course it will be printed by any other log message as well. Author: Christian Kruse, some tweaks by Álvaro Herrera Reviewed-by: Amit Kapila, Andres Freund, Tom Lane, Robert Haas
2014-03-19 19:10:36 +01:00
*
* Note that we don't set up an error context callback ourselves,
* but instead we pass the info down to XactLockTableWait. This
* might seem a bit wasteful because the context is set up and
* tore down for each member of the multixact, but in reality it
* should be barely noticeable, and it avoids duplicate code.
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
*/
if (nowait)
{
result = ConditionalXactLockTableWait(memxid);
if (!result)
break;
}
else
Setup error context callback for transaction lock waits With this in place, a session blocking behind another one because of tuple locks will get a context line mentioning the relation name, tuple TID, and operation being done on tuple. For example: LOG: process 11367 still waiting for ShareLock on transaction 717 after 1000.108 ms DETAIL: Process holding the lock: 11366. Wait queue: 11367. CONTEXT: while updating tuple (0,2) in relation "foo" STATEMENT: UPDATE foo SET value = 3; Most usefully, the new line is displayed by log entries due to log_lock_waits, although of course it will be printed by any other log message as well. Author: Christian Kruse, some tweaks by Álvaro Herrera Reviewed-by: Amit Kapila, Andres Freund, Tom Lane, Robert Haas
2014-03-19 19:10:36 +01:00
XactLockTableWait(memxid, rel, ctid, oper);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
pfree(members);
}
if (remaining)
*remaining = remain;
return result;
}
/*
* MultiXactIdWait
* Sleep on a MultiXactId.
*
* By the time we finish sleeping, someone else may have changed the Xmax
* of the containing tuple, so the caller needs to iterate on us somehow.
*
* We return (in *remaining, if not NULL) the number of members that are still
* running, including any (non-aborted) subtransactions of our own transaction.
*/
static void
Setup error context callback for transaction lock waits With this in place, a session blocking behind another one because of tuple locks will get a context line mentioning the relation name, tuple TID, and operation being done on tuple. For example: LOG: process 11367 still waiting for ShareLock on transaction 717 after 1000.108 ms DETAIL: Process holding the lock: 11366. Wait queue: 11367. CONTEXT: while updating tuple (0,2) in relation "foo" STATEMENT: UPDATE foo SET value = 3; Most usefully, the new line is displayed by log entries due to log_lock_waits, although of course it will be printed by any other log message as well. Author: Christian Kruse, some tweaks by Álvaro Herrera Reviewed-by: Amit Kapila, Andres Freund, Tom Lane, Robert Haas
2014-03-19 19:10:36 +01:00
MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
Relation rel, ItemPointer ctid, XLTW_Oper oper,
int *remaining)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
Setup error context callback for transaction lock waits With this in place, a session blocking behind another one because of tuple locks will get a context line mentioning the relation name, tuple TID, and operation being done on tuple. For example: LOG: process 11367 still waiting for ShareLock on transaction 717 after 1000.108 ms DETAIL: Process holding the lock: 11366. Wait queue: 11367. CONTEXT: while updating tuple (0,2) in relation "foo" STATEMENT: UPDATE foo SET value = 3; Most usefully, the new line is displayed by log entries due to log_lock_waits, although of course it will be printed by any other log message as well. Author: Christian Kruse, some tweaks by Álvaro Herrera Reviewed-by: Amit Kapila, Andres Freund, Tom Lane, Robert Haas
2014-03-19 19:10:36 +01:00
(void) Do_MultiXactIdWait(multi, status, infomask, false,
rel, ctid, oper, remaining);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
/*
* ConditionalMultiXactIdWait
* As above, but only lock if we can get the lock without blocking.
*
* By the time we finish sleeping, someone else may have changed the Xmax
* of the containing tuple, so the caller needs to iterate on us somehow.
*
* If the multixact is now all gone, return true. Returns false if some
* transactions might still be running.
*
* We return (in *remaining, if not NULL) the number of members that are still
* running, including any (non-aborted) subtransactions of our own transaction.
*/
static bool
ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
Revert XactLockTableWait context setup in conditional multixact wait There's no point in setting up a context error callback when doing conditional lock acquisition, because we never actually wait and so the user wouldn't be able to see the context message anywhere. In fact, this is more in line with what ConditionalXactLockTableWait is doing. Backpatch to 9.4, where this was added.
2014-08-25 21:33:17 +02:00
uint16 infomask, Relation rel, int *remaining)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
Setup error context callback for transaction lock waits With this in place, a session blocking behind another one because of tuple locks will get a context line mentioning the relation name, tuple TID, and operation being done on tuple. For example: LOG: process 11367 still waiting for ShareLock on transaction 717 after 1000.108 ms DETAIL: Process holding the lock: 11366. Wait queue: 11367. CONTEXT: while updating tuple (0,2) in relation "foo" STATEMENT: UPDATE foo SET value = 3; Most usefully, the new line is displayed by log entries due to log_lock_waits, although of course it will be printed by any other log message as well. Author: Christian Kruse, some tweaks by Álvaro Herrera Reviewed-by: Amit Kapila, Andres Freund, Tom Lane, Robert Haas
2014-03-19 19:10:36 +01:00
return Do_MultiXactIdWait(multi, status, infomask, true,
Revert XactLockTableWait context setup in conditional multixact wait There's no point in setting up a context error callback when doing conditional lock acquisition, because we never actually wait and so the user wouldn't be able to see the context message anywhere. In fact, this is more in line with what ConditionalXactLockTableWait is doing. Backpatch to 9.4, where this was added.
2014-08-25 21:33:17 +02:00
rel, NULL, XLTW_None, remaining);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
Change the format of the VM fork to add a second bit per page. The new bit indicates whether every tuple on the page is already frozen. It is cleared only when the all-visible bit is cleared, and it can be set only when we vacuum a page and find that every tuple on that page is both visible to every transaction and in no need of any future vacuuming. A future commit will use this new bit to optimize away full-table scans that would otherwise be triggered by XID wraparound considerations. A page which is merely all-visible must still be scanned in that case, but a page which is all-frozen need not be. This commit does not attempt that optimization, although that optimization is the goal here. It seems better to get the basic infrastructure in place first. Per discussion, it's very desirable for pg_upgrade to automatically migrate existing VM forks from the old format to the new format. That, too, will be handled in a follow-on patch. Masahiko Sawada, reviewed by Kyotaro Horiguchi, Fujii Masao, Amit Kapila, Simon Riggs, Andres Freund, and others, and substantially revised by me.
2016-03-02 03:49:41 +01:00
/*
* heap_tuple_needs_eventual_freeze
*
* Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
* will eventually require freezing (if tuple isn't removed by pruning first).
Change the format of the VM fork to add a second bit per page. The new bit indicates whether every tuple on the page is already frozen. It is cleared only when the all-visible bit is cleared, and it can be set only when we vacuum a page and find that every tuple on that page is both visible to every transaction and in no need of any future vacuuming. A future commit will use this new bit to optimize away full-table scans that would otherwise be triggered by XID wraparound considerations. A page which is merely all-visible must still be scanned in that case, but a page which is all-frozen need not be. This commit does not attempt that optimization, although that optimization is the goal here. It seems better to get the basic infrastructure in place first. Per discussion, it's very desirable for pg_upgrade to automatically migrate existing VM forks from the old format to the new format. That, too, will be handled in a follow-on patch. Masahiko Sawada, reviewed by Kyotaro Horiguchi, Fujii Masao, Amit Kapila, Simon Riggs, Andres Freund, and others, and substantially revised by me.
2016-03-02 03:49:41 +01:00
*/
bool
heap_tuple_needs_eventual_freeze(HeapTupleHeader tuple)
{
TransactionId xid;
/*
* If xmin is a normal transaction ID, this tuple is definitely not
* frozen.
*/
xid = HeapTupleHeaderGetXmin(tuple);
if (TransactionIdIsNormal(xid))
return true;
/*
* If xmax is a valid xact or multixact, this tuple is also not frozen.
*/
if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
{
MultiXactId multi;
multi = HeapTupleHeaderGetRawXmax(tuple);
if (MultiXactIdIsValid(multi))
return true;
}
else
{
xid = HeapTupleHeaderGetRawXmax(tuple);
if (TransactionIdIsNormal(xid))
return true;
}
if (tuple->t_infomask & HEAP_MOVED)
{
xid = HeapTupleHeaderGetXvac(tuple);
if (TransactionIdIsNormal(xid))
return true;
}
return false;
}
Make VACUUM avoid waiting for a cleanup lock, where possible. In a regular VACUUM, it's OK to skip pages for which a cleanup lock isn't immediately available; the next VACUUM will deal with them. If we're scanning the entire relation to advance relfrozenxid, we might need to wait, but only if there are tuples on the page that actually require freezing. These changes should greatly reduce the incidence of of vacuum processes getting "stuck". Simon Riggs and Robert Haas
2011-11-08 03:39:40 +01:00
/*
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* heap_tuple_should_freeze
Make VACUUM avoid waiting for a cleanup lock, where possible. In a regular VACUUM, it's OK to skip pages for which a cleanup lock isn't immediately available; the next VACUUM will deal with them. If we're scanning the entire relation to advance relfrozenxid, we might need to wait, but only if there are tuples on the page that actually require freezing. These changes should greatly reduce the incidence of of vacuum processes getting "stuck". Simon Riggs and Robert Haas
2011-11-08 03:39:40 +01:00
*
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
* Return value indicates if heap_prepare_freeze_tuple sibling function would
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* (or should) force freezing of the heap page that contains caller's tuple.
* Tuple header XIDs/MXIDs < FreezeLimit/MultiXactCutoff trigger freezing.
* This includes (xmin, xmax, xvac) fields, as well as MultiXact member XIDs.
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
*
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
* The *NoFreezePageRelfrozenXid and *NoFreezePageRelminMxid input/output
* arguments help VACUUM track the oldest extant XID/MXID remaining in rel.
* Our working assumption is that caller won't decide to freeze this tuple.
* It's up to caller to only ratchet back its own top-level trackers after the
* point that it fully commits to not freezing the tuple/page in question.
Make VACUUM avoid waiting for a cleanup lock, where possible. In a regular VACUUM, it's OK to skip pages for which a cleanup lock isn't immediately available; the next VACUUM will deal with them. If we're scanning the entire relation to advance relfrozenxid, we might need to wait, but only if there are tuples on the page that actually require freezing. These changes should greatly reduce the incidence of of vacuum processes getting "stuck". Simon Riggs and Robert Haas
2011-11-08 03:39:40 +01:00
*/
bool
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
heap_tuple_should_freeze(HeapTupleHeader tuple,
const struct VacuumCutoffs *cutoffs,
TransactionId *NoFreezePageRelfrozenXid,
MultiXactId *NoFreezePageRelminMxid)
Make VACUUM avoid waiting for a cleanup lock, where possible. In a regular VACUUM, it's OK to skip pages for which a cleanup lock isn't immediately available; the next VACUUM will deal with them. If we're scanning the entire relation to advance relfrozenxid, we might need to wait, but only if there are tuples on the page that actually require freezing. These changes should greatly reduce the incidence of of vacuum processes getting "stuck". Simon Riggs and Robert Haas
2011-11-08 03:39:40 +01:00
{
TransactionId xid;
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
MultiXactId multi;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
bool freeze = false;
Make VACUUM avoid waiting for a cleanup lock, where possible. In a regular VACUUM, it's OK to skip pages for which a cleanup lock isn't immediately available; the next VACUUM will deal with them. If we're scanning the entire relation to advance relfrozenxid, we might need to wait, but only if there are tuples on the page that actually require freezing. These changes should greatly reduce the incidence of of vacuum processes getting "stuck". Simon Riggs and Robert Haas
2011-11-08 03:39:40 +01:00
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
/* First deal with xmin */
Make VACUUM avoid waiting for a cleanup lock, where possible. In a regular VACUUM, it's OK to skip pages for which a cleanup lock isn't immediately available; the next VACUUM will deal with them. If we're scanning the entire relation to advance relfrozenxid, we might need to wait, but only if there are tuples on the page that actually require freezing. These changes should greatly reduce the incidence of of vacuum processes getting "stuck". Simon Riggs and Robert Haas
2011-11-08 03:39:40 +01:00
xid = HeapTupleHeaderGetXmin(tuple);
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
if (TransactionIdIsNormal(xid))
Make VACUUM avoid waiting for a cleanup lock, where possible. In a regular VACUUM, it's OK to skip pages for which a cleanup lock isn't immediately available; the next VACUUM will deal with them. If we're scanning the entire relation to advance relfrozenxid, we might need to wait, but only if there are tuples on the page that actually require freezing. These changes should greatly reduce the incidence of of vacuum processes getting "stuck". Simon Riggs and Robert Haas
2011-11-08 03:39:40 +01:00
{
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
*NoFreezePageRelfrozenXid = xid;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
freeze = true;
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
}
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
/* Now deal with xmax */
xid = InvalidTransactionId;
multi = InvalidMultiXactId;
if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
multi = HeapTupleHeaderGetRawXmax(tuple);
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
else
xid = HeapTupleHeaderGetRawXmax(tuple);
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
if (TransactionIdIsNormal(xid))
{
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
/* xmax is a non-permanent XID */
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
*NoFreezePageRelfrozenXid = xid;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
freeze = true;
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
}
else if (!MultiXactIdIsValid(multi))
{
/* xmax is a permanent XID or invalid MultiXactId/XID */
}
else if (HEAP_LOCKED_UPGRADED(tuple->t_infomask))
{
/* xmax is a pg_upgrade'd MultiXact, which can't have updater XID */
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
*NoFreezePageRelminMxid = multi;
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
/* heap_prepare_freeze_tuple always freezes pg_upgrade'd xmax */
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
freeze = true;
Make VACUUM avoid waiting for a cleanup lock, where possible. In a regular VACUUM, it's OK to skip pages for which a cleanup lock isn't immediately available; the next VACUUM will deal with them. If we're scanning the entire relation to advance relfrozenxid, we might need to wait, but only if there are tuples on the page that actually require freezing. These changes should greatly reduce the incidence of of vacuum processes getting "stuck". Simon Riggs and Robert Haas
2011-11-08 03:39:40 +01:00
}
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
else
{
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
/* xmax is a MultiXactId that may have an updater XID */
MultiXactMember *members;
int nmembers;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
Assert(MultiXactIdPrecedesOrEquals(cutoffs->relminmxid, multi));
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
*NoFreezePageRelminMxid = multi;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
if (MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff))
freeze = true;
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
/* need to check whether any member of the mxact is old */
nmembers = GetMultiXactIdMembers(multi, &members, false,
HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask));
for (int i = 0; i < nmembers; i++)
{
xid = members[i].xid;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
*NoFreezePageRelfrozenXid = xid;
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
freeze = true;
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
}
if (nmembers > 0)
pfree(members);
Fix a couple of bugs in MultiXactId freezing Both heap_freeze_tuple() and heap_tuple_needs_freeze() neglected to look into a multixact to check the members against cutoff_xid. This means that a very old Xid could survive hidden within a multi, possibly outliving its CLOG storage. In the distant future, this would cause clog lookup failures: ERROR: could not access status of transaction 3883960912 DETAIL: Could not open file "pg_clog/0E78": No such file or directory. This mostly was problematic when the updating transaction aborted, since in that case the row wouldn't get pruned away earlier in vacuum and the multixact could possibly survive for a long time. In many cases, data that is inaccessible for this reason way can be brought back heuristically. As a second bug, heap_freeze_tuple() didn't properly handle multixacts that need to be frozen according to cutoff_multi, but whose updater xid is still alive. Instead of preserving the update Xid, it just set Xmax invalid, which leads to both old and new tuple versions becoming visible. This is pretty rare in practice, but a real threat nonetheless. Existing corrupted rows, unfortunately, cannot be repaired in an automated fashion. Existing physical replicas might have already incorrectly frozen tuples because of different behavior than in master, which might only become apparent in the future once pg_multixact/ is truncated; it is recommended that all clones be rebuilt after upgrading. Following code analysis caused by bug report by J Smith in message CADFUPgc5bmtv-yg9znxV-vcfkb+JPRqs7m2OesQXaM_4Z1JpdQ@mail.gmail.com and privately by F-Secure. Backpatch to 9.3, where freezing of MultiXactIds was introduced. Analysis and patch by Andres Freund, with some tweaks by Álvaro.
2013-11-28 23:17:21 +01:00
}
Make VACUUM avoid waiting for a cleanup lock, where possible. In a regular VACUUM, it's OK to skip pages for which a cleanup lock isn't immediately available; the next VACUUM will deal with them. If we're scanning the entire relation to advance relfrozenxid, we might need to wait, but only if there are tuples on the page that actually require freezing. These changes should greatly reduce the incidence of of vacuum processes getting "stuck". Simon Riggs and Robert Haas
2011-11-08 03:39:40 +01:00
if (tuple->t_infomask & HEAP_MOVED)
{
xid = HeapTupleHeaderGetXvac(tuple);
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
if (TransactionIdIsNormal(xid))
{
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
Add page-level freezing to VACUUM. Teach VACUUM to decide on whether or not to trigger freezing at the level of whole heap pages. Individual XIDs and MXIDs fields from tuple headers now trigger freezing of whole pages, rather than independently triggering freezing of each individual tuple header field. Managing the cost of freezing over time now significantly influences when and how VACUUM freezes. The overall amount of WAL written is the single most important freezing related cost, in general. Freezing each page's tuples together in batch allows VACUUM to take full advantage of the freeze plan WAL deduplication optimization added by commit 9e540599. Also teach VACUUM to trigger page-level freezing whenever it detects that heap pruning generated an FPI. We'll have already written a large amount of WAL just to do that much, so it's very likely a good idea to get freezing out of the way for the page early. This only happens in cases where it will directly lead to marking the page all-frozen in the visibility map. In most cases "freezing a page" removes all XIDs < OldestXmin, and all MXIDs < OldestMxact. It doesn't quite work that way in certain rare cases involving MultiXacts, though. It is convenient to define "freeze the page" in a way that gives FreezeMultiXactId the leeway to put off the work of processing an individual tuple's xmax whenever it happens to be a MultiXactId that would require an expensive second pass to process aggressively (allocating a new multi is especially worth avoiding here). FreezeMultiXactId is eager when processing is cheap (as it usually is), and lazy in the event of an individual multi that happens to require expensive second pass processing. This avoids regressions related to processing of multis that page-level freezing might otherwise cause. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-WzkFok_6EAHuK39GaW4FjEFQsY=3J0AAd6FXk93u-Xq3Fg@mail.gmail.com
2022-12-28 17:50:47 +01:00
if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
*NoFreezePageRelfrozenXid = xid;
/* heap_prepare_freeze_tuple forces xvac freezing */
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
freeze = true;
Set relfrozenxid to oldest extant XID seen by VACUUM. When VACUUM set relfrozenxid before now, it set it to whatever value was used to determine which tuples to freeze -- the FreezeLimit cutoff. This approach was very naive. The relfrozenxid invariant only requires that new relfrozenxid values be <= the oldest extant XID remaining in the table (at the point that the VACUUM operation ends), which in general might be much more recent than FreezeLimit. VACUUM now carefully tracks the oldest remaining XID/MultiXactId as it goes (the oldest remaining values _after_ lazy_scan_prune processing). The final values are set as the table's new relfrozenxid and new relminmxid in pg_class at the end of each VACUUM. The oldest XID might come from a tuple's xmin, xmax, or xvac fields. It might even come from one of the table's remaining MultiXacts. Final relfrozenxid values must still be >= FreezeLimit in an aggressive VACUUM (FreezeLimit still acts as a lower bound on the final value that aggressive VACUUM can set relfrozenxid to). Since standard VACUUMs still make no guarantees about advancing relfrozenxid, they might as well set relfrozenxid to a value from well before FreezeLimit when the opportunity presents itself. In general standard VACUUMs may now set relfrozenxid to any value > the original relfrozenxid and <= OldestXmin. Credit for the general idea of using the oldest extant XID to set pg_class.relfrozenxid at the end of VACUUM goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Robert Haas <robertmhaas@gmail.com> Discussion: https://postgr.es/m/CAH2-WzkymFbz6D_vL+jmqSn_5q1wsFvFrE+37yLgL_Rkfd6Gzg@mail.gmail.com
2022-04-03 18:57:21 +02:00
}
Make VACUUM avoid waiting for a cleanup lock, where possible. In a regular VACUUM, it's OK to skip pages for which a cleanup lock isn't immediately available; the next VACUUM will deal with them. If we're scanning the entire relation to advance relfrozenxid, we might need to wait, but only if there are tuples on the page that actually require freezing. These changes should greatly reduce the incidence of of vacuum processes getting "stuck". Simon Riggs and Robert Haas
2011-11-08 03:39:40 +01:00
}
Refactor how VACUUM passes around its XID cutoffs. Use a dedicated struct for the XID/MXID cutoffs used by VACUUM, such as FreezeLimit and OldestXmin. This state is initialized in vacuum.c, and then passed around by code from vacuumlazy.c to heapam.c freezing related routines. The new convention is that everybody works off of the same cutoff state, which is passed around via pointers to const. Also simplify some of the logic for dealing with frozen xmin in heap_prepare_freeze_tuple: add dedicated "xmin_already_frozen" state to clearly distinguish xmin XIDs that we're going to freeze from those that were already frozen from before. That way the routine's xmin handling code is symmetrical with the existing xmax handling code. This is preparation for an upcoming commit that will add page level freezing. Also refactor the control flow within FreezeMultiXactId(), while adding stricter sanity checks. We now test OldestXmin directly, instead of using FreezeLimit as an inexact proxy for OldestXmin. This is further preparation for the page level freezing work, which will make the function's caller cede control of page level freezing to the function where appropriate (where heap_prepare_freeze_tuple sees a tuple that happens to contain a MultiXactId in its xmax). Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Jeff Davis <pgsql@j-davis.com> Discussion: https://postgr.es/m/CAH2-WznS9TxXmz2_=SY+SyJyDFbiOftKofM9=aDo68BbXNBUMA@mail.gmail.com
2022-12-22 18:37:59 +01:00
return freeze;
Make VACUUM avoid waiting for a cleanup lock, where possible. In a regular VACUUM, it's OK to skip pages for which a cleanup lock isn't immediately available; the next VACUUM will deal with them. If we're scanning the entire relation to advance relfrozenxid, we might need to wait, but only if there are tuples on the page that actually require freezing. These changes should greatly reduce the incidence of of vacuum processes getting "stuck". Simon Riggs and Robert Haas
2011-11-08 03:39:40 +01:00
}
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
Allow read only connections during recovery, known as Hot Standby. Enabled by recovery_connections = on (default) and forcing archive recovery using a recovery.conf. Recovery processing now emulates the original transactions as they are replayed, providing full locking and MVCC behaviour for read only queries. Recovery must enter consistent state before connections are allowed, so there is a delay, typically short, before connections succeed. Replay of recovering transactions can conflict and in some cases deadlock with queries during recovery; these result in query cancellation after max_standby_delay seconds have expired. Infrastructure changes have minor effects on normal running, though introduce four new types of WAL record. New test mode "make standbycheck" allows regression tests of static command behaviour on a standby server while in recovery. Typical and extreme dynamic behaviours have been checked via code inspection and manual testing. Few port specific behaviours have been utilised, though primary testing has been on Linux only so far. This commit is the basic patch. Additional changes will follow in this release to enhance some aspects of behaviour, notably improved handling of conflicts, deadlock detection and query cancellation. Changes to VACUUM FULL are also required. Simon Riggs, with significant and lengthy review by Heikki Linnakangas, including streamlined redesign of snapshot creation and two-phase commit. Important contributions from Florian Pflug, Mark Kirkwood, Merlin Moncure, Greg Stark, Gianni Ciolli, Gabriele Bartolini, Hannu Krosing, Robert Haas, Tatsuo Ishii, Hiroyuki Yamada plus support and feedback from many other community members.
2009-12-19 02:32:45 +01:00
/*
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
* Maintain snapshotConflictHorizon for caller by ratcheting forward its value
* using any committed XIDs contained in 'tuple', an obsolescent heap tuple
* that caller is in the process of physically removing, e.g. via HOT pruning
* or index deletion.
*
* Caller must initialize its value to InvalidTransactionId, which is
* generally interpreted as "definitely no need for a recovery conflict".
* Final value must reflect all heap tuples that caller will physically remove
* (or remove TID references to) via its ongoing pruning/deletion operation.
* ResolveRecoveryConflictWithSnapshot() is passed the final value (taken from
* caller's WAL record) by REDO routine when it replays caller's operation.
Allow read only connections during recovery, known as Hot Standby. Enabled by recovery_connections = on (default) and forcing archive recovery using a recovery.conf. Recovery processing now emulates the original transactions as they are replayed, providing full locking and MVCC behaviour for read only queries. Recovery must enter consistent state before connections are allowed, so there is a delay, typically short, before connections succeed. Replay of recovering transactions can conflict and in some cases deadlock with queries during recovery; these result in query cancellation after max_standby_delay seconds have expired. Infrastructure changes have minor effects on normal running, though introduce four new types of WAL record. New test mode "make standbycheck" allows regression tests of static command behaviour on a standby server while in recovery. Typical and extreme dynamic behaviours have been checked via code inspection and manual testing. Few port specific behaviours have been utilised, though primary testing has been on Linux only so far. This commit is the basic patch. Additional changes will follow in this release to enhance some aspects of behaviour, notably improved handling of conflicts, deadlock detection and query cancellation. Changes to VACUUM FULL are also required. Simon Riggs, with significant and lengthy review by Heikki Linnakangas, including streamlined redesign of snapshot creation and two-phase commit. Important contributions from Florian Pflug, Mark Kirkwood, Merlin Moncure, Greg Stark, Gianni Ciolli, Gabriele Bartolini, Hannu Krosing, Robert Haas, Tatsuo Ishii, Hiroyuki Yamada plus support and feedback from many other community members.
2009-12-19 02:32:45 +01:00
*/
void
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
HeapTupleHeaderAdvanceConflictHorizon(HeapTupleHeader tuple,
TransactionId *snapshotConflictHorizon)
Allow read only connections during recovery, known as Hot Standby. Enabled by recovery_connections = on (default) and forcing archive recovery using a recovery.conf. Recovery processing now emulates the original transactions as they are replayed, providing full locking and MVCC behaviour for read only queries. Recovery must enter consistent state before connections are allowed, so there is a delay, typically short, before connections succeed. Replay of recovering transactions can conflict and in some cases deadlock with queries during recovery; these result in query cancellation after max_standby_delay seconds have expired. Infrastructure changes have minor effects on normal running, though introduce four new types of WAL record. New test mode "make standbycheck" allows regression tests of static command behaviour on a standby server while in recovery. Typical and extreme dynamic behaviours have been checked via code inspection and manual testing. Few port specific behaviours have been utilised, though primary testing has been on Linux only so far. This commit is the basic patch. Additional changes will follow in this release to enhance some aspects of behaviour, notably improved handling of conflicts, deadlock detection and query cancellation. Changes to VACUUM FULL are also required. Simon Riggs, with significant and lengthy review by Heikki Linnakangas, including streamlined redesign of snapshot creation and two-phase commit. Important contributions from Florian Pflug, Mark Kirkwood, Merlin Moncure, Greg Stark, Gianni Ciolli, Gabriele Bartolini, Hannu Krosing, Robert Haas, Tatsuo Ishii, Hiroyuki Yamada plus support and feedback from many other community members.
2009-12-19 02:32:45 +01:00
{
TransactionId xmin = HeapTupleHeaderGetXmin(tuple);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
TransactionId xmax = HeapTupleHeaderGetUpdateXid(tuple);
Allow read only connections during recovery, known as Hot Standby. Enabled by recovery_connections = on (default) and forcing archive recovery using a recovery.conf. Recovery processing now emulates the original transactions as they are replayed, providing full locking and MVCC behaviour for read only queries. Recovery must enter consistent state before connections are allowed, so there is a delay, typically short, before connections succeed. Replay of recovering transactions can conflict and in some cases deadlock with queries during recovery; these result in query cancellation after max_standby_delay seconds have expired. Infrastructure changes have minor effects on normal running, though introduce four new types of WAL record. New test mode "make standbycheck" allows regression tests of static command behaviour on a standby server while in recovery. Typical and extreme dynamic behaviours have been checked via code inspection and manual testing. Few port specific behaviours have been utilised, though primary testing has been on Linux only so far. This commit is the basic patch. Additional changes will follow in this release to enhance some aspects of behaviour, notably improved handling of conflicts, deadlock detection and query cancellation. Changes to VACUUM FULL are also required. Simon Riggs, with significant and lengthy review by Heikki Linnakangas, including streamlined redesign of snapshot creation and two-phase commit. Important contributions from Florian Pflug, Mark Kirkwood, Merlin Moncure, Greg Stark, Gianni Ciolli, Gabriele Bartolini, Hannu Krosing, Robert Haas, Tatsuo Ishii, Hiroyuki Yamada plus support and feedback from many other community members.
2009-12-19 02:32:45 +01:00
TransactionId xvac = HeapTupleHeaderGetXvac(tuple);
Remove old-style VACUUM FULL (which was known for a little while as VACUUM FULL INPLACE), along with a boatload of subsidiary code and complexity. Per discussion, the use case for this method of vacuuming is no longer large enough to justify maintaining it; not to mention that we don't wish to invest the work that would be needed to make it play nicely with Hot Standby. Aside from the code directly related to old-style VACUUM FULL, this commit removes support for certain WAL record types that could only be generated within VACUUM FULL, redirect-pointer removal in heap_page_prune, and nontransactional generation of cache invalidation sinval messages (the last being the sticking point for Hot Standby). We still have to retain all code that copes with finding HEAP_MOVED_OFF and HEAP_MOVED_IN flag bits on existing tuples. This can't be removed as long as we want to support in-place update from pre-9.0 databases.
2010-02-08 05:33:55 +01:00
if (tuple->t_infomask & HEAP_MOVED)
Allow read only connections during recovery, known as Hot Standby. Enabled by recovery_connections = on (default) and forcing archive recovery using a recovery.conf. Recovery processing now emulates the original transactions as they are replayed, providing full locking and MVCC behaviour for read only queries. Recovery must enter consistent state before connections are allowed, so there is a delay, typically short, before connections succeed. Replay of recovering transactions can conflict and in some cases deadlock with queries during recovery; these result in query cancellation after max_standby_delay seconds have expired. Infrastructure changes have minor effects on normal running, though introduce four new types of WAL record. New test mode "make standbycheck" allows regression tests of static command behaviour on a standby server while in recovery. Typical and extreme dynamic behaviours have been checked via code inspection and manual testing. Few port specific behaviours have been utilised, though primary testing has been on Linux only so far. This commit is the basic patch. Additional changes will follow in this release to enhance some aspects of behaviour, notably improved handling of conflicts, deadlock detection and query cancellation. Changes to VACUUM FULL are also required. Simon Riggs, with significant and lengthy review by Heikki Linnakangas, including streamlined redesign of snapshot creation and two-phase commit. Important contributions from Florian Pflug, Mark Kirkwood, Merlin Moncure, Greg Stark, Gianni Ciolli, Gabriele Bartolini, Hannu Krosing, Robert Haas, Tatsuo Ishii, Hiroyuki Yamada plus support and feedback from many other community members.
2009-12-19 02:32:45 +01:00
{
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
if (TransactionIdPrecedes(*snapshotConflictHorizon, xvac))
*snapshotConflictHorizon = xvac;
Allow read only connections during recovery, known as Hot Standby. Enabled by recovery_connections = on (default) and forcing archive recovery using a recovery.conf. Recovery processing now emulates the original transactions as they are replayed, providing full locking and MVCC behaviour for read only queries. Recovery must enter consistent state before connections are allowed, so there is a delay, typically short, before connections succeed. Replay of recovering transactions can conflict and in some cases deadlock with queries during recovery; these result in query cancellation after max_standby_delay seconds have expired. Infrastructure changes have minor effects on normal running, though introduce four new types of WAL record. New test mode "make standbycheck" allows regression tests of static command behaviour on a standby server while in recovery. Typical and extreme dynamic behaviours have been checked via code inspection and manual testing. Few port specific behaviours have been utilised, though primary testing has been on Linux only so far. This commit is the basic patch. Additional changes will follow in this release to enhance some aspects of behaviour, notably improved handling of conflicts, deadlock detection and query cancellation. Changes to VACUUM FULL are also required. Simon Riggs, with significant and lengthy review by Heikki Linnakangas, including streamlined redesign of snapshot creation and two-phase commit. Important contributions from Florian Pflug, Mark Kirkwood, Merlin Moncure, Greg Stark, Gianni Ciolli, Gabriele Bartolini, Hannu Krosing, Robert Haas, Tatsuo Ishii, Hiroyuki Yamada plus support and feedback from many other community members.
2009-12-19 02:32:45 +01:00
}
Reduce spurious Hot Standby conflicts from never-visible records. Hot Standby conflicts only with tuples that were visible at some point. So ignore tuples from aborted transactions or for tuples updated/deleted during the inserting transaction when generating the conflict transaction ids. Following detailed analysis and test case by Noah Misch. Original report covered btree delete records, correctly observed by Heikki Linnakangas that this applies to other cases also. Fix covers all sources of cleanup records via common code.
2010-12-09 10:41:47 +01:00
/*
* Ignore tuples inserted by an aborted transaction or if the tuple was
* updated/deleted by the inserting transaction.
*
* Look for a committed hint bit, or if no xmin bit is set, check clog.
*/
Change the way we mark tuples as frozen. Instead of changing the tuple xmin to FrozenTransactionId, the combination of HEAP_XMIN_COMMITTED and HEAP_XMIN_INVALID, which were previously never set together, is now defined as HEAP_XMIN_FROZEN. A variety of previous proposals to freeze tuples opportunistically before vacuum_freeze_min_age is reached have foundered on the objection that replacing xmin by FrozenTransactionId might hinder debugging efforts when things in this area go awry; this patch is intended to solve that problem by keeping the XID around (but largely ignoring the value to which it is set). Third-party code that checks for HEAP_XMIN_INVALID on tuples where HEAP_XMIN_COMMITTED might be set will be broken by this change. To fix, use the new accessor macros in htup_details.h rather than consulting the bits directly. HeapTupleHeaderGetXmin has been modified to return FrozenTransactionId when the infomask bits indicate that the tuple is frozen; use HeapTupleHeaderGetRawXmin when you already know that the tuple isn't marked commited or frozen, or want the raw value anyway. We currently do this in routines that display the xmin for user consumption, in tqual.c where it's known to be safe and important for the avoidance of extra cycles, and in the function-caching code for various procedural languages, which shouldn't invalidate the cache just because the tuple gets frozen. Robert Haas and Andres Freund
2013-12-22 21:49:09 +01:00
if (HeapTupleHeaderXminCommitted(tuple) ||
(!HeapTupleHeaderXminInvalid(tuple) && TransactionIdDidCommit(xmin)))
Reduce spurious Hot Standby conflicts from never-visible records. Hot Standby conflicts only with tuples that were visible at some point. So ignore tuples from aborted transactions or for tuples updated/deleted during the inserting transaction when generating the conflict transaction ids. Following detailed analysis and test case by Noah Misch. Original report covered btree delete records, correctly observed by Heikki Linnakangas that this applies to other cases also. Fix covers all sources of cleanup records via common code.
2010-12-09 10:41:47 +01:00
{
Self review of previous patch. Fix assumption that xmax >= xmin.
2010-12-09 11:20:49 +01:00
if (xmax != xmin &&
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
TransactionIdFollows(xmax, *snapshotConflictHorizon))
*snapshotConflictHorizon = xmax;
Reduce spurious Hot Standby conflicts from never-visible records. Hot Standby conflicts only with tuples that were visible at some point. So ignore tuples from aborted transactions or for tuples updated/deleted during the inserting transaction when generating the conflict transaction ids. Following detailed analysis and test case by Noah Misch. Original report covered btree delete records, correctly observed by Heikki Linnakangas that this applies to other cases also. Fix covers all sources of cleanup records via common code.
2010-12-09 10:41:47 +01:00
}
Allow read only connections during recovery, known as Hot Standby. Enabled by recovery_connections = on (default) and forcing archive recovery using a recovery.conf. Recovery processing now emulates the original transactions as they are replayed, providing full locking and MVCC behaviour for read only queries. Recovery must enter consistent state before connections are allowed, so there is a delay, typically short, before connections succeed. Replay of recovering transactions can conflict and in some cases deadlock with queries during recovery; these result in query cancellation after max_standby_delay seconds have expired. Infrastructure changes have minor effects on normal running, though introduce four new types of WAL record. New test mode "make standbycheck" allows regression tests of static command behaviour on a standby server while in recovery. Typical and extreme dynamic behaviours have been checked via code inspection and manual testing. Few port specific behaviours have been utilised, though primary testing has been on Linux only so far. This commit is the basic patch. Additional changes will follow in this release to enhance some aspects of behaviour, notably improved handling of conflicts, deadlock detection and query cancellation. Changes to VACUUM FULL are also required. Simon Riggs, with significant and lengthy review by Heikki Linnakangas, including streamlined redesign of snapshot creation and two-phase commit. Important contributions from Florian Pflug, Mark Kirkwood, Merlin Moncure, Greg Stark, Gianni Ciolli, Gabriele Bartolini, Hannu Krosing, Robert Haas, Tatsuo Ishii, Hiroyuki Yamada plus support and feedback from many other community members.
2009-12-19 02:32:45 +01:00
}
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
#ifdef USE_PREFETCH
/*
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
* Helper function for heap_index_delete_tuples. Issues prefetch requests for
* prefetch_count buffers. The prefetch_state keeps track of all the buffers
* we can prefetch, and which have already been prefetched; each call to this
* function picks up where the previous call left off.
*
* Note: we expect the deltids array to be sorted in an order that groups TIDs
* by heap block, with all TIDs for each block appearing together in exactly
* one group.
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
*/
static void
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
index_delete_prefetch_buffer(Relation rel,
IndexDeletePrefetchState *prefetch_state,
int prefetch_count)
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
{
BlockNumber cur_hblkno = prefetch_state->cur_hblkno;
int count = 0;
int i;
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
int ndeltids = prefetch_state->ndeltids;
TM_IndexDelete *deltids = prefetch_state->deltids;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
for (i = prefetch_state->next_item;
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
i < ndeltids && count < prefetch_count;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
i++)
{
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
ItemPointer htid = &deltids[i].tid;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
if (cur_hblkno == InvalidBlockNumber ||
ItemPointerGetBlockNumber(htid) != cur_hblkno)
{
cur_hblkno = ItemPointerGetBlockNumber(htid);
PrefetchBuffer(rel, MAIN_FORKNUM, cur_hblkno);
count++;
}
}
/*
* Save the prefetch position so that next time we can continue from that
* position.
*/
prefetch_state->next_item = i;
prefetch_state->cur_hblkno = cur_hblkno;
}
#endif
Add hardening to catch invalid TIDs in indexes. Add hardening to the heapam index tuple deletion path to catch TIDs in index pages that point to a heap item that index tuples should never point to. The corruption we're trying to catch here is particularly tricky to detect, since it typically involves "extra" (corrupt) index tuples, as opposed to the absence of required index tuples in the index. For example, a heap TID from an index page that turns out to point to an LP_UNUSED item in the heap page has a good chance of being caught by one of the new checks. There is a decent chance that the recently fixed parallel VACUUM bug (see commit 9bacec15) would have been caught had that particular check been in place for Postgres 14. No backpatch of this extra hardening for now, though. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzk-4_raTzawWGaiqNvkpwDXxv3y1AQhQyUeHfkU=tFCeA@mail.gmail.com
2021-11-05 03:54:05 +01:00
/*
* Helper function for heap_index_delete_tuples. Checks for index corruption
* involving an invalid TID in index AM caller's index page.
*
* This is an ideal place for these checks. The index AM must hold a buffer
* lock on the index page containing the TIDs we examine here, so we don't
* have to worry about concurrent VACUUMs at all. We can be sure that the
* index is corrupt when htid points directly to an LP_UNUSED item or
* heap-only tuple, which is not the case during standard index scans.
*/
static inline void
index_delete_check_htid(TM_IndexDeleteOp *delstate,
Page page, OffsetNumber maxoff,
ItemPointer htid, TM_IndexStatus *istatus)
{
OffsetNumber indexpagehoffnum = ItemPointerGetOffsetNumber(htid);
ItemId iid;
Assert(OffsetNumberIsValid(istatus->idxoffnum));
if (unlikely(indexpagehoffnum > maxoff))
ereport(ERROR,
(errcode(ERRCODE_INDEX_CORRUPTED),
errmsg_internal("heap tid from index tuple (%u,%u) points past end of heap page line pointer array at offset %u of block %u in index \"%s\"",
ItemPointerGetBlockNumber(htid),
indexpagehoffnum,
istatus->idxoffnum, delstate->iblknum,
RelationGetRelationName(delstate->irel))));
iid = PageGetItemId(page, indexpagehoffnum);
if (unlikely(!ItemIdIsUsed(iid)))
ereport(ERROR,
(errcode(ERRCODE_INDEX_CORRUPTED),
errmsg_internal("heap tid from index tuple (%u,%u) points to unused heap page item at offset %u of block %u in index \"%s\"",
ItemPointerGetBlockNumber(htid),
indexpagehoffnum,
istatus->idxoffnum, delstate->iblknum,
RelationGetRelationName(delstate->irel))));
if (ItemIdHasStorage(iid))
{
HeapTupleHeader htup;
Assert(ItemIdIsNormal(iid));
htup = (HeapTupleHeader) PageGetItem(page, iid);
if (unlikely(HeapTupleHeaderIsHeapOnly(htup)))
ereport(ERROR,
(errcode(ERRCODE_INDEX_CORRUPTED),
errmsg_internal("heap tid from index tuple (%u,%u) points to heap-only tuple at offset %u of block %u in index \"%s\"",
ItemPointerGetBlockNumber(htid),
indexpagehoffnum,
istatus->idxoffnum, delstate->iblknum,
RelationGetRelationName(delstate->irel))));
}
}
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
/*
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
* heapam implementation of tableam's index_delete_tuples interface.
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
*
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
* This helper function is called by index AMs during index tuple deletion.
* See tableam header comments for an explanation of the interface implemented
* here and a general theory of operation. Note that each call here is either
* a simple index deletion call, or a bottom-up index deletion call.
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
*
* It's possible for this to generate a fair amount of I/O, since we may be
* deleting hundreds of tuples from a single index block. To amortize that
* cost to some degree, this uses prefetching and combines repeat accesses to
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
* the same heap block.
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
*/
TransactionId
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
heap_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
{
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
/* Initial assumption is that earlier pruning took care of conflict */
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
TransactionId snapshotConflictHorizon = InvalidTransactionId;
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
BlockNumber blkno = InvalidBlockNumber;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
Buffer buf = InvalidBuffer;
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
Page page = NULL;
OffsetNumber maxoff = InvalidOffsetNumber;
TransactionId priorXmax;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
#ifdef USE_PREFETCH
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
IndexDeletePrefetchState prefetch_state;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
int prefetch_distance;
#endif
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
SnapshotData SnapshotNonVacuumable;
int finalndeltids = 0,
nblocksaccessed = 0;
/* State that's only used in bottom-up index deletion case */
int nblocksfavorable = 0;
int curtargetfreespace = delstate->bottomupfreespace,
lastfreespace = 0,
actualfreespace = 0;
bool bottomup_final_block = false;
InitNonVacuumableSnapshot(SnapshotNonVacuumable, GlobalVisTestFor(rel));
/* Sort caller's deltids array by TID for further processing */
index_delete_sort(delstate);
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
/*
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
* Bottom-up case: resort deltids array in an order attuned to where the
* greatest number of promising TIDs are to be found, and determine how
* many blocks from the start of sorted array should be considered
* favorable. This will also shrink the deltids array in order to
* eliminate completely unfavorable blocks up front.
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
*/
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
if (delstate->bottomup)
nblocksfavorable = bottomup_sort_and_shrink(delstate);
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
#ifdef USE_PREFETCH
/* Initialize prefetch state. */
prefetch_state.cur_hblkno = InvalidBlockNumber;
prefetch_state.next_item = 0;
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
prefetch_state.ndeltids = delstate->ndeltids;
prefetch_state.deltids = delstate->deltids;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
/*
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
* Determine the prefetch distance that we will attempt to maintain.
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
*
Fix deadlock in heap_compute_xid_horizon_for_tuples(). We can't call code that uses syscache while we hold buffer locks on a catalog relation. If passed such a relation, just fall back to the general effective_io_concurrency GUC rather than trying to look up the containing tablespace's IO concurrency setting. We might find a better way to control prefetching in follow-up work, but for now this is enough to avoid the deadlock introduced by commit 558a9165e0. Reviewed-by: Andres Freund Diagnosed-by: Peter Geoghegan Discussion: https://postgr.es/m/CA%2BhUKGLCwPF0S4Mk7S8qw%2BDK0Bq65LueN9rofAA3HHSYikW-Zw%40mail.gmail.com Discussion: https://postgr.es/m/962831d8-c18d-180d-75fb-8b842e3a2742%40chrullrich.net
2019-04-01 22:08:15 +02:00
* Since the caller holds a buffer lock somewhere in rel, we'd better make
* sure that isn't a catalog relation before we call code that does
* syscache lookups, to avoid risk of deadlock.
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
*/
Fix deadlock in heap_compute_xid_horizon_for_tuples(). We can't call code that uses syscache while we hold buffer locks on a catalog relation. If passed such a relation, just fall back to the general effective_io_concurrency GUC rather than trying to look up the containing tablespace's IO concurrency setting. We might find a better way to control prefetching in follow-up work, but for now this is enough to avoid the deadlock introduced by commit 558a9165e0. Reviewed-by: Andres Freund Diagnosed-by: Peter Geoghegan Discussion: https://postgr.es/m/CA%2BhUKGLCwPF0S4Mk7S8qw%2BDK0Bq65LueN9rofAA3HHSYikW-Zw%40mail.gmail.com Discussion: https://postgr.es/m/962831d8-c18d-180d-75fb-8b842e3a2742%40chrullrich.net
2019-04-01 22:08:15 +02:00
if (IsCatalogRelation(rel))
Introduce a maintenance_io_concurrency setting. Introduce a GUC and a tablespace option to control I/O prefetching, much like effective_io_concurrency, but for work that is done on behalf of many client sessions. Use the new setting in heapam.c instead of the hard-coded formula effective_io_concurrency + 10 introduced by commit 558a9165e08. Go with a default value of 10 for now, because it's a round number pretty close to the value used for that existing case. Discussion: https://postgr.es/m/CA%2BhUKGJUw08dPs_3EUcdO6M90GnjofPYrWp4YSLaBkgYwS-AqA%40mail.gmail.com
2020-03-16 00:31:34 +01:00
prefetch_distance = maintenance_io_concurrency;
Fix deadlock in heap_compute_xid_horizon_for_tuples(). We can't call code that uses syscache while we hold buffer locks on a catalog relation. If passed such a relation, just fall back to the general effective_io_concurrency GUC rather than trying to look up the containing tablespace's IO concurrency setting. We might find a better way to control prefetching in follow-up work, but for now this is enough to avoid the deadlock introduced by commit 558a9165e0. Reviewed-by: Andres Freund Diagnosed-by: Peter Geoghegan Discussion: https://postgr.es/m/CA%2BhUKGLCwPF0S4Mk7S8qw%2BDK0Bq65LueN9rofAA3HHSYikW-Zw%40mail.gmail.com Discussion: https://postgr.es/m/962831d8-c18d-180d-75fb-8b842e3a2742%40chrullrich.net
2019-04-01 22:08:15 +02:00
else
Introduce a maintenance_io_concurrency setting. Introduce a GUC and a tablespace option to control I/O prefetching, much like effective_io_concurrency, but for work that is done on behalf of many client sessions. Use the new setting in heapam.c instead of the hard-coded formula effective_io_concurrency + 10 introduced by commit 558a9165e08. Go with a default value of 10 for now, because it's a round number pretty close to the value used for that existing case. Discussion: https://postgr.es/m/CA%2BhUKGJUw08dPs_3EUcdO6M90GnjofPYrWp4YSLaBkgYwS-AqA%40mail.gmail.com
2020-03-16 00:31:34 +01:00
prefetch_distance =
get_tablespace_maintenance_io_concurrency(rel->rd_rel->reltablespace);
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
/* Cap initial prefetch distance for bottom-up deletion caller */
if (delstate->bottomup)
{
Assert(nblocksfavorable >= 1);
Assert(nblocksfavorable <= BOTTOMUP_MAX_NBLOCKS);
prefetch_distance = Min(prefetch_distance, nblocksfavorable);
}
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
/* Start prefetching. */
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
index_delete_prefetch_buffer(rel, &prefetch_state, prefetch_distance);
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
#endif
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
/* Iterate over deltids, determine which to delete, check their horizon */
Assert(delstate->ndeltids > 0);
for (int i = 0; i < delstate->ndeltids; i++)
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
{
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
TM_IndexDelete *ideltid = &delstate->deltids[i];
TM_IndexStatus *istatus = delstate->status + ideltid->id;
ItemPointer htid = &ideltid->tid;
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
OffsetNumber offnum;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
/*
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
* Read buffer, and perform required extra steps each time a new block
* is encountered. Avoid refetching if it's the same block as the one
* from the last htid.
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
*/
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
if (blkno == InvalidBlockNumber ||
ItemPointerGetBlockNumber(htid) != blkno)
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
{
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
/*
* Consider giving up early for bottom-up index deletion caller
* first. (Only prefetch next-next block afterwards, when it
* becomes clear that we're at least going to access the next
* block in line.)
*
* Sometimes the first block frees so much space for bottom-up
* caller that the deletion process can end without accessing any
* more blocks. It is usually necessary to access 2 or 3 blocks
* per bottom-up deletion operation, though.
*/
if (delstate->bottomup)
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
{
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
/*
* We often allow caller to delete a few additional items
* whose entries we reached after the point that space target
* from caller was satisfied. The cost of accessing the page
* was already paid at that point, so it made sense to finish
* it off. When that happened, we finalize everything here
* (by finishing off the whole bottom-up deletion operation
* without needlessly paying the cost of accessing any more
* blocks).
*/
if (bottomup_final_block)
break;
/*
* Give up when we didn't enable our caller to free any
* additional space as a result of processing the page that we
* just finished up with. This rule is the main way in which
* we keep the cost of bottom-up deletion under control.
*/
if (nblocksaccessed >= 1 && actualfreespace == lastfreespace)
break;
lastfreespace = actualfreespace; /* for next time */
/*
* Deletion operation (which is bottom-up) will definitely
* access the next block in line. Prepare for that now.
*
* Decay target free space so that we don't hang on for too
* long with a marginal case. (Space target is only truly
* helpful when it allows us to recognize that we don't need
* to access more than 1 or 2 blocks to satisfy caller due to
* agreeable workload characteristics.)
*
* We are a bit more patient when we encounter contiguous
* blocks, though: these are treated as favorable blocks. The
* decay process is only applied when the next block in line
* is not a favorable/contiguous block. This is not an
* exception to the general rule; we still insist on finding
* at least one deletable item per block accessed. See
* bottomup_nblocksfavorable() for full details of the theory
* behind favorable blocks and heap block locality in general.
*
* Note: The first block in line is always treated as a
* favorable block, so the earliest possible point that the
* decay can be applied is just before we access the second
* block in line. The Assert() verifies this for us.
*/
Assert(nblocksaccessed > 0 || nblocksfavorable > 0);
if (nblocksfavorable > 0)
nblocksfavorable--;
else
curtargetfreespace /= 2;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
}
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
/* release old buffer */
if (BufferIsValid(buf))
UnlockReleaseBuffer(buf);
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
blkno = ItemPointerGetBlockNumber(htid);
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
buf = ReadBuffer(rel, blkno);
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
nblocksaccessed++;
Assert(!delstate->bottomup ||
nblocksaccessed <= BOTTOMUP_MAX_NBLOCKS);
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
#ifdef USE_PREFETCH
/*
* To maintain the prefetch distance, prefetch one more page for
* each page we read.
*/
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
index_delete_prefetch_buffer(rel, &prefetch_state, 1);
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
#endif
Get heap page max offset with buffer lock held. On further reflection it seems better to call PageGetMaxOffsetNumber() after acquiring a buffer lock on the page. This shouldn't really matter, but doing it this way is cleaner. Follow-up to commit 42288174. Backpatch: 12-, just like commit 42288174
2020-12-31 02:21:42 +01:00
LockBuffer(buf, BUFFER_LOCK_SHARE);
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
page = BufferGetPage(buf);
maxoff = PageGetMaxOffsetNumber(page);
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
}
Add hardening to catch invalid TIDs in indexes. Add hardening to the heapam index tuple deletion path to catch TIDs in index pages that point to a heap item that index tuples should never point to. The corruption we're trying to catch here is particularly tricky to detect, since it typically involves "extra" (corrupt) index tuples, as opposed to the absence of required index tuples in the index. For example, a heap TID from an index page that turns out to point to an LP_UNUSED item in the heap page has a good chance of being caught by one of the new checks. There is a decent chance that the recently fixed parallel VACUUM bug (see commit 9bacec15) would have been caught had that particular check been in place for Postgres 14. No backpatch of this extra hardening for now, though. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzk-4_raTzawWGaiqNvkpwDXxv3y1AQhQyUeHfkU=tFCeA@mail.gmail.com
2021-11-05 03:54:05 +01:00
/*
* In passing, detect index corruption involving an index page with a
* TID that points to a location in the heap that couldn't possibly be
* correct. We only do this with actual TIDs from caller's index page
* (not items reached by traversing through a HOT chain).
*/
index_delete_check_htid(delstate, page, maxoff, htid, istatus);
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
if (istatus->knowndeletable)
Assert(!delstate->bottomup && !istatus->promising);
else
{
ItemPointerData tmp = *htid;
HeapTupleData heapTuple;
/* Are any tuples from this HOT chain non-vacuumable? */
if (heap_hot_search_buffer(&tmp, rel, buf, &SnapshotNonVacuumable,
&heapTuple, NULL, true))
continue; /* can't delete entry */
/* Caller will delete, since whole HOT chain is vacuumable */
istatus->knowndeletable = true;
/* Maintain index free space info for bottom-up deletion case */
if (delstate->bottomup)
{
Assert(istatus->freespace > 0);
actualfreespace += istatus->freespace;
if (actualfreespace >= curtargetfreespace)
bottomup_final_block = true;
}
}
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
/*
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
* Maintain snapshotConflictHorizon value for deletion operation as a
* whole by advancing current value using heap tuple headers. This is
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
* loosely based on the logic for pruning a HOT chain.
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
*/
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
offnum = ItemPointerGetOffsetNumber(htid);
priorXmax = InvalidTransactionId; /* cannot check first XMIN */
for (;;)
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
{
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
ItemId lp;
HeapTupleHeader htup;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
Document issue with heapam line pointer truncation. Checking that an offset number isn't past the end of a heap page's line pointer array was just a defensive sanity check for HOT-chain traversal code before commit 3c3b8a4b. It's etrictly necessary now, though. Add comments that reference the issue to code in heapam that needs to get it right. Per suggestion from Alexander Lakhin. Discussion: https://postgr.es/m/f76a292c-9170-1aef-91a0-59d9443b99a3@gmail.com
2021-09-23 04:21:36 +02:00
/* Sanity check (pure paranoia) */
if (offnum < FirstOffsetNumber)
break;
/*
* An offset past the end of page's line pointer array is possible
* when the array was truncated
*/
if (offnum > maxoff)
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
break;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
lp = PageGetItemId(page, offnum);
if (ItemIdIsRedirected(lp))
{
offnum = ItemIdGetRedirect(lp);
continue;
}
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
/*
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
* We'll often encounter LP_DEAD line pointers (especially with an
* entry marked knowndeletable by our caller up front). No heap
* tuple headers get examined for an htid that leads us to an
* LP_DEAD item. This is okay because the earlier pruning
* operation that made the line pointer LP_DEAD in the first place
* must have considered the original tuple header as part of
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
* generating its own snapshotConflictHorizon value.
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
*
Remove tupgone special case from vacuumlazy.c. Retry the call to heap_prune_page() in rare cases where there is disagreement between the heap_prune_page() call and the call to HeapTupleSatisfiesVacuum() that immediately follows. Disagreement is possible when a concurrently-aborted transaction makes a tuple DEAD during the tiny window between each step. This was the only case where a tuple considered DEAD by VACUUM still had storage following pruning. VACUUM's definition of dead tuples is now uniformly simple and unambiguous: dead tuples from each page are always LP_DEAD line pointers that were encountered just after we performed pruning (and just before we considered freezing remaining items with tuple storage). Eliminating the tupgone=true special case enables INDEX_CLEANUP=off style skipping of index vacuuming that takes place based on flexible, dynamic criteria. The INDEX_CLEANUP=off case had to know about skipping indexes up-front before now, due to a subtle interaction with the special case (see commit dd695979) -- this was a special case unto itself. Now there are no special cases. And so now it won't matter when or how we decide to skip index vacuuming: it won't affect how pruning behaves, and it won't be affected by any of the implementation details of pruning or freezing. Also remove XLOG_HEAP2_CLEANUP_INFO records. These are no longer necessary because we now rely entirely on heap pruning taking care of recovery conflicts. There is no longer any need to generate recovery conflicts for DEAD tuples that pruning just missed. This also means that heap vacuuming now uses exactly the same strategy for recovery conflicts as index vacuuming always has: REDO routines never need to process a latestRemovedXid from the WAL record, since earlier REDO of the WAL record from pruning is sufficient in all cases. The generic XLOG_HEAP2_CLEAN record type is now split into two new record types to reflect this new division (these are called XLOG_HEAP2_PRUNE and XLOG_HEAP2_VACUUM). Also stop acquiring a super-exclusive lock for heap pages when they're vacuumed during VACUUM's second heap pass. A regular exclusive lock is enough. This is correct because heap page vacuuming is now strictly a matter of setting the LP_DEAD line pointers to LP_UNUSED. No other backend can have a pointer to a tuple located in a pinned buffer that can be invalidated by a concurrent heap page vacuum operation. Heap vacuuming can now be thought of as conceptually similar to index vacuuming and conceptually dissimilar to heap pruning. Heap pruning now has sole responsibility for anything involving the logical contents of the database (e.g., managing transaction status information, recovery conflicts, considering what to do with HOT chains). Index vacuuming and heap vacuuming are now only concerned with recycling garbage items from physical data structures that back the logical database. Bump XLOG_PAGE_MAGIC due to pruning and heap page vacuum WAL record changes. Credit for the idea of retrying pruning a page to avoid the tupgone case goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Discussion: https://postgr.es/m/CAH2-WznneCXTzuFmcwx_EyRQgfsfJAAsu+CsqRFmFXCAar=nJw@mail.gmail.com
2021-04-06 17:49:22 +02:00
* Relying on XLOG_HEAP2_PRUNE records like this is the same
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
* strategy that index vacuuming uses in all cases. Index VACUUM
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
* WAL records don't even have a snapshotConflictHorizon field of
* their own for this reason.
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
*/
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
if (!ItemIdIsNormal(lp))
break;
htup = (HeapTupleHeader) PageGetItem(page, lp);
/*
* Check the tuple XMIN against prior XMAX, if any
*/
if (TransactionIdIsValid(priorXmax) &&
!TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
break;
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
HeapTupleHeaderAdvanceConflictHorizon(htup,
&snapshotConflictHorizon);
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
Fix index deletion latestRemovedXid bug. The logic for determining the latest removed XID for the purposes of generating recovery conflicts in REDO routines was subtly broken. It failed to follow links from HOT chains, and so failed to consider all relevant heap tuple headers in some cases. To fix, expand the loop that deals with LP_REDIRECT line pointers to also deal with HOT chains. The new version of the loop is loosely based on a similar loop from heap_prune_chain(). The impact of this bug is probably quite limited, since the horizon code necessarily deals with heap tuples that are pointed to by LP_DEAD-set index tuples. The process of setting LP_DEAD index tuples (e.g. within the kill_prior_tuple mechanism) is highly correlated with opportunistic pruning of pointed-to heap tuples. Plus the question of generating a recovery conflict usually comes up some time after index tuple LP_DEAD bits were initially set, unlike heap pruning, where a latestRemovedXid is generated at the point of the pruning operation (heap pruning has no deferred "would-be page split" style processing that produces conflicts lazily). Only backpatch to Postgres 12, the first version where this logic runs during original execution (following commit 558a9165e08). The index latestRemovedXid mechanism has had the same bug since it first appeared over 10 years ago (in commit a760893d), but backpatching to all supported versions now seems like a bad idea on balance. Running the new improved code during recovery seems risky, especially given the lack of complaints from the field. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-Wz=Eib393+HHcERK_9MtgNS7Ew1HY=RDC_g6GL46zM5C6Q@mail.gmail.com Backpatch: 12-
2020-12-31 01:29:05 +01:00
/*
* If the tuple is not HOT-updated, then we are at the end of this
* HOT-chain. No need to visit later tuples from the same update
* chain (they get their own index entries) -- just move on to
* next htid from index AM caller.
*/
if (!HeapTupleHeaderIsHotUpdated(htup))
break;
/* Advance to next HOT chain member */
Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blkno);
offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
priorXmax = HeapTupleHeaderGetUpdateXid(htup);
}
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
/* Enable further/final shrinking of deltids for caller */
finalndeltids = i + 1;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
}
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
UnlockReleaseBuffer(buf);
/*
* Shrink deltids array to exclude non-deletable entries at the end. This
* is not just a minor optimization. Final deltids array size might be
* zero for a bottom-up caller. Index AM is explicitly allowed to rely on
* ndeltids being zero in all cases with zero total deletable entries.
*/
Assert(finalndeltids > 0 || delstate->bottomup);
delstate->ndeltids = finalndeltids;
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
return snapshotConflictHorizon;
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
}
/*
* Specialized inlineable comparison function for index_delete_sort()
*/
static inline int
index_delete_sort_cmp(TM_IndexDelete *deltid1, TM_IndexDelete *deltid2)
{
ItemPointer tid1 = &deltid1->tid;
ItemPointer tid2 = &deltid2->tid;
{
BlockNumber blk1 = ItemPointerGetBlockNumber(tid1);
BlockNumber blk2 = ItemPointerGetBlockNumber(tid2);
if (blk1 != blk2)
return (blk1 < blk2) ? -1 : 1;
}
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
{
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
OffsetNumber pos1 = ItemPointerGetOffsetNumber(tid1);
OffsetNumber pos2 = ItemPointerGetOffsetNumber(tid2);
if (pos1 != pos2)
return (pos1 < pos2) ? -1 : 1;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
}
Demote pg_unreachable() in heapam to an assertion. Commit d168b66682, which overhauled index deletion, added a pg_unreachable() to the end of a sort comparator used when sorting heap TIDs from an index page. This allows the compiler to apply optimizations that assume that the heap TIDs from the index AM must always be unique. That doesn't seem like a good idea now, given recent reports of corruption involving duplicate TIDs in indexes on Postgres 14. Demote to an assertion, just in case. Backpatch: 14-, where index deletion was overhauled.
2021-10-29 19:53:48 +02:00
Assert(false);
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
return 0;
}
/*
* Sort deltids array from delstate by TID. This prepares it for further
* processing by heap_index_delete_tuples().
*
* This operation becomes a noticeable consumer of CPU cycles with some
* workloads, so we go to the trouble of specialization/micro optimization.
* We use shellsort for this because it's easy to specialize, compiles to
* relatively few instructions, and is adaptive to presorted inputs/subsets
* (which are typical here).
*/
static void
index_delete_sort(TM_IndexDeleteOp *delstate)
{
TM_IndexDelete *deltids = delstate->deltids;
int ndeltids = delstate->ndeltids;
int low = 0;
/*
* Shellsort gap sequence (taken from Sedgewick-Incerpi paper).
*
* This implementation is fast with array sizes up to ~4500. This covers
* all supported BLCKSZ values.
*/
const int gaps[9] = {1968, 861, 336, 112, 48, 21, 7, 3, 1};
/* Think carefully before changing anything here -- keep swaps cheap */
Static assertions cleanup Because we added StaticAssertStmt() first before StaticAssertDecl(), some uses as well as the instructions in c.h are now a bit backwards from the "native" way static assertions are meant to be used in C. This updates the guidance and moves some static assertions to better places. Specifically, since the addition of StaticAssertDecl(), we can put static assertions at the file level. This moves a number of static assertions out of function bodies, where they might have been stuck out of necessity, to perhaps better places at the file level or in header files. Also, when the static assertion appears in a position where a declaration is allowed, then using StaticAssertDecl() is more native than StaticAssertStmt(). Reviewed-by: John Naylor <john.naylor@enterprisedb.com> Discussion: https://www.postgresql.org/message-id/flat/941a04e7-dd6f-c0e4-8cdf-a33b3338cbda%40enterprisedb.com
2022-12-08 14:30:01 +01:00
StaticAssertDecl(sizeof(TM_IndexDelete) <= 8,
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
"element size exceeds 8 bytes");
for (int g = 0; g < lengthof(gaps); g++)
{
for (int hi = gaps[g], i = low + hi; i < ndeltids; i++)
{
TM_IndexDelete d = deltids[i];
int j = i;
while (j >= hi && index_delete_sort_cmp(&deltids[j - hi], &d) >= 0)
{
deltids[j] = deltids[j - hi];
j -= hi;
}
deltids[j] = d;
}
}
}
/*
* Returns how many blocks should be considered favorable/contiguous for a
* bottom-up index deletion pass. This is a number of heap blocks that starts
* from and includes the first block in line.
*
* There is always at least one favorable block during bottom-up index
* deletion. In the worst case (i.e. with totally random heap blocks) the
* first block in line (the only favorable block) can be thought of as a
* degenerate array of contiguous blocks that consists of a single block.
* heap_index_delete_tuples() will expect this.
*
* Caller passes blockgroups, a description of the final order that deltids
* will be sorted in for heap_index_delete_tuples() bottom-up index deletion
* processing. Note that deltids need not actually be sorted just yet (caller
* only passes deltids to us so that we can interpret blockgroups).
*
* You might guess that the existence of contiguous blocks cannot matter much,
* since in general the main factor that determines which blocks we visit is
* the number of promising TIDs, which is a fixed hint from the index AM.
* We're not really targeting the general case, though -- the actual goal is
* to adapt our behavior to a wide variety of naturally occurring conditions.
* The effects of most of the heuristics we apply are only noticeable in the
* aggregate, over time and across many _related_ bottom-up index deletion
* passes.
*
* Deeming certain blocks favorable allows heapam to recognize and adapt to
* workloads where heap blocks visited during bottom-up index deletion can be
* accessed contiguously, in the sense that each newly visited block is the
* neighbor of the block that bottom-up deletion just finished processing (or
* close enough to it). It will likely be cheaper to access more favorable
* blocks sooner rather than later (e.g. in this pass, not across a series of
* related bottom-up passes). Either way it is probably only a matter of time
* (or a matter of further correlated version churn) before all blocks that
* appear together as a single large batch of favorable blocks get accessed by
* _some_ bottom-up pass. Large batches of favorable blocks tend to either
* appear almost constantly or not even once (it all depends on per-index
* workload characteristics).
*
* Note that the blockgroups sort order applies a power-of-two bucketing
* scheme that creates opportunities for contiguous groups of blocks to get
* batched together, at least with workloads that are naturally amenable to
* being driven by heap block locality. This doesn't just enhance the spatial
* locality of bottom-up heap block processing in the obvious way. It also
* enables temporal locality of access, since sorting by heap block number
* naturally tends to make the bottom-up processing order deterministic.
*
* Consider the following example to get a sense of how temporal locality
* might matter: There is a heap relation with several indexes, each of which
* is low to medium cardinality. It is subject to constant non-HOT updates.
* The updates are skewed (in one part of the primary key, perhaps). None of
* the indexes are logically modified by the UPDATE statements (if they were
* then bottom-up index deletion would not be triggered in the first place).
* Naturally, each new round of index tuples (for each heap tuple that gets a
* heap_update() call) will have the same heap TID in each and every index.
* Since these indexes are low cardinality and never get logically modified,
* heapam processing during bottom-up deletion passes will access heap blocks
* in approximately sequential order. Temporal locality of access occurs due
* to bottom-up deletion passes behaving very similarly across each of the
* indexes at any given moment. This keeps the number of buffer misses needed
* to visit heap blocks to a minimum.
*/
static int
bottomup_nblocksfavorable(IndexDeleteCounts *blockgroups, int nblockgroups,
TM_IndexDelete *deltids)
{
int64 lastblock = -1;
int nblocksfavorable = 0;
Assert(nblockgroups >= 1);
Assert(nblockgroups <= BOTTOMUP_MAX_NBLOCKS);
/*
* We tolerate heap blocks that will be accessed only slightly out of
* physical order. Small blips occur when a pair of almost-contiguous
* blocks happen to fall into different buckets (perhaps due only to a
* small difference in npromisingtids that the bucketing scheme didn't
* quite manage to ignore). We effectively ignore these blips by applying
* a small tolerance. The precise tolerance we use is a little arbitrary,
* but it works well enough in practice.
*/
for (int b = 0; b < nblockgroups; b++)
{
IndexDeleteCounts *group = blockgroups + b;
TM_IndexDelete *firstdtid = deltids + group->ifirsttid;
BlockNumber block = ItemPointerGetBlockNumber(&firstdtid->tid);
if (lastblock != -1 &&
((int64) block < lastblock - BOTTOMUP_TOLERANCE_NBLOCKS ||
(int64) block > lastblock + BOTTOMUP_TOLERANCE_NBLOCKS))
break;
nblocksfavorable++;
lastblock = block;
}
/* Always indicate that there is at least 1 favorable block */
Assert(nblocksfavorable >= 1);
return nblocksfavorable;
}
/*
* qsort comparison function for bottomup_sort_and_shrink()
*/
static int
bottomup_sort_and_shrink_cmp(const void *arg1, const void *arg2)
{
const IndexDeleteCounts *group1 = (const IndexDeleteCounts *) arg1;
const IndexDeleteCounts *group2 = (const IndexDeleteCounts *) arg2;
/*
* Most significant field is npromisingtids (which we invert the order of
* so as to sort in desc order).
*
* Caller should have already normalized npromisingtids fields into
* power-of-two values (buckets).
*/
if (group1->npromisingtids > group2->npromisingtids)
return -1;
if (group1->npromisingtids < group2->npromisingtids)
return 1;
/*
* Tiebreak: desc ntids sort order.
*
* We cannot expect power-of-two values for ntids fields. We should
* behave as if they were already rounded up for us instead.
*/
if (group1->ntids != group2->ntids)
{
uint32 ntids1 = pg_nextpower2_32((uint32) group1->ntids);
uint32 ntids2 = pg_nextpower2_32((uint32) group2->ntids);
if (ntids1 > ntids2)
return -1;
if (ntids1 < ntids2)
return 1;
}
/*
* Tiebreak: asc offset-into-deltids-for-block (offset to first TID for
* block in deltids array) order.
*
* This is equivalent to sorting in ascending heap block number order
* (among otherwise equal subsets of the array). This approach allows us
* to avoid accessing the out-of-line TID. (We rely on the assumption
* that the deltids array was sorted in ascending heap TID order when
* these offsets to the first TID from each heap block group were formed.)
*/
if (group1->ifirsttid > group2->ifirsttid)
return 1;
if (group1->ifirsttid < group2->ifirsttid)
return -1;
pg_unreachable();
return 0;
}
/*
* heap_index_delete_tuples() helper function for bottom-up deletion callers.
*
* Sorts deltids array in the order needed for useful processing by bottom-up
* deletion. The array should already be sorted in TID order when we're
* called. The sort process groups heap TIDs from deltids into heap block
* groupings. Earlier/more-promising groups/blocks are usually those that are
* known to have the most "promising" TIDs.
*
* Sets new size of deltids array (ndeltids) in state. deltids will only have
* TIDs from the BOTTOMUP_MAX_NBLOCKS most promising heap blocks when we
* return. This often means that deltids will be shrunk to a small fraction
* of its original size (we eliminate many heap blocks from consideration for
* caller up front).
*
* Returns the number of "favorable" blocks. See bottomup_nblocksfavorable()
* for a definition and full details.
*/
static int
bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate)
{
IndexDeleteCounts *blockgroups;
TM_IndexDelete *reordereddeltids;
BlockNumber curblock = InvalidBlockNumber;
int nblockgroups = 0;
int ncopied = 0;
int nblocksfavorable = 0;
Assert(delstate->bottomup);
Assert(delstate->ndeltids > 0);
/* Calculate per-heap-block count of TIDs */
blockgroups = palloc(sizeof(IndexDeleteCounts) * delstate->ndeltids);
for (int i = 0; i < delstate->ndeltids; i++)
{
TM_IndexDelete *ideltid = &delstate->deltids[i];
TM_IndexStatus *istatus = delstate->status + ideltid->id;
ItemPointer htid = &ideltid->tid;
bool promising = istatus->promising;
if (curblock != ItemPointerGetBlockNumber(htid))
{
/* New block group */
nblockgroups++;
Assert(curblock < ItemPointerGetBlockNumber(htid) ||
!BlockNumberIsValid(curblock));
curblock = ItemPointerGetBlockNumber(htid);
blockgroups[nblockgroups - 1].ifirsttid = i;
blockgroups[nblockgroups - 1].ntids = 1;
blockgroups[nblockgroups - 1].npromisingtids = 0;
}
else
{
blockgroups[nblockgroups - 1].ntids++;
}
if (promising)
blockgroups[nblockgroups - 1].npromisingtids++;
}
/*
* We're about ready to sort block groups to determine the optimal order
* for visiting heap blocks. But before we do, round the number of
Fix comment about promising tuples. Oversight in commit d168b666823, which added bottom-up index deletion.
2021-03-16 21:38:52 +01:00
* promising tuples for each block group up to the next power-of-two,
* unless it is very low (less than 4), in which case we round up to 4.
* npromisingtids is far too noisy to trust when choosing between a pair
* of block groups that both have very low values.
Enhance nbtree index tuple deletion. Teach nbtree and heapam to cooperate in order to eagerly remove duplicate tuples representing dead MVCC versions. This is "bottom-up deletion". Each bottom-up deletion pass is triggered lazily in response to a flood of versions on an nbtree leaf page. This usually involves a "logically unchanged index" hint (these are produced by the executor mechanism added by commit 9dc718bd). The immediate goal of bottom-up index deletion is to avoid "unnecessary" page splits caused entirely by version duplicates. It naturally has an even more useful effect, though: it acts as a backstop against accumulating an excessive number of index tuple versions for any given _logical row_. Bottom-up index deletion complements what we might now call "top-down index deletion": index vacuuming performed by VACUUM. Bottom-up index deletion responds to the immediate local needs of queries, while leaving it up to autovacuum to perform infrequent clean sweeps of the index. The overall effect is to avoid certain pathological performance issues related to "version churn" from UPDATEs. The previous tableam interface used by index AMs to perform tuple deletion (the table_compute_xid_horizon_for_tuples() function) has been replaced with a new interface that supports certain new requirements. Many (perhaps all) of the capabilities added to nbtree by this commit could also be extended to other index AMs. That is left as work for a later commit. Extend deletion of LP_DEAD-marked index tuples in nbtree by adding logic to consider extra index tuples (that are not LP_DEAD-marked) for deletion in passing. This increases the number of index tuples deleted significantly in many cases. The LP_DEAD deletion process (which is now called "simple deletion" to clearly distinguish it from bottom-up deletion) won't usually need to visit any extra table blocks to check these extra tuples. We have to visit the same table blocks anyway to generate a latestRemovedXid value (at least in the common case where the index deletion operation's WAL record needs such a value). Testing has shown that the "extra tuples" simple deletion enhancement increases the number of index tuples deleted with almost any workload that has LP_DEAD bits set in leaf pages. That is, it almost never fails to delete at least a few extra index tuples. It helps most of all in cases that happen to naturally have a lot of delete-safe tuples. It's not uncommon for an individual deletion operation to end up deleting an order of magnitude more index tuples compared to the old naive approach (e.g., custom instrumentation of the patch shows that this happens fairly often when the regression tests are run). Add a further enhancement that augments simple deletion and bottom-up deletion in indexes that make use of deduplication: Teach nbtree's _bt_delitems_delete() function to support granular TID deletion in posting list tuples. It is now possible to delete individual TIDs from posting list tuples provided the TIDs have a tableam block number of a table block that gets visited as part of the deletion process (visiting the table block can be triggered directly or indirectly). Setting the LP_DEAD bit of a posting list tuple is still an all-or-nothing thing, but that matters much less now that deletion only needs to start out with the right _general_ idea about which index tuples are deletable. Bump XLOG_PAGE_MAGIC because xl_btree_delete changed. No bump in BTREE_VERSION, since there are no changes to the on-disk representation of nbtree indexes. Indexes built on PostgreSQL 12 or PostgreSQL 13 will automatically benefit from bottom-up index deletion (i.e. no reindexing required) following a pg_upgrade. The enhancement to simple deletion is available with all B-Tree indexes following a pg_upgrade, no matter what PostgreSQL version the user upgrades from. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Heikki Linnakangas <hlinnaka@iki.fi> Reviewed-By: Victor Yegorov <vyegorov@gmail.com> Discussion: https://postgr.es/m/CAH2-Wzm+maE3apHB8NOtmM=p-DO65j2V5GzAWCOEEuy3JZgb2g@mail.gmail.com
2021-01-13 18:21:32 +01:00
*
* This scheme divides heap blocks/block groups into buckets. Each bucket
* contains blocks that have _approximately_ the same number of promising
* TIDs as each other. The goal is to ignore relatively small differences
* in the total number of promising entries, so that the whole process can
* give a little weight to heapam factors (like heap block locality)
* instead. This isn't a trade-off, really -- we have nothing to lose. It
* would be foolish to interpret small differences in npromisingtids
* values as anything more than noise.
*
* We tiebreak on nhtids when sorting block group subsets that have the
* same npromisingtids, but this has the same issues as npromisingtids,
* and so nhtids is subject to the same power-of-two bucketing scheme. The
* only reason that we don't fix nhtids in the same way here too is that
* we'll need accurate nhtids values after the sort. We handle nhtids
* bucketization dynamically instead (in the sort comparator).
*
* See bottomup_nblocksfavorable() for a full explanation of when and how
* heap locality/favorable blocks can significantly influence when and how
* heap blocks are accessed.
*/
for (int b = 0; b < nblockgroups; b++)
{
IndexDeleteCounts *group = blockgroups + b;
/* Better off falling back on nhtids with low npromisingtids */
if (group->npromisingtids <= 4)
group->npromisingtids = 4;
else
group->npromisingtids =
pg_nextpower2_32((uint32) group->npromisingtids);
}
/* Sort groups and rearrange caller's deltids array */
qsort(blockgroups, nblockgroups, sizeof(IndexDeleteCounts),
bottomup_sort_and_shrink_cmp);
reordereddeltids = palloc(delstate->ndeltids * sizeof(TM_IndexDelete));
nblockgroups = Min(BOTTOMUP_MAX_NBLOCKS, nblockgroups);
/* Determine number of favorable blocks at the start of final deltids */
nblocksfavorable = bottomup_nblocksfavorable(blockgroups, nblockgroups,
delstate->deltids);
for (int b = 0; b < nblockgroups; b++)
{
IndexDeleteCounts *group = blockgroups + b;
TM_IndexDelete *firstdtid = delstate->deltids + group->ifirsttid;
memcpy(reordereddeltids + ncopied, firstdtid,
sizeof(TM_IndexDelete) * group->ntids);
ncopied += group->ntids;
}
/* Copy final grouped and sorted TIDs back into start of caller's array */
memcpy(delstate->deltids, reordereddeltids,
sizeof(TM_IndexDelete) * ncopied);
delstate->ndeltids = ncopied;
pfree(reordereddeltids);
pfree(blockgroups);
return nblocksfavorable;
Compute XID horizon for page level index vacuum on primary. Previously the xid horizon was only computed during WAL replay. That had two major problems: 1) It relied on knowing what the table pointed to looks like. That was easy enough before the introducing of tableam (we knew it had to be heap, although some trickery around logging the heap relfilenodes was required). But to properly handle table AMs we need per-database catalog access to look up the AM handler, which recovery doesn't allow. 2) Not knowing the xid horizon also makes it hard to support logical decoding on standbys. When on a catalog table, we need to be able to conflict with slots that have an xid horizon that's too old. But computing the horizon by visiting the heap only works once consistency is reached, but we always need to be able to detect conflicts. There's also a secondary problem, in that the current method performs redundant work on every standby. But that's counterbalanced by potentially computing the value when not necessary (either because there's no standby, or because there's no connected backends). Solve 1) and 2) by moving computation of the xid horizon to the primary and by involving tableam in the computation of the horizon. To address the potentially increased overhead, increase the efficiency of the xid horizon computation for heap by sorting the tids, and eliminating redundant buffer accesses. When prefetching is available, additionally perform prefetching of buffers. As this is more of a maintenance task, rather than something routinely done in every read only query, we add an arbitrary 10 to the effective concurrency - thereby using IO concurrency, when not globally enabled. That's possibly not the perfect formula, but seems good enough for now. Bumps WAL format, as latestRemovedXid is now part of the records, and the heap's relfilenode isn't anymore. Author: Andres Freund, Amit Khandekar, Robert Haas Reviewed-By: Robert Haas Discussion: https://postgr.es/m/20181212204154.nsxf3gzqv3gesl32@alap3.anarazel.de https://postgr.es/m/20181214014235.dal5ogljs3bmlq44@alap3.anarazel.de https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
2019-03-26 22:41:46 +01:00
}
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
/*
* Perform XLogInsert for a heap-visible operation. 'block' is the block
* being marked all-visible, and vm_buffer is the buffer containing the
* corresponding visibility map block. Both should have already been modified
* and dirtied.
Allow I/O reliability checks using 16-bit checksums Checksums are set immediately prior to flush out of shared buffers and checked when pages are read in again. Hint bit setting will require full page write when block is dirtied, which causes various infrastructure changes. Extensive comments, docs and README. WARNING message thrown if checksum fails on non-all zeroes page; ERROR thrown but can be disabled with ignore_checksum_failure = on. Feature enabled by an initdb option, since transition from option off to option on is long and complex and has not yet been implemented. Default is not to use checksums. Checksum used is WAL CRC-32 truncated to 16-bits. Simon Riggs, Jeff Davis, Greg Smith Wide input and assistance from many community members. Thank you.
2013-03-22 14:54:07 +01:00
*
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
* snapshotConflictHorizon comes from the largest xmin on the page being
* marked all-visible. REDO routine uses it to generate recovery conflicts.
*
Document WAL rules related to PD_ALL_VISIBLE in README. Also improve comments. Discussion: https://postgr.es/m/a50005c1c537f89bb359057fd70e66bb83bce969.camel@j-davis.com Reviewed-by: Peter Geoghegan
2022-11-12 17:37:50 +01:00
* If checksums or wal_log_hints are enabled, we may also generate a full-page
* image of heap_buffer. Otherwise, we optimize away the FPI (by specifying
* REGBUF_NO_IMAGE for the heap buffer), in which case the caller should *not*
* update the heap page's LSN.
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
*/
XLogRecPtr
Pass down table relation into more index relation functions This is done in preparation for logical decoding on standby, which needs to include whether visibility affecting WAL records are about a (user) catalog table. Which is only known for the table, not the indexes. It's also nice to be able to pass the heap relation to GlobalVisTestFor() in vacuumRedirectAndPlaceholder(). Author: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Discussion: https://postgr.es/m/21b700c3-eecf-2e05-a699-f8c78dd31ec7@gmail.com
2023-04-02 05:12:26 +02:00
log_heap_visible(Relation rel, Buffer heap_buffer, Buffer vm_buffer,
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
TransactionId snapshotConflictHorizon, uint8 vmflags)
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
{
xl_heap_visible xlrec;
XLogRecPtr recptr;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
uint8 flags;
Allow I/O reliability checks using 16-bit checksums Checksums are set immediately prior to flush out of shared buffers and checked when pages are read in again. Hint bit setting will require full page write when block is dirtied, which causes various infrastructure changes. Extensive comments, docs and README. WARNING message thrown if checksum fails on non-all zeroes page; ERROR thrown but can be disabled with ignore_checksum_failure = on. Feature enabled by an initdb option, since transition from option off to option on is long and complex and has not yet been implemented. Default is not to use checksums. Checksum used is WAL CRC-32 truncated to 16-bits. Simon Riggs, Jeff Davis, Greg Smith Wide input and assistance from many community members. Thank you.
2013-03-22 14:54:07 +01:00
Assert(BufferIsValid(heap_buffer));
Assert(BufferIsValid(vm_buffer));
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
xlrec.snapshotConflictHorizon = snapshotConflictHorizon;
Change the format of the VM fork to add a second bit per page. The new bit indicates whether every tuple on the page is already frozen. It is cleared only when the all-visible bit is cleared, and it can be set only when we vacuum a page and find that every tuple on that page is both visible to every transaction and in no need of any future vacuuming. A future commit will use this new bit to optimize away full-table scans that would otherwise be triggered by XID wraparound considerations. A page which is merely all-visible must still be scanned in that case, but a page which is all-frozen need not be. This commit does not attempt that optimization, although that optimization is the goal here. It seems better to get the basic infrastructure in place first. Per discussion, it's very desirable for pg_upgrade to automatically migrate existing VM forks from the old format to the new format. That, too, will be handled in a follow-on patch. Masahiko Sawada, reviewed by Kyotaro Horiguchi, Fujii Masao, Amit Kapila, Simon Riggs, Andres Freund, and others, and substantially revised by me.
2016-03-02 03:49:41 +01:00
xlrec.flags = vmflags;
Add info in WAL records in preparation for logical slot conflict handling This commit only implements one prerequisite part for allowing logical decoding. The commit message contains an explanation of the overall design, which later commits will refer back to. Overall design: 1. We want to enable logical decoding on standbys, but replay of WAL from the primary might remove data that is needed by logical decoding, causing error(s) on the standby. To prevent those errors, a new replication conflict scenario needs to be addressed (as much as hot standby does). 2. Our chosen strategy for dealing with this type of replication slot is to invalidate logical slots for which needed data has been removed. 3. To do this we need the latestRemovedXid for each change, just as we do for physical replication conflicts, but we also need to know whether any particular change was to data that logical replication might access. That way, during WAL replay, we know when there is a risk of conflict and, if so, if there is a conflict. 4. We can't rely on the standby's relcache entries for this purpose in any way, because the startup process can't access catalog contents. 5. Therefore every WAL record that potentially removes data from the index or heap must carry a flag indicating whether or not it is one that might be accessed during logical decoding. Why do we need this for logical decoding on standby? First, let's forget about logical decoding on standby and recall that on a primary database, any catalog rows that may be needed by a logical decoding replication slot are not removed. This is done thanks to the catalog_xmin associated with the logical replication slot. But, with logical decoding on standby, in the following cases: - hot_standby_feedback is off - hot_standby_feedback is on but there is no a physical slot between the primary and the standby. Then, hot_standby_feedback will work, but only while the connection is alive (for example a node restart would break it) Then, the primary may delete system catalog rows that could be needed by the logical decoding on the standby (as it does not know about the catalog_xmin on the standby). So, it’s mandatory to identify those rows and invalidate the slots that may need them if any. Identifying those rows is the purpose of this commit. Implementation: When a WAL replay on standby indicates that a catalog table tuple is to be deleted by an xid that is greater than a logical slot's catalog_xmin, then that means the slot's catalog_xmin conflicts with the xid, and we need to handle the conflict. While subsequent commits will do the actual conflict handling, this commit adds a new field isCatalogRel in such WAL records (and a new bit set in the xl_heap_visible flags field), that is true for catalog tables, so as to arrange for conflict handling. The affected WAL records are the ones that already contain the snapshotConflictHorizon field, namely: - gistxlogDelete - gistxlogPageReuse - xl_hash_vacuum_one_page - xl_heap_prune - xl_heap_freeze_page - xl_heap_visible - xl_btree_reuse_page - xl_btree_delete - spgxlogVacuumRedirect Due to this new field being added, xl_hash_vacuum_one_page and gistxlogDelete do now contain the offsets to be deleted as a FLEXIBLE_ARRAY_MEMBER. This is needed to ensure correct alignment. It's not needed on the others struct where isCatalogRel has been added. This commit just introduces the WAL format changes mentioned above. Handling the actual conflicts will follow in future commits. Bumps XLOG_PAGE_MAGIC as the several WAL records are changed. Author: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Author: Andres Freund <andres@anarazel.de> (in an older version) Author: Amit Khandekar <amitdkhan.pg@gmail.com> (in an older version) Reviewed-by: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Reviewed-by: Andres Freund <andres@anarazel.de> Reviewed-by: Robert Haas <robertmhaas@gmail.com> Reviewed-by: Fabrízio de Royes Mello <fabriziomello@gmail.com> Reviewed-by: Melanie Plageman <melanieplageman@gmail.com>
2023-04-02 21:32:19 +02:00
if (RelationIsAccessibleInLogicalDecoding(rel))
xlrec.flags |= VISIBILITYMAP_XLOG_CATALOG_REL;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHeapVisible);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterBuffer(0, vm_buffer, 0);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
flags = REGBUF_STANDARD;
if (!XLogHintBitIsNeeded())
flags |= REGBUF_NO_IMAGE;
XLogRegisterBuffer(1, heap_buffer, flags);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_VISIBLE);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
return recptr;
}
Clean up WAL/buffer interactions as per my recent proposal. Get rid of the misleadingly-named WriteBuffer routine, and instead require routines that change buffer pages to call MarkBufferDirty (which does exactly what it says). We also require that they do so before calling XLogInsert; this takes care of the synchronization requirement documented in SyncOneBuffer. Note that because bufmgr takes the buffer content lock (in shared mode) while writing out any buffer, it doesn't matter whether MarkBufferDirty is executed before the buffer content change is complete, so long as the content change is completed before releasing exclusive lock on the buffer. So it's OK to set the dirtybit before we fill in the LSN. This eliminates the former kluge of needing to set the dirtybit in LockBuffer. Aside from making the code more transparent, we can also add some new debugging assertions, in particular that the caller of MarkBufferDirty must hold the buffer content lock, not merely a pin.
2006-04-01 01:32:07 +02:00
/*
* Perform XLogInsert for a heap-update operation. Caller must already
* have modified the buffer(s) and marked them dirty.
*/
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
static XLogRecPtr
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
log_heap_update(Relation reln, Buffer oldbuf,
Buffer newbuf, HeapTuple oldtup, HeapTuple newtup,
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
HeapTuple old_key_tuple,
Fix a violation of WAL coding rules in the recent patch to include an "all tuples visible" flag in heap page headers. The flag update *must* be applied before calling XLogInsert, but heap_update and the tuple moving routines in VACUUM FULL were ignoring this rule. A crash and replay could therefore leave the flag incorrectly set, causing rows to appear visible in seqscans when they should not be. This might explain recent reports of data corruption from Jeff Ross and others. In passing, do a bit of editorialization on comments in visibilitymap.c.
2009-08-24 04:18:32 +02:00
bool all_visible_cleared, bool new_all_visible_cleared)
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
{
Fix unportable assumptions about alignment of local char[n] variables.
2001-03-26 01:23:59 +02:00
xl_heap_update xlrec;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
xl_heap_header xlhdr;
xl_heap_header xlhdr_idx;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
uint8 info;
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
uint16 prefix_suffix[2];
uint16 prefixlen = 0,
suffixlen = 0;
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
XLogRecPtr recptr;
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
Page page = BufferGetPage(newbuf);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
bool need_tuple_data = RelationIsLogicallyLogged(reln);
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
bool init;
int bufflags;
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
Generalize concept of temporary relations to "relation persistence". This commit replaces pg_class.relistemp with pg_class.relpersistence; and also modifies the RangeVar node type to carry relpersistence rather than istemp. It also removes removes rd_istemp from RelationData and instead performs the correct computation based on relpersistence. For clarity, we add three new macros: RelationNeedsWAL(), RelationUsesLocalBuffers(), and RelationUsesTempNamespace(), so that we can clarify the purpose of each check that previous depended on rd_istemp. This is intended as infrastructure for the upcoming unlogged tables patch, as well as for future possible work on global temporary tables.
2010-12-13 18:34:26 +01:00
/* Caller should not call me on a non-WAL-logged relation */
Assert(RelationNeedsWAL(reln));
Restructure local-buffer handling per recent pghackers discussion. The local buffer manager is no longer used for newly-created relations (unless they are TEMP); a new non-TEMP relation goes through the shared bufmgr and thus will participate normally in checkpoints. But TEMP relations use the local buffer manager throughout their lifespan. Also, operations in TEMP relations are not logged in WAL, thus improving performance. Since it's no longer necessary to fsync relations as they move out of the local buffers into shared buffers, quite a lot of smgr.c/md.c/fd.c code is no longer needed and has been removed: there's no concept of a dirty relation anymore in md.c/fd.c, and we never fsync anything but WAL. Still TODO: improve local buffer management algorithms so that it would be reasonable to increase NLocBuffer.
2002-08-06 04:36:35 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogBeginInsert();
Remove old-style VACUUM FULL (which was known for a little while as VACUUM FULL INPLACE), along with a boatload of subsidiary code and complexity. Per discussion, the use case for this method of vacuuming is no longer large enough to justify maintaining it; not to mention that we don't wish to invest the work that would be needed to make it play nicely with Hot Standby. Aside from the code directly related to old-style VACUUM FULL, this commit removes support for certain WAL record types that could only be generated within VACUUM FULL, redirect-pointer removal in heap_page_prune, and nontransactional generation of cache invalidation sinval messages (the last being the sticking point for Hot Standby). We still have to retain all code that copes with finding HEAP_MOVED_OFF and HEAP_MOVED_IN flag bits on existing tuples. This can't be removed as long as we want to support in-place update from pre-9.0 databases.
2010-02-08 05:33:55 +01:00
if (HeapTupleIsHeapOnly(newtup))
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
info = XLOG_HEAP_HOT_UPDATE;
else
info = XLOG_HEAP_UPDATE;
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
/*
* If the old and new tuple are on the same page, we only need to log the
* parts of the new tuple that were changed. That saves on the amount of
* WAL we need to write. Currently, we just count any unchanged bytes in
* the beginning and end of the tuple. That's quick to check, and
* perfectly covers the common case that only one field is updated.
*
* We could do this even if the old and new tuple are on different pages,
* but only if we don't make a full-page image of the old page, which is
* difficult to know in advance. Also, if the old tuple is corrupt for
* some reason, it would allow the corruption to propagate the new page,
* so it seems best to avoid. Under the general assumption that most
* updates tend to create the new tuple version on the same page, there
* isn't much to be gained by doing this across pages anyway.
*
* Skip this if we're taking a full-page image of the new page, as we
* don't include the new tuple in the WAL record in that case. Also
* disable if wal_level='logical', as logical decoding needs to be able to
* read the new tuple in whole from the WAL record alone.
*/
if (oldbuf == newbuf && !need_tuple_data &&
!XLogCheckBufferNeedsBackup(newbuf))
{
char *oldp = (char *) oldtup->t_data + oldtup->t_data->t_hoff;
char *newp = (char *) newtup->t_data + newtup->t_data->t_hoff;
int oldlen = oldtup->t_len - oldtup->t_data->t_hoff;
int newlen = newtup->t_len - newtup->t_data->t_hoff;
/* Check for common prefix between old and new tuple */
for (prefixlen = 0; prefixlen < Min(oldlen, newlen); prefixlen++)
{
if (newp[prefixlen] != oldp[prefixlen])
break;
}
pgindent run for 9.4 This includes removing tabs after periods in C comments, which was applied to back branches, so this change should not effect backpatching.
2014-05-06 18:12:18 +02:00
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
/*
* Storing the length of the prefix takes 2 bytes, so we need to save
* at least 3 bytes or there's no point.
*/
if (prefixlen < 3)
prefixlen = 0;
/* Same for suffix */
for (suffixlen = 0; suffixlen < Min(oldlen, newlen) - prefixlen; suffixlen++)
{
if (newp[newlen - suffixlen - 1] != oldp[oldlen - suffixlen - 1])
break;
}
if (suffixlen < 3)
suffixlen = 0;
}
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
/* Prepare main WAL data chain */
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
xlrec.flags = 0;
if (all_visible_cleared)
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
xlrec.flags |= XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
if (new_all_visible_cleared)
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
xlrec.flags |= XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED;
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
if (prefixlen > 0)
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
xlrec.flags |= XLH_UPDATE_PREFIX_FROM_OLD;
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
if (suffixlen > 0)
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
xlrec.flags |= XLH_UPDATE_SUFFIX_FROM_OLD;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (need_tuple_data)
{
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
xlrec.flags |= XLH_UPDATE_CONTAINS_NEW_TUPLE;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (old_key_tuple)
{
if (reln->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_TUPLE;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
else
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_KEY;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
}
}
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
/* If new tuple is the single and first tuple on page... */
if (ItemPointerGetOffsetNumber(&(newtup->t_self)) == FirstOffsetNumber &&
PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
{
info |= XLOG_HEAP_INIT_PAGE;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
init = true;
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
}
else
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
init = false;
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
/* Prepare WAL data for the old page */
xlrec.old_offnum = ItemPointerGetOffsetNumber(&oldtup->t_self);
xlrec.old_xmax = HeapTupleHeaderGetRawXmax(oldtup->t_data);
xlrec.old_infobits_set = compute_infobits(oldtup->t_data->t_infomask,
oldtup->t_data->t_infomask2);
/* Prepare WAL data for the new page */
xlrec.new_offnum = ItemPointerGetOffsetNumber(&newtup->t_self);
xlrec.new_xmax = HeapTupleHeaderGetRawXmax(newtup->t_data);
bufflags = REGBUF_STANDARD;
if (init)
bufflags |= REGBUF_WILL_INIT;
if (need_tuple_data)
bufflags |= REGBUF_KEEP_DATA;
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterBuffer(0, newbuf, bufflags);
if (oldbuf != newbuf)
XLogRegisterBuffer(1, oldbuf, REGBUF_STANDARD);
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterData((char *) &xlrec, SizeOfHeapUpdate);
/*
* Prepare WAL data for the new tuple.
*/
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
if (prefixlen > 0 || suffixlen > 0)
{
if (prefixlen > 0 && suffixlen > 0)
{
prefix_suffix[0] = prefixlen;
prefix_suffix[1] = suffixlen;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterBufData(0, (char *) &prefix_suffix, sizeof(uint16) * 2);
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
}
else if (prefixlen > 0)
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterBufData(0, (char *) &prefixlen, sizeof(uint16));
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
}
else
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterBufData(0, (char *) &suffixlen, sizeof(uint16));
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
}
}
pgindent run.
2003-08-04 02:43:34 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
xlhdr.t_infomask2 = newtup->t_data->t_infomask2;
xlhdr.t_infomask = newtup->t_data->t_infomask;
xlhdr.t_hoff = newtup->t_data->t_hoff;
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
Assert(SizeofHeapTupleHeader + prefixlen + suffixlen <= newtup->t_len);
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
/*
* PG73FORMAT: write bitmap [+ padding] [+ oid] + data
*
* The 'data' doesn't include the common prefix or suffix.
*/
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterBufData(0, (char *) &xlhdr, SizeOfHeapHeader);
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
if (prefixlen == 0)
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterBufData(0,
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
((char *) newtup->t_data) + SizeofHeapTupleHeader,
newtup->t_len - SizeofHeapTupleHeader - suffixlen);
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
}
else
{
/*
* Have to write the null bitmap and data after the common prefix as
* two separate rdata entries.
*/
/* bitmap [+ padding] [+ oid] */
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
if (newtup->t_data->t_hoff - SizeofHeapTupleHeader > 0)
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterBufData(0,
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
((char *) newtup->t_data) + SizeofHeapTupleHeader,
newtup->t_data->t_hoff - SizeofHeapTupleHeader);
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
}
/* data after common prefix */
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterBufData(0,
((char *) newtup->t_data) + newtup->t_data->t_hoff + prefixlen,
newtup->t_len - newtup->t_data->t_hoff - prefixlen - suffixlen);
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
}
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
/* We need to log a tuple identity */
if (need_tuple_data && old_key_tuple)
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
/* don't really need this, but its more comfy to decode */
xlhdr_idx.t_infomask2 = old_key_tuple->t_data->t_infomask2;
xlhdr_idx.t_infomask = old_key_tuple->t_data->t_infomask;
xlhdr_idx.t_hoff = old_key_tuple->t_data->t_hoff;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRegisterData((char *) &xlhdr_idx, SizeOfHeapHeader);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
/* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
XLogRegisterData((char *) old_key_tuple->t_data + SizeofHeapTupleHeader,
old_key_tuple->t_len - SizeofHeapTupleHeader);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
}
Introduce replication progress tracking infrastructure. When implementing a replication solution ontop of logical decoding, two related problems exist: * How to safely keep track of replication progress * How to change replication behavior, based on the origin of a row; e.g. to avoid loops in bi-directional replication setups The solution to these problems, as implemented here, consist out of three parts: 1) 'replication origins', which identify nodes in a replication setup. 2) 'replication progress tracking', which remembers, for each replication origin, how far replay has progressed in a efficient and crash safe manner. 3) The ability to filter out changes performed on the behest of a replication origin during logical decoding; this allows complex replication topologies. E.g. by filtering all replayed changes out. Most of this could also be implemented in "userspace", e.g. by inserting additional rows contain origin information, but that ends up being much less efficient and more complicated. We don't want to require various replication solutions to reimplement logic for this independently. The infrastructure is intended to be generic enough to be reusable. This infrastructure also replaces the 'nodeid' infrastructure of commit timestamps. It is intended to provide all the former capabilities, except that there's only 2^16 different origins; but now they integrate with logical decoding. Additionally more functionality is accessible via SQL. Since the commit timestamp infrastructure has also been introduced in 9.5 (commit 73c986add) changing the API is not a problem. For now the number of origins for which the replication progress can be tracked simultaneously is determined by the max_replication_slots GUC. That GUC is not a perfect match to configure this, but there doesn't seem to be sufficient reason to introduce a separate new one. Bumps both catversion and wal page magic. Author: Andres Freund, with contributions from Petr Jelinek and Craig Ringer Reviewed-By: Heikki Linnakangas, Petr Jelinek, Robert Haas, Steve Singer Discussion: 20150216002155.GI15326@awork2.anarazel.de, 20140923182422.GA15776@alap3.anarazel.de, 20131114172632.GE7522@alap2.anarazel.de
2015-04-29 19:30:53 +02:00
/* filtering by origin on a row level is much more efficient */
Skip checkpoints, archiving on idle systems. Some background activity (like checkpoints, archive timeout, standby snapshots) is not supposed to happen on an idle system. Unfortunately so far it was not easy to determine when a system is idle, which defeated some of the attempts to avoid redundant activity on an idle system. To make that easier, allow to make individual WAL insertions as not being "important". By checking whether any important activity happened since the last time an activity was performed, it now is easy to check whether some action needs to be repeated. Use the new facility for checkpoints, archive timeout and standby snapshots. The lack of a facility causes some issues in older releases, but in my opinion the consequences (superflous checkpoints / archived segments) aren't grave enough to warrant backpatching. Author: Michael Paquier, editorialized by Andres Freund Reviewed-By: Andres Freund, David Steele, Amit Kapila, Kyotaro HORIGUCHI Bug: #13685 Discussion: https://www.postgresql.org/message-id/20151016203031.3019.72930@wrigleys.postgresql.org https://www.postgresql.org/message-id/CAB7nPqQcPqxEM3S735Bd2RzApNqSNJVietAC=6kfkYv_45dKwA@mail.gmail.com Backpatch: -
2016-12-22 20:31:50 +01:00
XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
Introduce replication progress tracking infrastructure. When implementing a replication solution ontop of logical decoding, two related problems exist: * How to safely keep track of replication progress * How to change replication behavior, based on the origin of a row; e.g. to avoid loops in bi-directional replication setups The solution to these problems, as implemented here, consist out of three parts: 1) 'replication origins', which identify nodes in a replication setup. 2) 'replication progress tracking', which remembers, for each replication origin, how far replay has progressed in a efficient and crash safe manner. 3) The ability to filter out changes performed on the behest of a replication origin during logical decoding; this allows complex replication topologies. E.g. by filtering all replayed changes out. Most of this could also be implemented in "userspace", e.g. by inserting additional rows contain origin information, but that ends up being much less efficient and more complicated. We don't want to require various replication solutions to reimplement logic for this independently. The infrastructure is intended to be generic enough to be reusable. This infrastructure also replaces the 'nodeid' infrastructure of commit timestamps. It is intended to provide all the former capabilities, except that there's only 2^16 different origins; but now they integrate with logical decoding. Additionally more functionality is accessible via SQL. Since the commit timestamp infrastructure has also been introduced in 9.5 (commit 73c986add) changing the API is not a problem. For now the number of origins for which the replication progress can be tracked simultaneously is determined by the max_replication_slots GUC. That GUC is not a perfect match to configure this, but there doesn't seem to be sufficient reason to introduce a separate new one. Bumps both catversion and wal page magic. Author: Andres Freund, with contributions from Petr Jelinek and Craig Ringer Reviewed-By: Heikki Linnakangas, Petr Jelinek, Robert Haas, Steve Singer Discussion: 20150216002155.GI15326@awork2.anarazel.de, 20140923182422.GA15776@alap3.anarazel.de, 20131114172632.GE7522@alap2.anarazel.de
2015-04-29 19:30:53 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
recptr = XLogInsert(RM_HEAP_ID, info);
WAL misc
2000-10-24 11:56:23 +02:00
Cosmetic code cleanup: fix a bunch of places that used "return (expr);" rather than "return expr;" -- the latter style is used in most of the tree. I kept the parentheses when they were necessary or useful because the return expression was complex.
2006-01-11 09:43:13 +01:00
return recptr;
WAL misc
2000-10-24 11:56:23 +02:00
}
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
/*
Collection of typo fixes. Use "a" and "an" correctly, mostly in comments. Two error messages were also fixed (they were just elogs, so no translation work required). Two function comments in pg_proc.h were also fixed. Etsuro Fujita reported one of these, but I found a lot more with grep. Also fix a few other typos spotted while grepping for the a/an typos. For example, "consists out of ..." -> "consists of ...". Plus a "though"/ "through" mixup reported by Euler Taveira. Many of these typos were in old code, which would be nice to backpatch to make future backpatching easier. But much of the code was new, and I didn't feel like crafting separate patches for each branch. So no backpatching.
2015-05-20 15:18:11 +02:00
* Perform XLogInsert of an XLOG_HEAP2_NEW_CID record
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
*
* This is only used in wal_level >= WAL_LEVEL_LOGICAL, and only for catalog
* tuples.
*/
static XLogRecPtr
log_heap_new_cid(Relation relation, HeapTuple tup)
{
xl_heap_new_cid xlrec;
XLogRecPtr recptr;
HeapTupleHeader hdr = tup->t_data;
Assert(ItemPointerIsValid(&tup->t_self));
Assert(tup->t_tableOid != InvalidOid);
xlrec.top_xid = GetTopTransactionId();
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
xlrec.target_locator = relation->rd_locator;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
xlrec.target_tid = tup->t_self;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
/*
* If the tuple got inserted & deleted in the same TX we definitely have a
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
* combo CID, set cmin and cmax.
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
*/
if (hdr->t_infomask & HEAP_COMBOCID)
{
Assert(!(hdr->t_infomask & HEAP_XMAX_INVALID));
Change the way we mark tuples as frozen. Instead of changing the tuple xmin to FrozenTransactionId, the combination of HEAP_XMIN_COMMITTED and HEAP_XMIN_INVALID, which were previously never set together, is now defined as HEAP_XMIN_FROZEN. A variety of previous proposals to freeze tuples opportunistically before vacuum_freeze_min_age is reached have foundered on the objection that replacing xmin by FrozenTransactionId might hinder debugging efforts when things in this area go awry; this patch is intended to solve that problem by keeping the XID around (but largely ignoring the value to which it is set). Third-party code that checks for HEAP_XMIN_INVALID on tuples where HEAP_XMIN_COMMITTED might be set will be broken by this change. To fix, use the new accessor macros in htup_details.h rather than consulting the bits directly. HeapTupleHeaderGetXmin has been modified to return FrozenTransactionId when the infomask bits indicate that the tuple is frozen; use HeapTupleHeaderGetRawXmin when you already know that the tuple isn't marked commited or frozen, or want the raw value anyway. We currently do this in routines that display the xmin for user consumption, in tqual.c where it's known to be safe and important for the avoidance of extra cycles, and in the function-caching code for various procedural languages, which shouldn't invalidate the cache just because the tuple gets frozen. Robert Haas and Andres Freund
2013-12-22 21:49:09 +01:00
Assert(!HeapTupleHeaderXminInvalid(hdr));
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
xlrec.cmin = HeapTupleHeaderGetCmin(hdr);
xlrec.cmax = HeapTupleHeaderGetCmax(hdr);
xlrec.combocid = HeapTupleHeaderGetRawCommandId(hdr);
}
Sanitize the term "combo CID" in code comments Combo CIDs were referred in the code comments using different terms across various places of the code, so unify a bit the term used with what is currently in use in some of the READMEs. Author: "Hou, Zhijie" Discussion: https://postgr.es/m/1d42865c91404f46af4562532fdbea31@G08CNEXMBPEKD05.g08.fujitsu.local
2021-03-25 08:08:03 +01:00
/* No combo CID, so only cmin or cmax can be set by this TX */
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
else
{
/*
* Tuple inserted.
*
* We need to check for LOCK ONLY because multixacts might be
* transferred to the new tuple in case of FOR KEY SHARE updates in
Collection of typo fixes. Use "a" and "an" correctly, mostly in comments. Two error messages were also fixed (they were just elogs, so no translation work required). Two function comments in pg_proc.h were also fixed. Etsuro Fujita reported one of these, but I found a lot more with grep. Also fix a few other typos spotted while grepping for the a/an typos. For example, "consists out of ..." -> "consists of ...". Plus a "though"/ "through" mixup reported by Euler Taveira. Many of these typos were in old code, which would be nice to backpatch to make future backpatching easier. But much of the code was new, and I didn't feel like crafting separate patches for each branch. So no backpatching.
2015-05-20 15:18:11 +02:00
* which case there will be an xmax, although the tuple just got
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
* inserted.
*/
if (hdr->t_infomask & HEAP_XMAX_INVALID ||
HEAP_XMAX_IS_LOCKED_ONLY(hdr->t_infomask))
{
xlrec.cmin = HeapTupleHeaderGetRawCommandId(hdr);
xlrec.cmax = InvalidCommandId;
}
/* Tuple from a different tx updated or deleted. */
else
{
xlrec.cmin = InvalidCommandId;
xlrec.cmax = HeapTupleHeaderGetRawCommandId(hdr);
}
xlrec.combocid = InvalidCommandId;
}
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
/*
* Note that we don't need to register the buffer here, because this
* operation does not modify the page. The insert/update/delete that
* called us certainly did, but that's WAL-logged separately.
*/
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHeapNewCid);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
Introduce replication progress tracking infrastructure. When implementing a replication solution ontop of logical decoding, two related problems exist: * How to safely keep track of replication progress * How to change replication behavior, based on the origin of a row; e.g. to avoid loops in bi-directional replication setups The solution to these problems, as implemented here, consist out of three parts: 1) 'replication origins', which identify nodes in a replication setup. 2) 'replication progress tracking', which remembers, for each replication origin, how far replay has progressed in a efficient and crash safe manner. 3) The ability to filter out changes performed on the behest of a replication origin during logical decoding; this allows complex replication topologies. E.g. by filtering all replayed changes out. Most of this could also be implemented in "userspace", e.g. by inserting additional rows contain origin information, but that ends up being much less efficient and more complicated. We don't want to require various replication solutions to reimplement logic for this independently. The infrastructure is intended to be generic enough to be reusable. This infrastructure also replaces the 'nodeid' infrastructure of commit timestamps. It is intended to provide all the former capabilities, except that there's only 2^16 different origins; but now they integrate with logical decoding. Additionally more functionality is accessible via SQL. Since the commit timestamp infrastructure has also been introduced in 9.5 (commit 73c986add) changing the API is not a problem. For now the number of origins for which the replication progress can be tracked simultaneously is determined by the max_replication_slots GUC. That GUC is not a perfect match to configure this, but there doesn't seem to be sufficient reason to introduce a separate new one. Bumps both catversion and wal page magic. Author: Andres Freund, with contributions from Petr Jelinek and Craig Ringer Reviewed-By: Heikki Linnakangas, Petr Jelinek, Robert Haas, Steve Singer Discussion: 20150216002155.GI15326@awork2.anarazel.de, 20140923182422.GA15776@alap3.anarazel.de, 20131114172632.GE7522@alap2.anarazel.de
2015-04-29 19:30:53 +02:00
/* will be looked at irrespective of origin */
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_NEW_CID);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
return recptr;
}
/*
* Build a heap tuple representing the configured REPLICA IDENTITY to represent
Fix various typos, grammar and code style in comments and docs This fixes a set of issues that have accumulated over the past months (or years) in various code areas. Most fixes are related to some recent additions, as of the development of v15. Author: Justin Pryzby Discussion: https://postgr.es/m/20220124030001.GQ23027@telsasoft.com
2022-01-25 01:40:04 +01:00
* the old tuple in an UPDATE or DELETE.
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
*
Collection of typo fixes. Use "a" and "an" correctly, mostly in comments. Two error messages were also fixed (they were just elogs, so no translation work required). Two function comments in pg_proc.h were also fixed. Etsuro Fujita reported one of these, but I found a lot more with grep. Also fix a few other typos spotted while grepping for the a/an typos. For example, "consists out of ..." -> "consists of ...". Plus a "though"/ "through" mixup reported by Euler Taveira. Many of these typos were in old code, which would be nice to backpatch to make future backpatching easier. But much of the code was new, and I didn't feel like crafting separate patches for each branch. So no backpatching.
2015-05-20 15:18:11 +02:00
* Returns NULL if there's no need to log an identity or if there's no suitable
Avoid touching replica identity index in ExtractReplicaIdentity(). In what seems like a fit of misplaced optimization, ExtractReplicaIdentity() accessed the relation's replica-identity index without taking any lock on it. Usually, the surrounding query already holds some lock so this is safe enough ... but in the case of a previously-planned delete, there might be no existing lock. Given a suitable test case, this is exposed in v12 and HEAD by an assertion added by commit b04aeb0a0. The whole thing's rather poorly thought out anyway; rather than looking directly at the index, we should use the index-attributes bitmap that's held by the parent table's relcache entry, as the caller functions do. This is more consistent and likely a bit faster, since it avoids a cache lookup. Hence, change to doing it that way. While at it, rather than blithely assuming that the identity columns are non-null (with catastrophic results if that's wrong), add assertion checks that they aren't null. Possibly those should be actual test-and-elog, but I'll leave it like this for now. In principle, this is a bug that's been there since this code was introduced (in 9.4). In practice, the risk seems quite low, since we do have a lock on the index's parent table, so concurrent changes to the index's catalog entries seem unlikely. Given the precedent that commit 9c703c169 wasn't back-patched, I won't risk back-patching this further than v12. Per report from Hadi Moshayedi. Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
2019-09-02 22:10:37 +02:00
* key defined.
*
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
* Pass key_required true if any replica identity columns changed value, or if
* any of them have any external data. Delete must always pass true.
Avoid touching replica identity index in ExtractReplicaIdentity(). In what seems like a fit of misplaced optimization, ExtractReplicaIdentity() accessed the relation's replica-identity index without taking any lock on it. Usually, the surrounding query already holds some lock so this is safe enough ... but in the case of a previously-planned delete, there might be no existing lock. Given a suitable test case, this is exposed in v12 and HEAD by an assertion added by commit b04aeb0a0. The whole thing's rather poorly thought out anyway; rather than looking directly at the index, we should use the index-attributes bitmap that's held by the parent table's relcache entry, as the caller functions do. This is more consistent and likely a bit faster, since it avoids a cache lookup. Hence, change to doing it that way. While at it, rather than blithely assuming that the identity columns are non-null (with catastrophic results if that's wrong), add assertion checks that they aren't null. Possibly those should be actual test-and-elog, but I'll leave it like this for now. In principle, this is a bug that's been there since this code was introduced (in 9.4). In practice, the risk seems quite low, since we do have a lock on the index's parent table, so concurrent changes to the index's catalog entries seem unlikely. Given the precedent that commit 9c703c169 wasn't back-patched, I won't risk back-patching this further than v12. Per report from Hadi Moshayedi. Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
2019-09-02 22:10:37 +02:00
*
* *copy is set to true if the returned tuple is a modified copy rather than
* the same tuple that was passed in.
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
*/
static HeapTuple
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
Avoid touching replica identity index in ExtractReplicaIdentity(). In what seems like a fit of misplaced optimization, ExtractReplicaIdentity() accessed the relation's replica-identity index without taking any lock on it. Usually, the surrounding query already holds some lock so this is safe enough ... but in the case of a previously-planned delete, there might be no existing lock. Given a suitable test case, this is exposed in v12 and HEAD by an assertion added by commit b04aeb0a0. The whole thing's rather poorly thought out anyway; rather than looking directly at the index, we should use the index-attributes bitmap that's held by the parent table's relcache entry, as the caller functions do. This is more consistent and likely a bit faster, since it avoids a cache lookup. Hence, change to doing it that way. While at it, rather than blithely assuming that the identity columns are non-null (with catastrophic results if that's wrong), add assertion checks that they aren't null. Possibly those should be actual test-and-elog, but I'll leave it like this for now. In principle, this is a bug that's been there since this code was introduced (in 9.4). In practice, the risk seems quite low, since we do have a lock on the index's parent table, so concurrent changes to the index's catalog entries seem unlikely. Given the precedent that commit 9c703c169 wasn't back-patched, I won't risk back-patching this further than v12. Per report from Hadi Moshayedi. Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
2019-09-02 22:10:37 +02:00
bool *copy)
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
{
TupleDesc desc = RelationGetDescr(relation);
char replident = relation->rd_rel->relreplident;
Avoid touching replica identity index in ExtractReplicaIdentity(). In what seems like a fit of misplaced optimization, ExtractReplicaIdentity() accessed the relation's replica-identity index without taking any lock on it. Usually, the surrounding query already holds some lock so this is safe enough ... but in the case of a previously-planned delete, there might be no existing lock. Given a suitable test case, this is exposed in v12 and HEAD by an assertion added by commit b04aeb0a0. The whole thing's rather poorly thought out anyway; rather than looking directly at the index, we should use the index-attributes bitmap that's held by the parent table's relcache entry, as the caller functions do. This is more consistent and likely a bit faster, since it avoids a cache lookup. Hence, change to doing it that way. While at it, rather than blithely assuming that the identity columns are non-null (with catastrophic results if that's wrong), add assertion checks that they aren't null. Possibly those should be actual test-and-elog, but I'll leave it like this for now. In principle, this is a bug that's been there since this code was introduced (in 9.4). In practice, the risk seems quite low, since we do have a lock on the index's parent table, so concurrent changes to the index's catalog entries seem unlikely. Given the precedent that commit 9c703c169 wasn't back-patched, I won't risk back-patching this further than v12. Per report from Hadi Moshayedi. Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
2019-09-02 22:10:37 +02:00
Bitmapset *idattrs;
HeapTuple key_tuple;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
bool nulls[MaxHeapAttributeNumber];
Datum values[MaxHeapAttributeNumber];
*copy = false;
if (!RelationIsLogicallyLogged(relation))
return NULL;
if (replident == REPLICA_IDENTITY_NOTHING)
return NULL;
if (replident == REPLICA_IDENTITY_FULL)
{
/*
* When logging the entire old tuple, it very well could contain
* toasted columns. If so, force them to be inlined.
*/
if (HeapTupleHasExternal(tp))
{
*copy = true;
Avoid touching replica identity index in ExtractReplicaIdentity(). In what seems like a fit of misplaced optimization, ExtractReplicaIdentity() accessed the relation's replica-identity index without taking any lock on it. Usually, the surrounding query already holds some lock so this is safe enough ... but in the case of a previously-planned delete, there might be no existing lock. Given a suitable test case, this is exposed in v12 and HEAD by an assertion added by commit b04aeb0a0. The whole thing's rather poorly thought out anyway; rather than looking directly at the index, we should use the index-attributes bitmap that's held by the parent table's relcache entry, as the caller functions do. This is more consistent and likely a bit faster, since it avoids a cache lookup. Hence, change to doing it that way. While at it, rather than blithely assuming that the identity columns are non-null (with catastrophic results if that's wrong), add assertion checks that they aren't null. Possibly those should be actual test-and-elog, but I'll leave it like this for now. In principle, this is a bug that's been there since this code was introduced (in 9.4). In practice, the risk seems quite low, since we do have a lock on the index's parent table, so concurrent changes to the index's catalog entries seem unlikely. Given the precedent that commit 9c703c169 wasn't back-patched, I won't risk back-patching this further than v12. Per report from Hadi Moshayedi. Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
2019-09-02 22:10:37 +02:00
tp = toast_flatten_tuple(tp, desc);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
}
return tp;
}
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
/* if the key isn't required and we're only logging the key, we're done */
if (!key_required)
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
return NULL;
Avoid touching replica identity index in ExtractReplicaIdentity(). In what seems like a fit of misplaced optimization, ExtractReplicaIdentity() accessed the relation's replica-identity index without taking any lock on it. Usually, the surrounding query already holds some lock so this is safe enough ... but in the case of a previously-planned delete, there might be no existing lock. Given a suitable test case, this is exposed in v12 and HEAD by an assertion added by commit b04aeb0a0. The whole thing's rather poorly thought out anyway; rather than looking directly at the index, we should use the index-attributes bitmap that's held by the parent table's relcache entry, as the caller functions do. This is more consistent and likely a bit faster, since it avoids a cache lookup. Hence, change to doing it that way. While at it, rather than blithely assuming that the identity columns are non-null (with catastrophic results if that's wrong), add assertion checks that they aren't null. Possibly those should be actual test-and-elog, but I'll leave it like this for now. In principle, this is a bug that's been there since this code was introduced (in 9.4). In practice, the risk seems quite low, since we do have a lock on the index's parent table, so concurrent changes to the index's catalog entries seem unlikely. Given the precedent that commit 9c703c169 wasn't back-patched, I won't risk back-patching this further than v12. Per report from Hadi Moshayedi. Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
2019-09-02 22:10:37 +02:00
/* find out the replica identity columns */
idattrs = RelationGetIndexAttrBitmap(relation,
INDEX_ATTR_BITMAP_IDENTITY_KEY);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
Avoid touching replica identity index in ExtractReplicaIdentity(). In what seems like a fit of misplaced optimization, ExtractReplicaIdentity() accessed the relation's replica-identity index without taking any lock on it. Usually, the surrounding query already holds some lock so this is safe enough ... but in the case of a previously-planned delete, there might be no existing lock. Given a suitable test case, this is exposed in v12 and HEAD by an assertion added by commit b04aeb0a0. The whole thing's rather poorly thought out anyway; rather than looking directly at the index, we should use the index-attributes bitmap that's held by the parent table's relcache entry, as the caller functions do. This is more consistent and likely a bit faster, since it avoids a cache lookup. Hence, change to doing it that way. While at it, rather than blithely assuming that the identity columns are non-null (with catastrophic results if that's wrong), add assertion checks that they aren't null. Possibly those should be actual test-and-elog, but I'll leave it like this for now. In principle, this is a bug that's been there since this code was introduced (in 9.4). In practice, the risk seems quite low, since we do have a lock on the index's parent table, so concurrent changes to the index's catalog entries seem unlikely. Given the precedent that commit 9c703c169 wasn't back-patched, I won't risk back-patching this further than v12. Per report from Hadi Moshayedi. Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
2019-09-02 22:10:37 +02:00
/*
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
* If there's no defined replica identity columns, treat as !key_required.
Avoid touching replica identity index in ExtractReplicaIdentity(). In what seems like a fit of misplaced optimization, ExtractReplicaIdentity() accessed the relation's replica-identity index without taking any lock on it. Usually, the surrounding query already holds some lock so this is safe enough ... but in the case of a previously-planned delete, there might be no existing lock. Given a suitable test case, this is exposed in v12 and HEAD by an assertion added by commit b04aeb0a0. The whole thing's rather poorly thought out anyway; rather than looking directly at the index, we should use the index-attributes bitmap that's held by the parent table's relcache entry, as the caller functions do. This is more consistent and likely a bit faster, since it avoids a cache lookup. Hence, change to doing it that way. While at it, rather than blithely assuming that the identity columns are non-null (with catastrophic results if that's wrong), add assertion checks that they aren't null. Possibly those should be actual test-and-elog, but I'll leave it like this for now. In principle, this is a bug that's been there since this code was introduced (in 9.4). In practice, the risk seems quite low, since we do have a lock on the index's parent table, so concurrent changes to the index's catalog entries seem unlikely. Given the precedent that commit 9c703c169 wasn't back-patched, I won't risk back-patching this further than v12. Per report from Hadi Moshayedi. Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
2019-09-02 22:10:37 +02:00
* (This case should not be reachable from heap_update, since that should
WAL log unchanged toasted replica identity key attributes. Currently, during UPDATE, the unchanged replica identity key attributes are not logged separately because they are getting logged as part of the new tuple. But if they are stored externally then the untoasted values are not getting logged as part of the new tuple and logical replication won't be able to replicate such UPDATEs. So we need to log such attributes as part of the old_key_tuple during UPDATE. Reported-by: Haiying Tang Author: Dilip Kumar and Amit Kapila Reviewed-by: Alvaro Herrera, Haiying Tang, Andres Freund Backpatch-through: 10 Discussion: https://postgr.es/m/OS0PR01MB611342D0A92D4F4BF26C0F47FB229@OS0PR01MB6113.jpnprd01.prod.outlook.com
2022-02-14 04:25:58 +01:00
* calculate key_required accurately. But heap_delete just passes
* constant true for key_required, so we can hit this case in deletes.)
Avoid touching replica identity index in ExtractReplicaIdentity(). In what seems like a fit of misplaced optimization, ExtractReplicaIdentity() accessed the relation's replica-identity index without taking any lock on it. Usually, the surrounding query already holds some lock so this is safe enough ... but in the case of a previously-planned delete, there might be no existing lock. Given a suitable test case, this is exposed in v12 and HEAD by an assertion added by commit b04aeb0a0. The whole thing's rather poorly thought out anyway; rather than looking directly at the index, we should use the index-attributes bitmap that's held by the parent table's relcache entry, as the caller functions do. This is more consistent and likely a bit faster, since it avoids a cache lookup. Hence, change to doing it that way. While at it, rather than blithely assuming that the identity columns are non-null (with catastrophic results if that's wrong), add assertion checks that they aren't null. Possibly those should be actual test-and-elog, but I'll leave it like this for now. In principle, this is a bug that's been there since this code was introduced (in 9.4). In practice, the risk seems quite low, since we do have a lock on the index's parent table, so concurrent changes to the index's catalog entries seem unlikely. Given the precedent that commit 9c703c169 wasn't back-patched, I won't risk back-patching this further than v12. Per report from Hadi Moshayedi. Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
2019-09-02 22:10:37 +02:00
*/
if (bms_is_empty(idattrs))
return NULL;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
/*
Avoid touching replica identity index in ExtractReplicaIdentity(). In what seems like a fit of misplaced optimization, ExtractReplicaIdentity() accessed the relation's replica-identity index without taking any lock on it. Usually, the surrounding query already holds some lock so this is safe enough ... but in the case of a previously-planned delete, there might be no existing lock. Given a suitable test case, this is exposed in v12 and HEAD by an assertion added by commit b04aeb0a0. The whole thing's rather poorly thought out anyway; rather than looking directly at the index, we should use the index-attributes bitmap that's held by the parent table's relcache entry, as the caller functions do. This is more consistent and likely a bit faster, since it avoids a cache lookup. Hence, change to doing it that way. While at it, rather than blithely assuming that the identity columns are non-null (with catastrophic results if that's wrong), add assertion checks that they aren't null. Possibly those should be actual test-and-elog, but I'll leave it like this for now. In principle, this is a bug that's been there since this code was introduced (in 9.4). In practice, the risk seems quite low, since we do have a lock on the index's parent table, so concurrent changes to the index's catalog entries seem unlikely. Given the precedent that commit 9c703c169 wasn't back-patched, I won't risk back-patching this further than v12. Per report from Hadi Moshayedi. Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
2019-09-02 22:10:37 +02:00
* Construct a new tuple containing only the replica identity columns,
* with nulls elsewhere. While we're at it, assert that the replica
* identity columns aren't null.
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
*/
Avoid touching replica identity index in ExtractReplicaIdentity(). In what seems like a fit of misplaced optimization, ExtractReplicaIdentity() accessed the relation's replica-identity index without taking any lock on it. Usually, the surrounding query already holds some lock so this is safe enough ... but in the case of a previously-planned delete, there might be no existing lock. Given a suitable test case, this is exposed in v12 and HEAD by an assertion added by commit b04aeb0a0. The whole thing's rather poorly thought out anyway; rather than looking directly at the index, we should use the index-attributes bitmap that's held by the parent table's relcache entry, as the caller functions do. This is more consistent and likely a bit faster, since it avoids a cache lookup. Hence, change to doing it that way. While at it, rather than blithely assuming that the identity columns are non-null (with catastrophic results if that's wrong), add assertion checks that they aren't null. Possibly those should be actual test-and-elog, but I'll leave it like this for now. In principle, this is a bug that's been there since this code was introduced (in 9.4). In practice, the risk seems quite low, since we do have a lock on the index's parent table, so concurrent changes to the index's catalog entries seem unlikely. Given the precedent that commit 9c703c169 wasn't back-patched, I won't risk back-patching this further than v12. Per report from Hadi Moshayedi. Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
2019-09-02 22:10:37 +02:00
heap_deform_tuple(tp, desc, values, nulls);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
Avoid touching replica identity index in ExtractReplicaIdentity(). In what seems like a fit of misplaced optimization, ExtractReplicaIdentity() accessed the relation's replica-identity index without taking any lock on it. Usually, the surrounding query already holds some lock so this is safe enough ... but in the case of a previously-planned delete, there might be no existing lock. Given a suitable test case, this is exposed in v12 and HEAD by an assertion added by commit b04aeb0a0. The whole thing's rather poorly thought out anyway; rather than looking directly at the index, we should use the index-attributes bitmap that's held by the parent table's relcache entry, as the caller functions do. This is more consistent and likely a bit faster, since it avoids a cache lookup. Hence, change to doing it that way. While at it, rather than blithely assuming that the identity columns are non-null (with catastrophic results if that's wrong), add assertion checks that they aren't null. Possibly those should be actual test-and-elog, but I'll leave it like this for now. In principle, this is a bug that's been there since this code was introduced (in 9.4). In practice, the risk seems quite low, since we do have a lock on the index's parent table, so concurrent changes to the index's catalog entries seem unlikely. Given the precedent that commit 9c703c169 wasn't back-patched, I won't risk back-patching this further than v12. Per report from Hadi Moshayedi. Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
2019-09-02 22:10:37 +02:00
for (int i = 0; i < desc->natts; i++)
{
if (bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
idattrs))
Assert(!nulls[i]);
else
nulls[i] = true;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
}
key_tuple = heap_form_tuple(desc, values, nulls);
*copy = true;
Avoid touching replica identity index in ExtractReplicaIdentity(). In what seems like a fit of misplaced optimization, ExtractReplicaIdentity() accessed the relation's replica-identity index without taking any lock on it. Usually, the surrounding query already holds some lock so this is safe enough ... but in the case of a previously-planned delete, there might be no existing lock. Given a suitable test case, this is exposed in v12 and HEAD by an assertion added by commit b04aeb0a0. The whole thing's rather poorly thought out anyway; rather than looking directly at the index, we should use the index-attributes bitmap that's held by the parent table's relcache entry, as the caller functions do. This is more consistent and likely a bit faster, since it avoids a cache lookup. Hence, change to doing it that way. While at it, rather than blithely assuming that the identity columns are non-null (with catastrophic results if that's wrong), add assertion checks that they aren't null. Possibly those should be actual test-and-elog, but I'll leave it like this for now. In principle, this is a bug that's been there since this code was introduced (in 9.4). In practice, the risk seems quite low, since we do have a lock on the index's parent table, so concurrent changes to the index's catalog entries seem unlikely. Given the precedent that commit 9c703c169 wasn't back-patched, I won't risk back-patching this further than v12. Per report from Hadi Moshayedi. Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
2019-09-02 22:10:37 +02:00
bms_free(idattrs);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
/*
* If the tuple, which by here only contains indexed columns, still has
* toasted columns, force them to be inlined. This is somewhat unlikely
* since there's limits on the size of indexed columns, so we don't
* duplicate toast_flatten_tuple()s functionality in the above loop over
* the indexed columns, even if it would be more efficient.
*/
if (HeapTupleHasExternal(key_tuple))
{
HeapTuple oldtup = key_tuple;
pgindent run for 9.4 This includes removing tabs after periods in C comments, which was applied to back branches, so this change should not effect backpatching.
2014-05-06 18:12:18 +02:00
Avoid touching replica identity index in ExtractReplicaIdentity(). In what seems like a fit of misplaced optimization, ExtractReplicaIdentity() accessed the relation's replica-identity index without taking any lock on it. Usually, the surrounding query already holds some lock so this is safe enough ... but in the case of a previously-planned delete, there might be no existing lock. Given a suitable test case, this is exposed in v12 and HEAD by an assertion added by commit b04aeb0a0. The whole thing's rather poorly thought out anyway; rather than looking directly at the index, we should use the index-attributes bitmap that's held by the parent table's relcache entry, as the caller functions do. This is more consistent and likely a bit faster, since it avoids a cache lookup. Hence, change to doing it that way. While at it, rather than blithely assuming that the identity columns are non-null (with catastrophic results if that's wrong), add assertion checks that they aren't null. Possibly those should be actual test-and-elog, but I'll leave it like this for now. In principle, this is a bug that's been there since this code was introduced (in 9.4). In practice, the risk seems quite low, since we do have a lock on the index's parent table, so concurrent changes to the index's catalog entries seem unlikely. Given the precedent that commit 9c703c169 wasn't back-patched, I won't risk back-patching this further than v12. Per report from Hadi Moshayedi. Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
2019-09-02 22:10:37 +02:00
key_tuple = toast_flatten_tuple(oldtup, desc);
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
heap_freetuple(oldtup);
}
return key_tuple;
}
Allow read only connections during recovery, known as Hot Standby. Enabled by recovery_connections = on (default) and forcing archive recovery using a recovery.conf. Recovery processing now emulates the original transactions as they are replayed, providing full locking and MVCC behaviour for read only queries. Recovery must enter consistent state before connections are allowed, so there is a delay, typically short, before connections succeed. Replay of recovering transactions can conflict and in some cases deadlock with queries during recovery; these result in query cancellation after max_standby_delay seconds have expired. Infrastructure changes have minor effects on normal running, though introduce four new types of WAL record. New test mode "make standbycheck" allows regression tests of static command behaviour on a standby server while in recovery. Typical and extreme dynamic behaviours have been checked via code inspection and manual testing. Few port specific behaviours have been utilised, though primary testing has been on Linux only so far. This commit is the basic patch. Additional changes will follow in this release to enhance some aspects of behaviour, notably improved handling of conflicts, deadlock detection and query cancellation. Changes to VACUUM FULL are also required. Simon Riggs, with significant and lengthy review by Heikki Linnakangas, including streamlined redesign of snapshot creation and two-phase commit. Important contributions from Florian Pflug, Mark Kirkwood, Merlin Moncure, Greg Stark, Gianni Ciolli, Gabriele Bartolini, Hannu Krosing, Robert Haas, Tatsuo Ishii, Hiroyuki Yamada plus support and feedback from many other community members.
2009-12-19 02:32:45 +01:00
/*
Remove tupgone special case from vacuumlazy.c. Retry the call to heap_prune_page() in rare cases where there is disagreement between the heap_prune_page() call and the call to HeapTupleSatisfiesVacuum() that immediately follows. Disagreement is possible when a concurrently-aborted transaction makes a tuple DEAD during the tiny window between each step. This was the only case where a tuple considered DEAD by VACUUM still had storage following pruning. VACUUM's definition of dead tuples is now uniformly simple and unambiguous: dead tuples from each page are always LP_DEAD line pointers that were encountered just after we performed pruning (and just before we considered freezing remaining items with tuple storage). Eliminating the tupgone=true special case enables INDEX_CLEANUP=off style skipping of index vacuuming that takes place based on flexible, dynamic criteria. The INDEX_CLEANUP=off case had to know about skipping indexes up-front before now, due to a subtle interaction with the special case (see commit dd695979) -- this was a special case unto itself. Now there are no special cases. And so now it won't matter when or how we decide to skip index vacuuming: it won't affect how pruning behaves, and it won't be affected by any of the implementation details of pruning or freezing. Also remove XLOG_HEAP2_CLEANUP_INFO records. These are no longer necessary because we now rely entirely on heap pruning taking care of recovery conflicts. There is no longer any need to generate recovery conflicts for DEAD tuples that pruning just missed. This also means that heap vacuuming now uses exactly the same strategy for recovery conflicts as index vacuuming always has: REDO routines never need to process a latestRemovedXid from the WAL record, since earlier REDO of the WAL record from pruning is sufficient in all cases. The generic XLOG_HEAP2_CLEAN record type is now split into two new record types to reflect this new division (these are called XLOG_HEAP2_PRUNE and XLOG_HEAP2_VACUUM). Also stop acquiring a super-exclusive lock for heap pages when they're vacuumed during VACUUM's second heap pass. A regular exclusive lock is enough. This is correct because heap page vacuuming is now strictly a matter of setting the LP_DEAD line pointers to LP_UNUSED. No other backend can have a pointer to a tuple located in a pinned buffer that can be invalidated by a concurrent heap page vacuum operation. Heap vacuuming can now be thought of as conceptually similar to index vacuuming and conceptually dissimilar to heap pruning. Heap pruning now has sole responsibility for anything involving the logical contents of the database (e.g., managing transaction status information, recovery conflicts, considering what to do with HOT chains). Index vacuuming and heap vacuuming are now only concerned with recycling garbage items from physical data structures that back the logical database. Bump XLOG_PAGE_MAGIC due to pruning and heap page vacuum WAL record changes. Credit for the idea of retrying pruning a page to avoid the tupgone case goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Discussion: https://postgr.es/m/CAH2-WznneCXTzuFmcwx_EyRQgfsfJAAsu+CsqRFmFXCAar=nJw@mail.gmail.com
2021-04-06 17:49:22 +02:00
* Handles XLOG_HEAP2_PRUNE record type.
*
Standardize cleanup lock terminology. The term "super-exclusive lock" is a synonym for "buffer cleanup lock" that first appeared in nbtree many years ago. Standardize things by consistently using the term cleanup lock. This finishes work started by commit 276db875. There is no good reason to have two terms. But there is a good reason to only have one: to avoid confusion around why VACUUM acquires a full cleanup lock (not just an ordinary exclusive lock) in index AMs, during ambulkdelete calls. This has nothing to do with protecting the physical index data structure itself. It is needed to implement a locking protocol that ensures that TIDs pointing to the heap/table structure cannot get marked for recycling by VACUUM before it is safe (which is somewhat similar to how VACUUM uses cleanup locks during its first heap pass). Note that it isn't strictly necessary for index AMs to implement this locking protocol -- several index AMs use an MVCC snapshot as their sole interlock to prevent unsafe TID recycling. In passing, update the nbtree README. Cleanly separate discussion of the aforementioned index vacuuming locking protocol from discussion of the "drop leaf page pin" optimization added by commit 2ed5b87f. We now structure discussion of the latter by describing how individual index scans may safely opt out of applying the standard locking protocol (and so can avoid blocking progress by VACUUM). Also document why the optimization is not safe to apply during nbtree index-only scans. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzngHgQa92tz6NQihf4nxJwRzCV36yMJO_i8dS+2mgEVKw@mail.gmail.com Discussion: https://postgr.es/m/CAH2-WzkHPgsBBvGWjz=8PjNhDefy7XRkDKiT5NxMs-n5ZCf2dA@mail.gmail.com
2021-12-09 02:24:45 +01:00
* Acquires a full cleanup lock.
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
*/
WAL fixes.
2000-10-29 19:33:41 +01:00
static void
Remove tupgone special case from vacuumlazy.c. Retry the call to heap_prune_page() in rare cases where there is disagreement between the heap_prune_page() call and the call to HeapTupleSatisfiesVacuum() that immediately follows. Disagreement is possible when a concurrently-aborted transaction makes a tuple DEAD during the tiny window between each step. This was the only case where a tuple considered DEAD by VACUUM still had storage following pruning. VACUUM's definition of dead tuples is now uniformly simple and unambiguous: dead tuples from each page are always LP_DEAD line pointers that were encountered just after we performed pruning (and just before we considered freezing remaining items with tuple storage). Eliminating the tupgone=true special case enables INDEX_CLEANUP=off style skipping of index vacuuming that takes place based on flexible, dynamic criteria. The INDEX_CLEANUP=off case had to know about skipping indexes up-front before now, due to a subtle interaction with the special case (see commit dd695979) -- this was a special case unto itself. Now there are no special cases. And so now it won't matter when or how we decide to skip index vacuuming: it won't affect how pruning behaves, and it won't be affected by any of the implementation details of pruning or freezing. Also remove XLOG_HEAP2_CLEANUP_INFO records. These are no longer necessary because we now rely entirely on heap pruning taking care of recovery conflicts. There is no longer any need to generate recovery conflicts for DEAD tuples that pruning just missed. This also means that heap vacuuming now uses exactly the same strategy for recovery conflicts as index vacuuming always has: REDO routines never need to process a latestRemovedXid from the WAL record, since earlier REDO of the WAL record from pruning is sufficient in all cases. The generic XLOG_HEAP2_CLEAN record type is now split into two new record types to reflect this new division (these are called XLOG_HEAP2_PRUNE and XLOG_HEAP2_VACUUM). Also stop acquiring a super-exclusive lock for heap pages when they're vacuumed during VACUUM's second heap pass. A regular exclusive lock is enough. This is correct because heap page vacuuming is now strictly a matter of setting the LP_DEAD line pointers to LP_UNUSED. No other backend can have a pointer to a tuple located in a pinned buffer that can be invalidated by a concurrent heap page vacuum operation. Heap vacuuming can now be thought of as conceptually similar to index vacuuming and conceptually dissimilar to heap pruning. Heap pruning now has sole responsibility for anything involving the logical contents of the database (e.g., managing transaction status information, recovery conflicts, considering what to do with HOT chains). Index vacuuming and heap vacuuming are now only concerned with recycling garbage items from physical data structures that back the logical database. Bump XLOG_PAGE_MAGIC due to pruning and heap page vacuum WAL record changes. Credit for the idea of retrying pruning a page to avoid the tupgone case goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Discussion: https://postgr.es/m/CAH2-WznneCXTzuFmcwx_EyRQgfsfJAAsu+CsqRFmFXCAar=nJw@mail.gmail.com
2021-04-06 17:49:22 +02:00
heap_xlog_prune(XLogReaderState *record)
WAL fixes.
2000-10-29 19:33:41 +01:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRecPtr lsn = record->EndRecPtr;
Remove tupgone special case from vacuumlazy.c. Retry the call to heap_prune_page() in rare cases where there is disagreement between the heap_prune_page() call and the call to HeapTupleSatisfiesVacuum() that immediately follows. Disagreement is possible when a concurrently-aborted transaction makes a tuple DEAD during the tiny window between each step. This was the only case where a tuple considered DEAD by VACUUM still had storage following pruning. VACUUM's definition of dead tuples is now uniformly simple and unambiguous: dead tuples from each page are always LP_DEAD line pointers that were encountered just after we performed pruning (and just before we considered freezing remaining items with tuple storage). Eliminating the tupgone=true special case enables INDEX_CLEANUP=off style skipping of index vacuuming that takes place based on flexible, dynamic criteria. The INDEX_CLEANUP=off case had to know about skipping indexes up-front before now, due to a subtle interaction with the special case (see commit dd695979) -- this was a special case unto itself. Now there are no special cases. And so now it won't matter when or how we decide to skip index vacuuming: it won't affect how pruning behaves, and it won't be affected by any of the implementation details of pruning or freezing. Also remove XLOG_HEAP2_CLEANUP_INFO records. These are no longer necessary because we now rely entirely on heap pruning taking care of recovery conflicts. There is no longer any need to generate recovery conflicts for DEAD tuples that pruning just missed. This also means that heap vacuuming now uses exactly the same strategy for recovery conflicts as index vacuuming always has: REDO routines never need to process a latestRemovedXid from the WAL record, since earlier REDO of the WAL record from pruning is sufficient in all cases. The generic XLOG_HEAP2_CLEAN record type is now split into two new record types to reflect this new division (these are called XLOG_HEAP2_PRUNE and XLOG_HEAP2_VACUUM). Also stop acquiring a super-exclusive lock for heap pages when they're vacuumed during VACUUM's second heap pass. A regular exclusive lock is enough. This is correct because heap page vacuuming is now strictly a matter of setting the LP_DEAD line pointers to LP_UNUSED. No other backend can have a pointer to a tuple located in a pinned buffer that can be invalidated by a concurrent heap page vacuum operation. Heap vacuuming can now be thought of as conceptually similar to index vacuuming and conceptually dissimilar to heap pruning. Heap pruning now has sole responsibility for anything involving the logical contents of the database (e.g., managing transaction status information, recovery conflicts, considering what to do with HOT chains). Index vacuuming and heap vacuuming are now only concerned with recycling garbage items from physical data structures that back the logical database. Bump XLOG_PAGE_MAGIC due to pruning and heap page vacuum WAL record changes. Credit for the idea of retrying pruning a page to avoid the tupgone case goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Discussion: https://postgr.es/m/CAH2-WznneCXTzuFmcwx_EyRQgfsfJAAsu+CsqRFmFXCAar=nJw@mail.gmail.com
2021-04-06 17:49:22 +02:00
xl_heap_prune *xlrec = (xl_heap_prune *) XLogRecGetData(record);
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
Buffer buffer;
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
RelFileLocator rlocator;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
BlockNumber blkno;
XLogRedoAction action;
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecGetBlockTag(record, 0, &rlocator, NULL, &blkno);
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
Allow read only connections during recovery, known as Hot Standby. Enabled by recovery_connections = on (default) and forcing archive recovery using a recovery.conf. Recovery processing now emulates the original transactions as they are replayed, providing full locking and MVCC behaviour for read only queries. Recovery must enter consistent state before connections are allowed, so there is a delay, typically short, before connections succeed. Replay of recovering transactions can conflict and in some cases deadlock with queries during recovery; these result in query cancellation after max_standby_delay seconds have expired. Infrastructure changes have minor effects on normal running, though introduce four new types of WAL record. New test mode "make standbycheck" allows regression tests of static command behaviour on a standby server while in recovery. Typical and extreme dynamic behaviours have been checked via code inspection and manual testing. Few port specific behaviours have been utilised, though primary testing has been on Linux only so far. This commit is the basic patch. Additional changes will follow in this release to enhance some aspects of behaviour, notably improved handling of conflicts, deadlock detection and query cancellation. Changes to VACUUM FULL are also required. Simon Riggs, with significant and lengthy review by Heikki Linnakangas, including streamlined redesign of snapshot creation and two-phase commit. Important contributions from Florian Pflug, Mark Kirkwood, Merlin Moncure, Greg Stark, Gianni Ciolli, Gabriele Bartolini, Hannu Krosing, Robert Haas, Tatsuo Ishii, Hiroyuki Yamada plus support and feedback from many other community members.
2009-12-19 02:32:45 +01:00
/*
* We're about to remove tuples. In Hot Standby mode, ensure that there's
* no queries running for which the removed tuples are still visible.
*/
Remove tupgone special case from vacuumlazy.c. Retry the call to heap_prune_page() in rare cases where there is disagreement between the heap_prune_page() call and the call to HeapTupleSatisfiesVacuum() that immediately follows. Disagreement is possible when a concurrently-aborted transaction makes a tuple DEAD during the tiny window between each step. This was the only case where a tuple considered DEAD by VACUUM still had storage following pruning. VACUUM's definition of dead tuples is now uniformly simple and unambiguous: dead tuples from each page are always LP_DEAD line pointers that were encountered just after we performed pruning (and just before we considered freezing remaining items with tuple storage). Eliminating the tupgone=true special case enables INDEX_CLEANUP=off style skipping of index vacuuming that takes place based on flexible, dynamic criteria. The INDEX_CLEANUP=off case had to know about skipping indexes up-front before now, due to a subtle interaction with the special case (see commit dd695979) -- this was a special case unto itself. Now there are no special cases. And so now it won't matter when or how we decide to skip index vacuuming: it won't affect how pruning behaves, and it won't be affected by any of the implementation details of pruning or freezing. Also remove XLOG_HEAP2_CLEANUP_INFO records. These are no longer necessary because we now rely entirely on heap pruning taking care of recovery conflicts. There is no longer any need to generate recovery conflicts for DEAD tuples that pruning just missed. This also means that heap vacuuming now uses exactly the same strategy for recovery conflicts as index vacuuming always has: REDO routines never need to process a latestRemovedXid from the WAL record, since earlier REDO of the WAL record from pruning is sufficient in all cases. The generic XLOG_HEAP2_CLEAN record type is now split into two new record types to reflect this new division (these are called XLOG_HEAP2_PRUNE and XLOG_HEAP2_VACUUM). Also stop acquiring a super-exclusive lock for heap pages when they're vacuumed during VACUUM's second heap pass. A regular exclusive lock is enough. This is correct because heap page vacuuming is now strictly a matter of setting the LP_DEAD line pointers to LP_UNUSED. No other backend can have a pointer to a tuple located in a pinned buffer that can be invalidated by a concurrent heap page vacuum operation. Heap vacuuming can now be thought of as conceptually similar to index vacuuming and conceptually dissimilar to heap pruning. Heap pruning now has sole responsibility for anything involving the logical contents of the database (e.g., managing transaction status information, recovery conflicts, considering what to do with HOT chains). Index vacuuming and heap vacuuming are now only concerned with recycling garbage items from physical data structures that back the logical database. Bump XLOG_PAGE_MAGIC due to pruning and heap page vacuum WAL record changes. Credit for the idea of retrying pruning a page to avoid the tupgone case goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Discussion: https://postgr.es/m/CAH2-WznneCXTzuFmcwx_EyRQgfsfJAAsu+CsqRFmFXCAar=nJw@mail.gmail.com
2021-04-06 17:49:22 +02:00
if (InHotStandby)
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
ResolveRecoveryConflictWithSnapshot(xlrec->snapshotConflictHorizon,
Handle logical slot conflicts on standby During WAL replay on the standby, when a conflict with a logical slot is identified, invalidate such slots. There are two sources of conflicts: 1) Using the information added in 6af1793954e, logical slots are invalidated if required rows are removed 2) wal_level on the primary server is reduced to below logical Uses the infrastructure introduced in the prior commit. FIXME: add commit reference. Change InvalidatePossiblyObsoleteSlot() to use a recovery conflict to interrupt use of a slot, if called in the startup process. The new recovery conflict is added to pg_stat_database_conflicts, as confl_active_logicalslot. See 6af1793954e for an overall design of logical decoding on a standby. Bumps catversion for the addition of the pg_stat_database_conflicts column. Bumps PGSTAT_FILE_FORMAT_ID for the same reason. Author: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Author: Andres Freund <andres@anarazel.de> Author: Amit Khandekar <amitdkhan.pg@gmail.com> (in an older version) Reviewed-by: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Reviewed-by: Andres Freund <andres@anarazel.de> Reviewed-by: Robert Haas <robertmhaas@gmail.com> Reviewed-by: Fabrízio de Royes Mello <fabriziomello@gmail.com> Reviewed-by: Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com> Reviewed-by: Amit Kapila <amit.kapila16@gmail.com> Reviewed-by: Alvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/20230407075009.igg7be27ha2htkbt@awork3.anarazel.de
2023-04-08 08:11:28 +02:00
xlrec->isCatalogRel,
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
rlocator);
Allow read only connections during recovery, known as Hot Standby. Enabled by recovery_connections = on (default) and forcing archive recovery using a recovery.conf. Recovery processing now emulates the original transactions as they are replayed, providing full locking and MVCC behaviour for read only queries. Recovery must enter consistent state before connections are allowed, so there is a delay, typically short, before connections succeed. Replay of recovering transactions can conflict and in some cases deadlock with queries during recovery; these result in query cancellation after max_standby_delay seconds have expired. Infrastructure changes have minor effects on normal running, though introduce four new types of WAL record. New test mode "make standbycheck" allows regression tests of static command behaviour on a standby server while in recovery. Typical and extreme dynamic behaviours have been checked via code inspection and manual testing. Few port specific behaviours have been utilised, though primary testing has been on Linux only so far. This commit is the basic patch. Additional changes will follow in this release to enhance some aspects of behaviour, notably improved handling of conflicts, deadlock detection and query cancellation. Changes to VACUUM FULL are also required. Simon Riggs, with significant and lengthy review by Heikki Linnakangas, including streamlined redesign of snapshot creation and two-phase commit. Important contributions from Florian Pflug, Mark Kirkwood, Merlin Moncure, Greg Stark, Gianni Ciolli, Gabriele Bartolini, Hannu Krosing, Robert Haas, Tatsuo Ishii, Hiroyuki Yamada plus support and feedback from many other community members.
2009-12-19 02:32:45 +01:00
Fix multiple problems in WAL replay. Most of the replay functions for WAL record types that modify more than one page failed to ensure that those pages were locked correctly to ensure that concurrent queries could not see inconsistent page states. This is a hangover from coding decisions made long before Hot Standby was added, when it was hardly necessary to acquire buffer locks during WAL replay at all, let alone hold them for carefully-chosen periods. The key problem was that RestoreBkpBlocks was written to hold lock on each page restored from a full-page image for only as long as it took to update that page. This was guaranteed to break any WAL replay function in which there was any update-ordering constraint between pages, because even if the nominal order of the pages is the right one, any mixture of full-page and non-full-page updates in the same record would result in out-of-order updates. Moreover, it wouldn't work for situations where there's a requirement to maintain lock on one page while updating another. Failure to honor an update ordering constraint in this way is thought to be the cause of bug #7648 from Daniel Farina: what seems to have happened there is that a btree page being split was rewritten from a full-page image before the new right sibling page was written, and because lock on the original page was not maintained it was possible for hot standby queries to try to traverse the page's right-link to the not-yet-existing sibling page. To fix, get rid of RestoreBkpBlocks as such, and instead create a new function RestoreBackupBlock that restores just one full-page image at a time. This function can be invoked by WAL replay functions at the points where they would otherwise perform non-full-page updates; in this way, the physical order of page updates remains the same no matter which pages are replaced by full-page images. We can then further adjust the logic in individual replay functions if it is necessary to hold buffer locks for overlapping periods. A side benefit is that we can simplify the handling of concurrency conflict resolution by moving that code into the record-type-specfic functions; there's no more need to contort the code layout to keep conflict resolution in front of the RestoreBkpBlocks call. In connection with that, standardize on zero-based numbering rather than one-based numbering for referencing the full-page images. In HEAD, I removed the macros XLR_BKP_BLOCK_1 through XLR_BKP_BLOCK_4. They are still there in the header files in previous branches, but are no longer used by the code. In addition, fix some other bugs identified in the course of making these changes: spgRedoAddNode could fail to update the parent downlink at all, if the parent tuple is in the same page as either the old or new split tuple and we're not doing a full-page image: it would get fooled by the LSN having been advanced already. This would result in permanent index corruption, not just transient failure of concurrent queries. Also, ginHeapTupleFastInsert's "merge lists" case failed to mark the old tail page as a candidate for a full-page image; in the worst case this could result in torn-page corruption. heap_xlog_freeze() was inconsistent about using a cleanup lock or plain exclusive lock: it did the former in the normal path but the latter for a full-page image. A plain exclusive lock seems sufficient, so change to that. Also, remove gistRedoPageDeleteRecord(), which has been dead code since VACUUM FULL was rewritten. Back-patch to 9.0, where hot standby was introduced. Note however that 9.0 had a significantly different WAL-logging scheme for GIST index updates, and it doesn't appear possible to make that scheme safe for concurrent hot standby queries, because it can leave inconsistent states in the index even between WAL records. Given the lack of complaints from the field, we won't work too hard on fixing that branch.
2012-11-13 04:05:08 +01:00
/*
* If we have a full-page image, restore it (using a cleanup lock) and
* we're done.
*/
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
action = XLogReadBufferForRedoExtended(record, 0, RBM_NORMAL, true,
&buffer);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (action == BLK_NEEDS_REDO)
{
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
Page page = (Page) BufferGetPage(buffer);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
OffsetNumber *end;
OffsetNumber *redirected;
OffsetNumber *nowdead;
OffsetNumber *nowunused;
int nredirected;
int ndead;
int nunused;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
Size datalen;
redirected = (OffsetNumber *) XLogRecGetBlockData(record, 0, &datalen);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
nredirected = xlrec->nredirected;
ndead = xlrec->ndead;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
end = (OffsetNumber *) ((char *) redirected + datalen);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
nowdead = redirected + (nredirected * 2);
nowunused = nowdead + ndead;
nunused = (end - nowunused);
Assert(nunused >= 0);
Standardize ItemIdData terminology. The term "item pointer" should not be used to refer to ItemIdData variables, since that is needlessly ambiguous. Only ItemPointerData/ItemPointer variables should be called item pointers. To fix, establish the convention that ItemIdData variables should always be referred to either as "item identifiers" or "line pointers". The term "item identifier" already predominates in docs and translatable messages, and so should be the preferred alternative there. Discussion: https://postgr.es/m/CAH2-Wz=c=MZQjUzde3o9+2PLAPuHTpVZPPdYxN=E4ndQ2--8ew@mail.gmail.com
2019-05-14 00:53:39 +02:00
/* Update all line pointers per the record, and repair fragmentation */
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
heap_page_prune_execute(buffer,
redirected, nredirected,
nowdead, ndead,
nowunused, nunused);
/*
* Note: we don't worry about updating the page's prunability hints.
* At worst this will cause an extra prune cycle to occur soon.
*/
WAL fixes.
2000-10-29 19:33:41 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageSetLSN(page, lsn);
MarkBufferDirty(buffer);
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
}
Update FSM on WAL replay of page all-visible/frozen We aren't very strict about keeping FSM up to date on WAL replay, because per-page freespace values aren't critical in replicas (can't write to heap in a replica; and if the replica is promoted, the values would be updated by VACUUM anyway). However, VACUUM since 9.6 can skip processing pages marked all-visible or all-frozen, and if such pages are recorded in FSM with wrong values, those values are blindly propagated to FSM's upper layers by VACUUM's FreeSpaceMapVacuum. (This rationale assumes that crashes are not very frequent, because those would cause outdated FSM to occur in the primary.) Even when the FSM is outdated in standby, things are not too bad normally, because, most per-page FSM values will be zero (other than those propagated with the base-backup that created the standby); only once the remaining free space is less than 0.2*BLCKSZ the per-page value is maintained by WAL replay of heap ins/upd/del. However, if wal_log_hints=on causes complete FSM pages to be propagated to a standby via full-page images, many too-optimistic per-page values can end up being registered in the standby. Incorrect per-page values aren't critical in most cases, since an inserter that is given a page that doesn't actually contain the claimed free space will update FSM with the correct value, and retry until it finds a usable page. However, if there are many such updates to do, an inserter can spend a long time doing them before a usable page is found; in a heavily trafficked insert-only table with many concurrent inserters this has been observed to cause several second stalls, causing visible application malfunction. To fix this problem, it seems sufficient to have heap_xlog_visible (replay of setting all-visible and all-frozen VM bits for a heap page) update the FSM value for the page being processed. This fixes the per-page counters together with making the page skippable to vacuum, so when vacuum does FreeSpaceMapVacuum, the values propagated to FSM upper layers are the correct ones, avoiding the problem. While at it, apply the same fix to heap_xlog_clean (replay of tuple removal by HOT pruning and vacuum). This makes any space freed by the cleaning available earlier than the next vacuum in the promoted replica. Backpatch to 9.6, where this problem was diagnosed on an insert-only table with all-frozen pages, which were introduced as a concept in that release. Theoretically it could apply with all-visible pages to older branches, but there's been no report of that and it doesn't backpatch cleanly anyway. Author: Álvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/20180802172857.5skoexsilnjvgruk@alvherre.pgsql
2018-08-15 23:09:29 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (BufferIsValid(buffer))
Update FSM on WAL replay of page all-visible/frozen We aren't very strict about keeping FSM up to date on WAL replay, because per-page freespace values aren't critical in replicas (can't write to heap in a replica; and if the replica is promoted, the values would be updated by VACUUM anyway). However, VACUUM since 9.6 can skip processing pages marked all-visible or all-frozen, and if such pages are recorded in FSM with wrong values, those values are blindly propagated to FSM's upper layers by VACUUM's FreeSpaceMapVacuum. (This rationale assumes that crashes are not very frequent, because those would cause outdated FSM to occur in the primary.) Even when the FSM is outdated in standby, things are not too bad normally, because, most per-page FSM values will be zero (other than those propagated with the base-backup that created the standby); only once the remaining free space is less than 0.2*BLCKSZ the per-page value is maintained by WAL replay of heap ins/upd/del. However, if wal_log_hints=on causes complete FSM pages to be propagated to a standby via full-page images, many too-optimistic per-page values can end up being registered in the standby. Incorrect per-page values aren't critical in most cases, since an inserter that is given a page that doesn't actually contain the claimed free space will update FSM with the correct value, and retry until it finds a usable page. However, if there are many such updates to do, an inserter can spend a long time doing them before a usable page is found; in a heavily trafficked insert-only table with many concurrent inserters this has been observed to cause several second stalls, causing visible application malfunction. To fix this problem, it seems sufficient to have heap_xlog_visible (replay of setting all-visible and all-frozen VM bits for a heap page) update the FSM value for the page being processed. This fixes the per-page counters together with making the page skippable to vacuum, so when vacuum does FreeSpaceMapVacuum, the values propagated to FSM upper layers are the correct ones, avoiding the problem. While at it, apply the same fix to heap_xlog_clean (replay of tuple removal by HOT pruning and vacuum). This makes any space freed by the cleaning available earlier than the next vacuum in the promoted replica. Backpatch to 9.6, where this problem was diagnosed on an insert-only table with all-frozen pages, which were introduced as a concept in that release. Theoretically it could apply with all-visible pages to older branches, but there's been no report of that and it doesn't backpatch cleanly anyway. Author: Álvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/20180802172857.5skoexsilnjvgruk@alvherre.pgsql
2018-08-15 23:09:29 +02:00
{
Size freespace = PageGetHeapFreeSpace(BufferGetPage(buffer));
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
UnlockReleaseBuffer(buffer);
Update FSM on WAL replay. This is a bit limited; the FSM is only updated on non-full-page-image WAL records, and quite arbitrarily, only if there's less than 20% free space on the page after the insert/update (not on HOT updates, though). The 20% cutoff should avoid most of the overhead, when replaying a bulk insertion, for example, while ensuring that pages that are full are marked as full in the FSM. This is mostly to avoid the nasty worst case scenario, where you replay from a PITR archive, and the FSM information in the base backup is really out of date. If there was a lot of pages that the outdated FSM claims to have free space, but don't actually have any, the first unlucky inserter after the recovery would traverse through all those pages, just to find out that they're full. We didn't have this problem with the old FSM implementation, because we simply threw the FSM information away on a non-clean shutdown.
2008-10-31 20:40:27 +01:00
Update FSM on WAL replay of page all-visible/frozen We aren't very strict about keeping FSM up to date on WAL replay, because per-page freespace values aren't critical in replicas (can't write to heap in a replica; and if the replica is promoted, the values would be updated by VACUUM anyway). However, VACUUM since 9.6 can skip processing pages marked all-visible or all-frozen, and if such pages are recorded in FSM with wrong values, those values are blindly propagated to FSM's upper layers by VACUUM's FreeSpaceMapVacuum. (This rationale assumes that crashes are not very frequent, because those would cause outdated FSM to occur in the primary.) Even when the FSM is outdated in standby, things are not too bad normally, because, most per-page FSM values will be zero (other than those propagated with the base-backup that created the standby); only once the remaining free space is less than 0.2*BLCKSZ the per-page value is maintained by WAL replay of heap ins/upd/del. However, if wal_log_hints=on causes complete FSM pages to be propagated to a standby via full-page images, many too-optimistic per-page values can end up being registered in the standby. Incorrect per-page values aren't critical in most cases, since an inserter that is given a page that doesn't actually contain the claimed free space will update FSM with the correct value, and retry until it finds a usable page. However, if there are many such updates to do, an inserter can spend a long time doing them before a usable page is found; in a heavily trafficked insert-only table with many concurrent inserters this has been observed to cause several second stalls, causing visible application malfunction. To fix this problem, it seems sufficient to have heap_xlog_visible (replay of setting all-visible and all-frozen VM bits for a heap page) update the FSM value for the page being processed. This fixes the per-page counters together with making the page skippable to vacuum, so when vacuum does FreeSpaceMapVacuum, the values propagated to FSM upper layers are the correct ones, avoiding the problem. While at it, apply the same fix to heap_xlog_clean (replay of tuple removal by HOT pruning and vacuum). This makes any space freed by the cleaning available earlier than the next vacuum in the promoted replica. Backpatch to 9.6, where this problem was diagnosed on an insert-only table with all-frozen pages, which were introduced as a concept in that release. Theoretically it could apply with all-visible pages to older branches, but there's been no report of that and it doesn't backpatch cleanly anyway. Author: Álvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/20180802172857.5skoexsilnjvgruk@alvherre.pgsql
2018-08-15 23:09:29 +02:00
/*
Remove tupgone special case from vacuumlazy.c. Retry the call to heap_prune_page() in rare cases where there is disagreement between the heap_prune_page() call and the call to HeapTupleSatisfiesVacuum() that immediately follows. Disagreement is possible when a concurrently-aborted transaction makes a tuple DEAD during the tiny window between each step. This was the only case where a tuple considered DEAD by VACUUM still had storage following pruning. VACUUM's definition of dead tuples is now uniformly simple and unambiguous: dead tuples from each page are always LP_DEAD line pointers that were encountered just after we performed pruning (and just before we considered freezing remaining items with tuple storage). Eliminating the tupgone=true special case enables INDEX_CLEANUP=off style skipping of index vacuuming that takes place based on flexible, dynamic criteria. The INDEX_CLEANUP=off case had to know about skipping indexes up-front before now, due to a subtle interaction with the special case (see commit dd695979) -- this was a special case unto itself. Now there are no special cases. And so now it won't matter when or how we decide to skip index vacuuming: it won't affect how pruning behaves, and it won't be affected by any of the implementation details of pruning or freezing. Also remove XLOG_HEAP2_CLEANUP_INFO records. These are no longer necessary because we now rely entirely on heap pruning taking care of recovery conflicts. There is no longer any need to generate recovery conflicts for DEAD tuples that pruning just missed. This also means that heap vacuuming now uses exactly the same strategy for recovery conflicts as index vacuuming always has: REDO routines never need to process a latestRemovedXid from the WAL record, since earlier REDO of the WAL record from pruning is sufficient in all cases. The generic XLOG_HEAP2_CLEAN record type is now split into two new record types to reflect this new division (these are called XLOG_HEAP2_PRUNE and XLOG_HEAP2_VACUUM). Also stop acquiring a super-exclusive lock for heap pages when they're vacuumed during VACUUM's second heap pass. A regular exclusive lock is enough. This is correct because heap page vacuuming is now strictly a matter of setting the LP_DEAD line pointers to LP_UNUSED. No other backend can have a pointer to a tuple located in a pinned buffer that can be invalidated by a concurrent heap page vacuum operation. Heap vacuuming can now be thought of as conceptually similar to index vacuuming and conceptually dissimilar to heap pruning. Heap pruning now has sole responsibility for anything involving the logical contents of the database (e.g., managing transaction status information, recovery conflicts, considering what to do with HOT chains). Index vacuuming and heap vacuuming are now only concerned with recycling garbage items from physical data structures that back the logical database. Bump XLOG_PAGE_MAGIC due to pruning and heap page vacuum WAL record changes. Credit for the idea of retrying pruning a page to avoid the tupgone case goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Discussion: https://postgr.es/m/CAH2-WznneCXTzuFmcwx_EyRQgfsfJAAsu+CsqRFmFXCAar=nJw@mail.gmail.com
2021-04-06 17:49:22 +02:00
* After pruning records from a page, it's useful to update the FSM
Update FSM on WAL replay of page all-visible/frozen We aren't very strict about keeping FSM up to date on WAL replay, because per-page freespace values aren't critical in replicas (can't write to heap in a replica; and if the replica is promoted, the values would be updated by VACUUM anyway). However, VACUUM since 9.6 can skip processing pages marked all-visible or all-frozen, and if such pages are recorded in FSM with wrong values, those values are blindly propagated to FSM's upper layers by VACUUM's FreeSpaceMapVacuum. (This rationale assumes that crashes are not very frequent, because those would cause outdated FSM to occur in the primary.) Even when the FSM is outdated in standby, things are not too bad normally, because, most per-page FSM values will be zero (other than those propagated with the base-backup that created the standby); only once the remaining free space is less than 0.2*BLCKSZ the per-page value is maintained by WAL replay of heap ins/upd/del. However, if wal_log_hints=on causes complete FSM pages to be propagated to a standby via full-page images, many too-optimistic per-page values can end up being registered in the standby. Incorrect per-page values aren't critical in most cases, since an inserter that is given a page that doesn't actually contain the claimed free space will update FSM with the correct value, and retry until it finds a usable page. However, if there are many such updates to do, an inserter can spend a long time doing them before a usable page is found; in a heavily trafficked insert-only table with many concurrent inserters this has been observed to cause several second stalls, causing visible application malfunction. To fix this problem, it seems sufficient to have heap_xlog_visible (replay of setting all-visible and all-frozen VM bits for a heap page) update the FSM value for the page being processed. This fixes the per-page counters together with making the page skippable to vacuum, so when vacuum does FreeSpaceMapVacuum, the values propagated to FSM upper layers are the correct ones, avoiding the problem. While at it, apply the same fix to heap_xlog_clean (replay of tuple removal by HOT pruning and vacuum). This makes any space freed by the cleaning available earlier than the next vacuum in the promoted replica. Backpatch to 9.6, where this problem was diagnosed on an insert-only table with all-frozen pages, which were introduced as a concept in that release. Theoretically it could apply with all-visible pages to older branches, but there's been no report of that and it doesn't backpatch cleanly anyway. Author: Álvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/20180802172857.5skoexsilnjvgruk@alvherre.pgsql
2018-08-15 23:09:29 +02:00
* about it, as it may cause the page become target for insertions
* later even if vacuum decides not to visit it (which is possible if
* gets marked all-visible.)
*
* Do this regardless of a full-page image being applied, since the
* FSM data is not in the page anyway.
*/
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecordPageWithFreeSpace(rlocator, blkno, freespace);
Update FSM on WAL replay of page all-visible/frozen We aren't very strict about keeping FSM up to date on WAL replay, because per-page freespace values aren't critical in replicas (can't write to heap in a replica; and if the replica is promoted, the values would be updated by VACUUM anyway). However, VACUUM since 9.6 can skip processing pages marked all-visible or all-frozen, and if such pages are recorded in FSM with wrong values, those values are blindly propagated to FSM's upper layers by VACUUM's FreeSpaceMapVacuum. (This rationale assumes that crashes are not very frequent, because those would cause outdated FSM to occur in the primary.) Even when the FSM is outdated in standby, things are not too bad normally, because, most per-page FSM values will be zero (other than those propagated with the base-backup that created the standby); only once the remaining free space is less than 0.2*BLCKSZ the per-page value is maintained by WAL replay of heap ins/upd/del. However, if wal_log_hints=on causes complete FSM pages to be propagated to a standby via full-page images, many too-optimistic per-page values can end up being registered in the standby. Incorrect per-page values aren't critical in most cases, since an inserter that is given a page that doesn't actually contain the claimed free space will update FSM with the correct value, and retry until it finds a usable page. However, if there are many such updates to do, an inserter can spend a long time doing them before a usable page is found; in a heavily trafficked insert-only table with many concurrent inserters this has been observed to cause several second stalls, causing visible application malfunction. To fix this problem, it seems sufficient to have heap_xlog_visible (replay of setting all-visible and all-frozen VM bits for a heap page) update the FSM value for the page being processed. This fixes the per-page counters together with making the page skippable to vacuum, so when vacuum does FreeSpaceMapVacuum, the values propagated to FSM upper layers are the correct ones, avoiding the problem. While at it, apply the same fix to heap_xlog_clean (replay of tuple removal by HOT pruning and vacuum). This makes any space freed by the cleaning available earlier than the next vacuum in the promoted replica. Backpatch to 9.6, where this problem was diagnosed on an insert-only table with all-frozen pages, which were introduced as a concept in that release. Theoretically it could apply with all-visible pages to older branches, but there's been no report of that and it doesn't backpatch cleanly anyway. Author: Álvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/20180802172857.5skoexsilnjvgruk@alvherre.pgsql
2018-08-15 23:09:29 +02:00
}
WAL fixes.
2000-10-29 19:33:41 +01:00
}
Remove tupgone special case from vacuumlazy.c. Retry the call to heap_prune_page() in rare cases where there is disagreement between the heap_prune_page() call and the call to HeapTupleSatisfiesVacuum() that immediately follows. Disagreement is possible when a concurrently-aborted transaction makes a tuple DEAD during the tiny window between each step. This was the only case where a tuple considered DEAD by VACUUM still had storage following pruning. VACUUM's definition of dead tuples is now uniformly simple and unambiguous: dead tuples from each page are always LP_DEAD line pointers that were encountered just after we performed pruning (and just before we considered freezing remaining items with tuple storage). Eliminating the tupgone=true special case enables INDEX_CLEANUP=off style skipping of index vacuuming that takes place based on flexible, dynamic criteria. The INDEX_CLEANUP=off case had to know about skipping indexes up-front before now, due to a subtle interaction with the special case (see commit dd695979) -- this was a special case unto itself. Now there are no special cases. And so now it won't matter when or how we decide to skip index vacuuming: it won't affect how pruning behaves, and it won't be affected by any of the implementation details of pruning or freezing. Also remove XLOG_HEAP2_CLEANUP_INFO records. These are no longer necessary because we now rely entirely on heap pruning taking care of recovery conflicts. There is no longer any need to generate recovery conflicts for DEAD tuples that pruning just missed. This also means that heap vacuuming now uses exactly the same strategy for recovery conflicts as index vacuuming always has: REDO routines never need to process a latestRemovedXid from the WAL record, since earlier REDO of the WAL record from pruning is sufficient in all cases. The generic XLOG_HEAP2_CLEAN record type is now split into two new record types to reflect this new division (these are called XLOG_HEAP2_PRUNE and XLOG_HEAP2_VACUUM). Also stop acquiring a super-exclusive lock for heap pages when they're vacuumed during VACUUM's second heap pass. A regular exclusive lock is enough. This is correct because heap page vacuuming is now strictly a matter of setting the LP_DEAD line pointers to LP_UNUSED. No other backend can have a pointer to a tuple located in a pinned buffer that can be invalidated by a concurrent heap page vacuum operation. Heap vacuuming can now be thought of as conceptually similar to index vacuuming and conceptually dissimilar to heap pruning. Heap pruning now has sole responsibility for anything involving the logical contents of the database (e.g., managing transaction status information, recovery conflicts, considering what to do with HOT chains). Index vacuuming and heap vacuuming are now only concerned with recycling garbage items from physical data structures that back the logical database. Bump XLOG_PAGE_MAGIC due to pruning and heap page vacuum WAL record changes. Credit for the idea of retrying pruning a page to avoid the tupgone case goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Discussion: https://postgr.es/m/CAH2-WznneCXTzuFmcwx_EyRQgfsfJAAsu+CsqRFmFXCAar=nJw@mail.gmail.com
2021-04-06 17:49:22 +02:00
/*
* Handles XLOG_HEAP2_VACUUM record type.
*
Standardize cleanup lock terminology. The term "super-exclusive lock" is a synonym for "buffer cleanup lock" that first appeared in nbtree many years ago. Standardize things by consistently using the term cleanup lock. This finishes work started by commit 276db875. There is no good reason to have two terms. But there is a good reason to only have one: to avoid confusion around why VACUUM acquires a full cleanup lock (not just an ordinary exclusive lock) in index AMs, during ambulkdelete calls. This has nothing to do with protecting the physical index data structure itself. It is needed to implement a locking protocol that ensures that TIDs pointing to the heap/table structure cannot get marked for recycling by VACUUM before it is safe (which is somewhat similar to how VACUUM uses cleanup locks during its first heap pass). Note that it isn't strictly necessary for index AMs to implement this locking protocol -- several index AMs use an MVCC snapshot as their sole interlock to prevent unsafe TID recycling. In passing, update the nbtree README. Cleanly separate discussion of the aforementioned index vacuuming locking protocol from discussion of the "drop leaf page pin" optimization added by commit 2ed5b87f. We now structure discussion of the latter by describing how individual index scans may safely opt out of applying the standard locking protocol (and so can avoid blocking progress by VACUUM). Also document why the optimization is not safe to apply during nbtree index-only scans. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzngHgQa92tz6NQihf4nxJwRzCV36yMJO_i8dS+2mgEVKw@mail.gmail.com Discussion: https://postgr.es/m/CAH2-WzkHPgsBBvGWjz=8PjNhDefy7XRkDKiT5NxMs-n5ZCf2dA@mail.gmail.com
2021-12-09 02:24:45 +01:00
* Acquires an ordinary exclusive lock only.
Remove tupgone special case from vacuumlazy.c. Retry the call to heap_prune_page() in rare cases where there is disagreement between the heap_prune_page() call and the call to HeapTupleSatisfiesVacuum() that immediately follows. Disagreement is possible when a concurrently-aborted transaction makes a tuple DEAD during the tiny window between each step. This was the only case where a tuple considered DEAD by VACUUM still had storage following pruning. VACUUM's definition of dead tuples is now uniformly simple and unambiguous: dead tuples from each page are always LP_DEAD line pointers that were encountered just after we performed pruning (and just before we considered freezing remaining items with tuple storage). Eliminating the tupgone=true special case enables INDEX_CLEANUP=off style skipping of index vacuuming that takes place based on flexible, dynamic criteria. The INDEX_CLEANUP=off case had to know about skipping indexes up-front before now, due to a subtle interaction with the special case (see commit dd695979) -- this was a special case unto itself. Now there are no special cases. And so now it won't matter when or how we decide to skip index vacuuming: it won't affect how pruning behaves, and it won't be affected by any of the implementation details of pruning or freezing. Also remove XLOG_HEAP2_CLEANUP_INFO records. These are no longer necessary because we now rely entirely on heap pruning taking care of recovery conflicts. There is no longer any need to generate recovery conflicts for DEAD tuples that pruning just missed. This also means that heap vacuuming now uses exactly the same strategy for recovery conflicts as index vacuuming always has: REDO routines never need to process a latestRemovedXid from the WAL record, since earlier REDO of the WAL record from pruning is sufficient in all cases. The generic XLOG_HEAP2_CLEAN record type is now split into two new record types to reflect this new division (these are called XLOG_HEAP2_PRUNE and XLOG_HEAP2_VACUUM). Also stop acquiring a super-exclusive lock for heap pages when they're vacuumed during VACUUM's second heap pass. A regular exclusive lock is enough. This is correct because heap page vacuuming is now strictly a matter of setting the LP_DEAD line pointers to LP_UNUSED. No other backend can have a pointer to a tuple located in a pinned buffer that can be invalidated by a concurrent heap page vacuum operation. Heap vacuuming can now be thought of as conceptually similar to index vacuuming and conceptually dissimilar to heap pruning. Heap pruning now has sole responsibility for anything involving the logical contents of the database (e.g., managing transaction status information, recovery conflicts, considering what to do with HOT chains). Index vacuuming and heap vacuuming are now only concerned with recycling garbage items from physical data structures that back the logical database. Bump XLOG_PAGE_MAGIC due to pruning and heap page vacuum WAL record changes. Credit for the idea of retrying pruning a page to avoid the tupgone case goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Discussion: https://postgr.es/m/CAH2-WznneCXTzuFmcwx_EyRQgfsfJAAsu+CsqRFmFXCAar=nJw@mail.gmail.com
2021-04-06 17:49:22 +02:00
*/
static void
heap_xlog_vacuum(XLogReaderState *record)
{
XLogRecPtr lsn = record->EndRecPtr;
xl_heap_vacuum *xlrec = (xl_heap_vacuum *) XLogRecGetData(record);
Buffer buffer;
BlockNumber blkno;
XLogRedoAction action;
/*
* If we have a full-page image, restore it (without using a cleanup lock)
* and we're done.
*/
action = XLogReadBufferForRedoExtended(record, 0, RBM_NORMAL, false,
&buffer);
if (action == BLK_NEEDS_REDO)
{
Page page = (Page) BufferGetPage(buffer);
OffsetNumber *nowunused;
Size datalen;
OffsetNumber *offnum;
nowunused = (OffsetNumber *) XLogRecGetBlockData(record, 0, &datalen);
/* Shouldn't be a record unless there's something to do */
Assert(xlrec->nunused > 0);
/* Update all now-unused line pointers */
offnum = nowunused;
for (int i = 0; i < xlrec->nunused; i++)
{
OffsetNumber off = *offnum++;
ItemId lp = PageGetItemId(page, off);
Assert(ItemIdIsDead(lp) && !ItemIdHasStorage(lp));
ItemIdSetUnused(lp);
}
Truncate line pointer array during VACUUM. Teach VACUUM to truncate the line pointer array of each heap page when a contiguous group of LP_UNUSED line pointers appear at the end of the array -- these unused and unreferenced items are excluded. This process occurs during VACUUM's second pass over the heap, right after LP_DEAD line pointers on the page (those encountered/pruned during the first pass) are marked LP_UNUSED. Truncation avoids line pointer bloat with certain workloads, particularly those involving continual range DELETEs and bulk INSERTs against the same table. Also harden heapam code to check for an out-of-range page offset number in places where we weren't already doing so. Author: Matthias van de Meent <boekewurm+postgres@gmail.com> Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAEze2WjgaQc55Y5f5CQd3L=eS5CZcff2Obxp=O6pto8-f0hC4w@mail.gmail.com Discussion: https://postgr.es/m/CAH2-Wzn6a64PJM1Ggzm=uvx2otsopJMhFQj_g1rAj4GWr3ZSzw@mail.gmail.com
2021-04-07 17:47:15 +02:00
/* Attempt to truncate line pointer array now */
PageTruncateLinePointerArray(page);
Remove tupgone special case from vacuumlazy.c. Retry the call to heap_prune_page() in rare cases where there is disagreement between the heap_prune_page() call and the call to HeapTupleSatisfiesVacuum() that immediately follows. Disagreement is possible when a concurrently-aborted transaction makes a tuple DEAD during the tiny window between each step. This was the only case where a tuple considered DEAD by VACUUM still had storage following pruning. VACUUM's definition of dead tuples is now uniformly simple and unambiguous: dead tuples from each page are always LP_DEAD line pointers that were encountered just after we performed pruning (and just before we considered freezing remaining items with tuple storage). Eliminating the tupgone=true special case enables INDEX_CLEANUP=off style skipping of index vacuuming that takes place based on flexible, dynamic criteria. The INDEX_CLEANUP=off case had to know about skipping indexes up-front before now, due to a subtle interaction with the special case (see commit dd695979) -- this was a special case unto itself. Now there are no special cases. And so now it won't matter when or how we decide to skip index vacuuming: it won't affect how pruning behaves, and it won't be affected by any of the implementation details of pruning or freezing. Also remove XLOG_HEAP2_CLEANUP_INFO records. These are no longer necessary because we now rely entirely on heap pruning taking care of recovery conflicts. There is no longer any need to generate recovery conflicts for DEAD tuples that pruning just missed. This also means that heap vacuuming now uses exactly the same strategy for recovery conflicts as index vacuuming always has: REDO routines never need to process a latestRemovedXid from the WAL record, since earlier REDO of the WAL record from pruning is sufficient in all cases. The generic XLOG_HEAP2_CLEAN record type is now split into two new record types to reflect this new division (these are called XLOG_HEAP2_PRUNE and XLOG_HEAP2_VACUUM). Also stop acquiring a super-exclusive lock for heap pages when they're vacuumed during VACUUM's second heap pass. A regular exclusive lock is enough. This is correct because heap page vacuuming is now strictly a matter of setting the LP_DEAD line pointers to LP_UNUSED. No other backend can have a pointer to a tuple located in a pinned buffer that can be invalidated by a concurrent heap page vacuum operation. Heap vacuuming can now be thought of as conceptually similar to index vacuuming and conceptually dissimilar to heap pruning. Heap pruning now has sole responsibility for anything involving the logical contents of the database (e.g., managing transaction status information, recovery conflicts, considering what to do with HOT chains). Index vacuuming and heap vacuuming are now only concerned with recycling garbage items from physical data structures that back the logical database. Bump XLOG_PAGE_MAGIC due to pruning and heap page vacuum WAL record changes. Credit for the idea of retrying pruning a page to avoid the tupgone case goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Discussion: https://postgr.es/m/CAH2-WznneCXTzuFmcwx_EyRQgfsfJAAsu+CsqRFmFXCAar=nJw@mail.gmail.com
2021-04-06 17:49:22 +02:00
PageSetLSN(page, lsn);
MarkBufferDirty(buffer);
}
if (BufferIsValid(buffer))
{
Size freespace = PageGetHeapFreeSpace(BufferGetPage(buffer));
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
RelFileLocator rlocator;
Remove tupgone special case from vacuumlazy.c. Retry the call to heap_prune_page() in rare cases where there is disagreement between the heap_prune_page() call and the call to HeapTupleSatisfiesVacuum() that immediately follows. Disagreement is possible when a concurrently-aborted transaction makes a tuple DEAD during the tiny window between each step. This was the only case where a tuple considered DEAD by VACUUM still had storage following pruning. VACUUM's definition of dead tuples is now uniformly simple and unambiguous: dead tuples from each page are always LP_DEAD line pointers that were encountered just after we performed pruning (and just before we considered freezing remaining items with tuple storage). Eliminating the tupgone=true special case enables INDEX_CLEANUP=off style skipping of index vacuuming that takes place based on flexible, dynamic criteria. The INDEX_CLEANUP=off case had to know about skipping indexes up-front before now, due to a subtle interaction with the special case (see commit dd695979) -- this was a special case unto itself. Now there are no special cases. And so now it won't matter when or how we decide to skip index vacuuming: it won't affect how pruning behaves, and it won't be affected by any of the implementation details of pruning or freezing. Also remove XLOG_HEAP2_CLEANUP_INFO records. These are no longer necessary because we now rely entirely on heap pruning taking care of recovery conflicts. There is no longer any need to generate recovery conflicts for DEAD tuples that pruning just missed. This also means that heap vacuuming now uses exactly the same strategy for recovery conflicts as index vacuuming always has: REDO routines never need to process a latestRemovedXid from the WAL record, since earlier REDO of the WAL record from pruning is sufficient in all cases. The generic XLOG_HEAP2_CLEAN record type is now split into two new record types to reflect this new division (these are called XLOG_HEAP2_PRUNE and XLOG_HEAP2_VACUUM). Also stop acquiring a super-exclusive lock for heap pages when they're vacuumed during VACUUM's second heap pass. A regular exclusive lock is enough. This is correct because heap page vacuuming is now strictly a matter of setting the LP_DEAD line pointers to LP_UNUSED. No other backend can have a pointer to a tuple located in a pinned buffer that can be invalidated by a concurrent heap page vacuum operation. Heap vacuuming can now be thought of as conceptually similar to index vacuuming and conceptually dissimilar to heap pruning. Heap pruning now has sole responsibility for anything involving the logical contents of the database (e.g., managing transaction status information, recovery conflicts, considering what to do with HOT chains). Index vacuuming and heap vacuuming are now only concerned with recycling garbage items from physical data structures that back the logical database. Bump XLOG_PAGE_MAGIC due to pruning and heap page vacuum WAL record changes. Credit for the idea of retrying pruning a page to avoid the tupgone case goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Discussion: https://postgr.es/m/CAH2-WznneCXTzuFmcwx_EyRQgfsfJAAsu+CsqRFmFXCAar=nJw@mail.gmail.com
2021-04-06 17:49:22 +02:00
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecGetBlockTag(record, 0, &rlocator, NULL, &blkno);
Remove tupgone special case from vacuumlazy.c. Retry the call to heap_prune_page() in rare cases where there is disagreement between the heap_prune_page() call and the call to HeapTupleSatisfiesVacuum() that immediately follows. Disagreement is possible when a concurrently-aborted transaction makes a tuple DEAD during the tiny window between each step. This was the only case where a tuple considered DEAD by VACUUM still had storage following pruning. VACUUM's definition of dead tuples is now uniformly simple and unambiguous: dead tuples from each page are always LP_DEAD line pointers that were encountered just after we performed pruning (and just before we considered freezing remaining items with tuple storage). Eliminating the tupgone=true special case enables INDEX_CLEANUP=off style skipping of index vacuuming that takes place based on flexible, dynamic criteria. The INDEX_CLEANUP=off case had to know about skipping indexes up-front before now, due to a subtle interaction with the special case (see commit dd695979) -- this was a special case unto itself. Now there are no special cases. And so now it won't matter when or how we decide to skip index vacuuming: it won't affect how pruning behaves, and it won't be affected by any of the implementation details of pruning or freezing. Also remove XLOG_HEAP2_CLEANUP_INFO records. These are no longer necessary because we now rely entirely on heap pruning taking care of recovery conflicts. There is no longer any need to generate recovery conflicts for DEAD tuples that pruning just missed. This also means that heap vacuuming now uses exactly the same strategy for recovery conflicts as index vacuuming always has: REDO routines never need to process a latestRemovedXid from the WAL record, since earlier REDO of the WAL record from pruning is sufficient in all cases. The generic XLOG_HEAP2_CLEAN record type is now split into two new record types to reflect this new division (these are called XLOG_HEAP2_PRUNE and XLOG_HEAP2_VACUUM). Also stop acquiring a super-exclusive lock for heap pages when they're vacuumed during VACUUM's second heap pass. A regular exclusive lock is enough. This is correct because heap page vacuuming is now strictly a matter of setting the LP_DEAD line pointers to LP_UNUSED. No other backend can have a pointer to a tuple located in a pinned buffer that can be invalidated by a concurrent heap page vacuum operation. Heap vacuuming can now be thought of as conceptually similar to index vacuuming and conceptually dissimilar to heap pruning. Heap pruning now has sole responsibility for anything involving the logical contents of the database (e.g., managing transaction status information, recovery conflicts, considering what to do with HOT chains). Index vacuuming and heap vacuuming are now only concerned with recycling garbage items from physical data structures that back the logical database. Bump XLOG_PAGE_MAGIC due to pruning and heap page vacuum WAL record changes. Credit for the idea of retrying pruning a page to avoid the tupgone case goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Discussion: https://postgr.es/m/CAH2-WznneCXTzuFmcwx_EyRQgfsfJAAsu+CsqRFmFXCAar=nJw@mail.gmail.com
2021-04-06 17:49:22 +02:00
UnlockReleaseBuffer(buffer);
/*
* After vacuuming LP_DEAD items from a page, it's useful to update
* the FSM about it, as it may cause the page become target for
* insertions later even if vacuum decides not to visit it (which is
* possible if gets marked all-visible.)
*
* Do this regardless of a full-page image being applied, since the
* FSM data is not in the page anyway.
*/
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecordPageWithFreeSpace(rlocator, blkno, freespace);
Remove tupgone special case from vacuumlazy.c. Retry the call to heap_prune_page() in rare cases where there is disagreement between the heap_prune_page() call and the call to HeapTupleSatisfiesVacuum() that immediately follows. Disagreement is possible when a concurrently-aborted transaction makes a tuple DEAD during the tiny window between each step. This was the only case where a tuple considered DEAD by VACUUM still had storage following pruning. VACUUM's definition of dead tuples is now uniformly simple and unambiguous: dead tuples from each page are always LP_DEAD line pointers that were encountered just after we performed pruning (and just before we considered freezing remaining items with tuple storage). Eliminating the tupgone=true special case enables INDEX_CLEANUP=off style skipping of index vacuuming that takes place based on flexible, dynamic criteria. The INDEX_CLEANUP=off case had to know about skipping indexes up-front before now, due to a subtle interaction with the special case (see commit dd695979) -- this was a special case unto itself. Now there are no special cases. And so now it won't matter when or how we decide to skip index vacuuming: it won't affect how pruning behaves, and it won't be affected by any of the implementation details of pruning or freezing. Also remove XLOG_HEAP2_CLEANUP_INFO records. These are no longer necessary because we now rely entirely on heap pruning taking care of recovery conflicts. There is no longer any need to generate recovery conflicts for DEAD tuples that pruning just missed. This also means that heap vacuuming now uses exactly the same strategy for recovery conflicts as index vacuuming always has: REDO routines never need to process a latestRemovedXid from the WAL record, since earlier REDO of the WAL record from pruning is sufficient in all cases. The generic XLOG_HEAP2_CLEAN record type is now split into two new record types to reflect this new division (these are called XLOG_HEAP2_PRUNE and XLOG_HEAP2_VACUUM). Also stop acquiring a super-exclusive lock for heap pages when they're vacuumed during VACUUM's second heap pass. A regular exclusive lock is enough. This is correct because heap page vacuuming is now strictly a matter of setting the LP_DEAD line pointers to LP_UNUSED. No other backend can have a pointer to a tuple located in a pinned buffer that can be invalidated by a concurrent heap page vacuum operation. Heap vacuuming can now be thought of as conceptually similar to index vacuuming and conceptually dissimilar to heap pruning. Heap pruning now has sole responsibility for anything involving the logical contents of the database (e.g., managing transaction status information, recovery conflicts, considering what to do with HOT chains). Index vacuuming and heap vacuuming are now only concerned with recycling garbage items from physical data structures that back the logical database. Bump XLOG_PAGE_MAGIC due to pruning and heap page vacuum WAL record changes. Credit for the idea of retrying pruning a page to avoid the tupgone case goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Discussion: https://postgr.es/m/CAH2-WznneCXTzuFmcwx_EyRQgfsfJAAsu+CsqRFmFXCAar=nJw@mail.gmail.com
2021-04-06 17:49:22 +02:00
}
}
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
/*
* Replay XLOG_HEAP2_VISIBLE record.
*
* The critical integrity requirement here is that we must never end up with
* a situation where the visibility map bit is set, and the page-level
* PD_ALL_VISIBLE bit is clear. If that were to occur, then a subsequent
* page modification would fail to clear the visibility map bit.
*/
static void
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_visible(XLogReaderState *record)
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRecPtr lsn = record->EndRecPtr;
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
xl_heap_visible *xlrec = (xl_heap_visible *) XLogRecGetData(record);
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
Buffer vmbuffer = InvalidBuffer;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
Buffer buffer;
Page page;
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
RelFileLocator rlocator;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
BlockNumber blkno;
XLogRedoAction action;
Add info in WAL records in preparation for logical slot conflict handling This commit only implements one prerequisite part for allowing logical decoding. The commit message contains an explanation of the overall design, which later commits will refer back to. Overall design: 1. We want to enable logical decoding on standbys, but replay of WAL from the primary might remove data that is needed by logical decoding, causing error(s) on the standby. To prevent those errors, a new replication conflict scenario needs to be addressed (as much as hot standby does). 2. Our chosen strategy for dealing with this type of replication slot is to invalidate logical slots for which needed data has been removed. 3. To do this we need the latestRemovedXid for each change, just as we do for physical replication conflicts, but we also need to know whether any particular change was to data that logical replication might access. That way, during WAL replay, we know when there is a risk of conflict and, if so, if there is a conflict. 4. We can't rely on the standby's relcache entries for this purpose in any way, because the startup process can't access catalog contents. 5. Therefore every WAL record that potentially removes data from the index or heap must carry a flag indicating whether or not it is one that might be accessed during logical decoding. Why do we need this for logical decoding on standby? First, let's forget about logical decoding on standby and recall that on a primary database, any catalog rows that may be needed by a logical decoding replication slot are not removed. This is done thanks to the catalog_xmin associated with the logical replication slot. But, with logical decoding on standby, in the following cases: - hot_standby_feedback is off - hot_standby_feedback is on but there is no a physical slot between the primary and the standby. Then, hot_standby_feedback will work, but only while the connection is alive (for example a node restart would break it) Then, the primary may delete system catalog rows that could be needed by the logical decoding on the standby (as it does not know about the catalog_xmin on the standby). So, it’s mandatory to identify those rows and invalidate the slots that may need them if any. Identifying those rows is the purpose of this commit. Implementation: When a WAL replay on standby indicates that a catalog table tuple is to be deleted by an xid that is greater than a logical slot's catalog_xmin, then that means the slot's catalog_xmin conflicts with the xid, and we need to handle the conflict. While subsequent commits will do the actual conflict handling, this commit adds a new field isCatalogRel in such WAL records (and a new bit set in the xl_heap_visible flags field), that is true for catalog tables, so as to arrange for conflict handling. The affected WAL records are the ones that already contain the snapshotConflictHorizon field, namely: - gistxlogDelete - gistxlogPageReuse - xl_hash_vacuum_one_page - xl_heap_prune - xl_heap_freeze_page - xl_heap_visible - xl_btree_reuse_page - xl_btree_delete - spgxlogVacuumRedirect Due to this new field being added, xl_hash_vacuum_one_page and gistxlogDelete do now contain the offsets to be deleted as a FLEXIBLE_ARRAY_MEMBER. This is needed to ensure correct alignment. It's not needed on the others struct where isCatalogRel has been added. This commit just introduces the WAL format changes mentioned above. Handling the actual conflicts will follow in future commits. Bumps XLOG_PAGE_MAGIC as the several WAL records are changed. Author: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Author: Andres Freund <andres@anarazel.de> (in an older version) Author: Amit Khandekar <amitdkhan.pg@gmail.com> (in an older version) Reviewed-by: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Reviewed-by: Andres Freund <andres@anarazel.de> Reviewed-by: Robert Haas <robertmhaas@gmail.com> Reviewed-by: Fabrízio de Royes Mello <fabriziomello@gmail.com> Reviewed-by: Melanie Plageman <melanieplageman@gmail.com>
2023-04-02 21:32:19 +02:00
Assert((xlrec->flags & VISIBILITYMAP_XLOG_VALID_BITS) == xlrec->flags);
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecGetBlockTag(record, 1, &rlocator, NULL, &blkno);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Prevent index-only scans from returning wrong answers under Hot Standby. The alternative of disallowing index-only scans in HS operation was discussed, but the consensus was that it was better to treat marking a page all-visible as a recovery conflict for snapshots that could still fail to see XIDs on that page. We may in the future try to soften this, so that we simply force index scans to do heap fetches in cases where this may be an issue, rather than throwing a hard conflict.
2012-04-27 02:00:21 +02:00
/*
* If there are any Hot Standby transactions running that have an xmin
* horizon old enough that this page isn't all-visible for them, they
* might incorrectly decide that an index-only scan can skip a heap fetch.
*
* NB: It might be better to throw some kind of "soft" conflict here that
* forces any index-only scan that is in flight to perform heap fetches,
* rather than killing the transaction outright.
*/
Only throw recovery conflicts when InHotStandby. Bug fix to recent patch to allow Index Only Scans on Hot Standby. Bug report from Jaime Casanova
2012-05-31 14:11:47 +02:00
if (InHotStandby)
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
ResolveRecoveryConflictWithSnapshot(xlrec->snapshotConflictHorizon,
Handle logical slot conflicts on standby During WAL replay on the standby, when a conflict with a logical slot is identified, invalidate such slots. There are two sources of conflicts: 1) Using the information added in 6af1793954e, logical slots are invalidated if required rows are removed 2) wal_level on the primary server is reduced to below logical Uses the infrastructure introduced in the prior commit. FIXME: add commit reference. Change InvalidatePossiblyObsoleteSlot() to use a recovery conflict to interrupt use of a slot, if called in the startup process. The new recovery conflict is added to pg_stat_database_conflicts, as confl_active_logicalslot. See 6af1793954e for an overall design of logical decoding on a standby. Bumps catversion for the addition of the pg_stat_database_conflicts column. Bumps PGSTAT_FILE_FORMAT_ID for the same reason. Author: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Author: Andres Freund <andres@anarazel.de> Author: Amit Khandekar <amitdkhan.pg@gmail.com> (in an older version) Reviewed-by: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Reviewed-by: Andres Freund <andres@anarazel.de> Reviewed-by: Robert Haas <robertmhaas@gmail.com> Reviewed-by: Fabrízio de Royes Mello <fabriziomello@gmail.com> Reviewed-by: Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com> Reviewed-by: Amit Kapila <amit.kapila16@gmail.com> Reviewed-by: Alvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/20230407075009.igg7be27ha2htkbt@awork3.anarazel.de
2023-04-08 08:11:28 +02:00
xlrec->flags & VISIBILITYMAP_XLOG_CATALOG_REL,
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
rlocator);
Prevent index-only scans from returning wrong answers under Hot Standby. The alternative of disallowing index-only scans in HS operation was discussed, but the consensus was that it was better to treat marking a page all-visible as a recovery conflict for snapshots that could still fail to see XIDs on that page. We may in the future try to soften this, so that we simply force index scans to do heap fetches in cases where this may be an issue, rather than throwing a hard conflict.
2012-04-27 02:00:21 +02:00
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
/*
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
* Read the heap page, if it still exists. If the heap file has dropped or
* truncated later in recovery, we don't need to update the page, but we'd
* better still update the visibility map.
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
*/
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
action = XLogReadBufferForRedo(record, 1, &buffer);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (action == BLK_NEEDS_REDO)
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
{
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
/*
* We don't bump the LSN of the heap page when setting the visibility
Fix a couple of bugs with wal_log_hints. 1. Replay of the WAL record for setting a bit in the visibility map contained an assertion that a full-page image of that record type can only occur with checksums enabled. But it can also happen with wal_log_hints, so remove the assertion. Unlike checksums, wal_log_hints can be changed on the fly, so it would be complicated to figure out if it was enabled at the time that the WAL record was generated. 2. wal_log_hints has the same effect on the locking needed to read the LSN of a page as data checksums. BufferGetLSNAtomic() didn't get the memo. Backpatch to 9.4, where wal_log_hints was added.
2015-06-26 11:38:24 +02:00
* map bit (unless checksums or wal_hint_bits is enabled, in which
Fix theoretical torn page hazard. The original report was concerned with a possible inconsistency between the heap and the visibility map, which I was unable to confirm. The concern has been retracted. However, there did seem to be a torn page hazard when using checksums. By not setting the heap page LSN during redo, the protections of minRecoveryPoint were bypassed. Fixed, along with a misleading comment. It may have been impossible to hit this problem in practice, because it would require a page tear between the checksum and the flags, so I am marking this as a theoretical risk. But, as discussed, it did violate expectations about the page LSN, so it may have other consequences. Backpatch to all supported versions. Reported-by: Konstantin Knizhnik Reviewed-by: Konstantin Knizhnik Discussion: https://postgr.es/m/fed17dac-8cb8-4f5b-d462-1bb4908c029e@garret.ru Backpatch-through: 11
2022-11-10 23:46:30 +01:00
* case we must). This exposes us to torn page hazards, but since
Fix a couple of bugs with wal_log_hints. 1. Replay of the WAL record for setting a bit in the visibility map contained an assertion that a full-page image of that record type can only occur with checksums enabled. But it can also happen with wal_log_hints, so remove the assertion. Unlike checksums, wal_log_hints can be changed on the fly, so it would be complicated to figure out if it was enabled at the time that the WAL record was generated. 2. wal_log_hints has the same effect on the locking needed to read the LSN of a page as data checksums. BufferGetLSNAtomic() didn't get the memo. Backpatch to 9.4, where wal_log_hints was added.
2015-06-26 11:38:24 +02:00
* we're not inspecting the existing page contents in any way, we
* don't care.
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
*/
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(buffer);
Change the format of the VM fork to add a second bit per page. The new bit indicates whether every tuple on the page is already frozen. It is cleared only when the all-visible bit is cleared, and it can be set only when we vacuum a page and find that every tuple on that page is both visible to every transaction and in no need of any future vacuuming. A future commit will use this new bit to optimize away full-table scans that would otherwise be triggered by XID wraparound considerations. A page which is merely all-visible must still be scanned in that case, but a page which is all-frozen need not be. This commit does not attempt that optimization, although that optimization is the goal here. It seems better to get the basic infrastructure in place first. Per discussion, it's very desirable for pg_upgrade to automatically migrate existing VM forks from the old format to the new format. That, too, will be handled in a follow-on patch. Masahiko Sawada, reviewed by Kyotaro Horiguchi, Fujii Masao, Amit Kapila, Simon Riggs, Andres Freund, and others, and substantially revised by me.
2016-03-02 03:49:41 +01:00
Department of second thoughts: remove PD_ALL_FROZEN. Commit a892234f830e832110f63fc0a2afce2fb21d1584 added a second bit per page to the visibility map, which still seems like a good idea, but it also added a second page-level bit alongside PD_ALL_VISIBLE to track whether the visibility map bit was set. That no longer seems like a clever plan, because we don't really need that bit for anything. We always clear both bits when the page is modified anyway. Patch by me, reviewed by Kyotaro Horiguchi and Masahiko Sawada.
2016-03-08 14:46:48 +01:00
PageSetAllVisible(page);
Change the format of the VM fork to add a second bit per page. The new bit indicates whether every tuple on the page is already frozen. It is cleared only when the all-visible bit is cleared, and it can be set only when we vacuum a page and find that every tuple on that page is both visible to every transaction and in no need of any future vacuuming. A future commit will use this new bit to optimize away full-table scans that would otherwise be triggered by XID wraparound considerations. A page which is merely all-visible must still be scanned in that case, but a page which is all-frozen need not be. This commit does not attempt that optimization, although that optimization is the goal here. It seems better to get the basic infrastructure in place first. Per discussion, it's very desirable for pg_upgrade to automatically migrate existing VM forks from the old format to the new format. That, too, will be handled in a follow-on patch. Masahiko Sawada, reviewed by Kyotaro Horiguchi, Fujii Masao, Amit Kapila, Simon Riggs, Andres Freund, and others, and substantially revised by me.
2016-03-02 03:49:41 +01:00
Fix theoretical torn page hazard. The original report was concerned with a possible inconsistency between the heap and the visibility map, which I was unable to confirm. The concern has been retracted. However, there did seem to be a torn page hazard when using checksums. By not setting the heap page LSN during redo, the protections of minRecoveryPoint were bypassed. Fixed, along with a misleading comment. It may have been impossible to hit this problem in practice, because it would require a page tear between the checksum and the flags, so I am marking this as a theoretical risk. But, as discussed, it did violate expectations about the page LSN, so it may have other consequences. Backpatch to all supported versions. Reported-by: Konstantin Knizhnik Reviewed-by: Konstantin Knizhnik Discussion: https://postgr.es/m/fed17dac-8cb8-4f5b-d462-1bb4908c029e@garret.ru Backpatch-through: 11
2022-11-10 23:46:30 +01:00
if (XLogHintBitIsNeeded())
PageSetLSN(page, lsn);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
MarkBufferDirty(buffer);
Allow I/O reliability checks using 16-bit checksums Checksums are set immediately prior to flush out of shared buffers and checked when pages are read in again. Hint bit setting will require full page write when block is dirtied, which causes various infrastructure changes. Extensive comments, docs and README. WARNING message thrown if checksum fails on non-all zeroes page; ERROR thrown but can be disabled with ignore_checksum_failure = on. Feature enabled by an initdb option, since transition from option off to option on is long and complex and has not yet been implemented. Default is not to use checksums. Checksum used is WAL CRC-32 truncated to 16-bits. Simon Riggs, Jeff Davis, Greg Smith Wide input and assistance from many community members. Thank you.
2013-03-22 14:54:07 +01:00
}
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
else if (action == BLK_RESTORED)
Allow I/O reliability checks using 16-bit checksums Checksums are set immediately prior to flush out of shared buffers and checked when pages are read in again. Hint bit setting will require full page write when block is dirtied, which causes various infrastructure changes. Extensive comments, docs and README. WARNING message thrown if checksum fails on non-all zeroes page; ERROR thrown but can be disabled with ignore_checksum_failure = on. Feature enabled by an initdb option, since transition from option off to option on is long and complex and has not yet been implemented. Default is not to use checksums. Checksum used is WAL CRC-32 truncated to 16-bits. Simon Riggs, Jeff Davis, Greg Smith Wide input and assistance from many community members. Thank you.
2013-03-22 14:54:07 +01:00
{
Fix multiple problems in WAL replay. Most of the replay functions for WAL record types that modify more than one page failed to ensure that those pages were locked correctly to ensure that concurrent queries could not see inconsistent page states. This is a hangover from coding decisions made long before Hot Standby was added, when it was hardly necessary to acquire buffer locks during WAL replay at all, let alone hold them for carefully-chosen periods. The key problem was that RestoreBkpBlocks was written to hold lock on each page restored from a full-page image for only as long as it took to update that page. This was guaranteed to break any WAL replay function in which there was any update-ordering constraint between pages, because even if the nominal order of the pages is the right one, any mixture of full-page and non-full-page updates in the same record would result in out-of-order updates. Moreover, it wouldn't work for situations where there's a requirement to maintain lock on one page while updating another. Failure to honor an update ordering constraint in this way is thought to be the cause of bug #7648 from Daniel Farina: what seems to have happened there is that a btree page being split was rewritten from a full-page image before the new right sibling page was written, and because lock on the original page was not maintained it was possible for hot standby queries to try to traverse the page's right-link to the not-yet-existing sibling page. To fix, get rid of RestoreBkpBlocks as such, and instead create a new function RestoreBackupBlock that restores just one full-page image at a time. This function can be invoked by WAL replay functions at the points where they would otherwise perform non-full-page updates; in this way, the physical order of page updates remains the same no matter which pages are replaced by full-page images. We can then further adjust the logic in individual replay functions if it is necessary to hold buffer locks for overlapping periods. A side benefit is that we can simplify the handling of concurrency conflict resolution by moving that code into the record-type-specfic functions; there's no more need to contort the code layout to keep conflict resolution in front of the RestoreBkpBlocks call. In connection with that, standardize on zero-based numbering rather than one-based numbering for referencing the full-page images. In HEAD, I removed the macros XLR_BKP_BLOCK_1 through XLR_BKP_BLOCK_4. They are still there in the header files in previous branches, but are no longer used by the code. In addition, fix some other bugs identified in the course of making these changes: spgRedoAddNode could fail to update the parent downlink at all, if the parent tuple is in the same page as either the old or new split tuple and we're not doing a full-page image: it would get fooled by the LSN having been advanced already. This would result in permanent index corruption, not just transient failure of concurrent queries. Also, ginHeapTupleFastInsert's "merge lists" case failed to mark the old tail page as a candidate for a full-page image; in the worst case this could result in torn-page corruption. heap_xlog_freeze() was inconsistent about using a cleanup lock or plain exclusive lock: it did the former in the normal path but the latter for a full-page image. A plain exclusive lock seems sufficient, so change to that. Also, remove gistRedoPageDeleteRecord(), which has been dead code since VACUUM FULL was rewritten. Back-patch to 9.0, where hot standby was introduced. Note however that 9.0 had a significantly different WAL-logging scheme for GIST index updates, and it doesn't appear possible to make that scheme safe for concurrent hot standby queries, because it can leave inconsistent states in the index even between WAL records. Given the lack of complaints from the field, we won't work too hard on fixing that branch.
2012-11-13 04:05:08 +01:00
/*
Fix a couple of bugs with wal_log_hints. 1. Replay of the WAL record for setting a bit in the visibility map contained an assertion that a full-page image of that record type can only occur with checksums enabled. But it can also happen with wal_log_hints, so remove the assertion. Unlike checksums, wal_log_hints can be changed on the fly, so it would be complicated to figure out if it was enabled at the time that the WAL record was generated. 2. wal_log_hints has the same effect on the locking needed to read the LSN of a page as data checksums. BufferGetLSNAtomic() didn't get the memo. Backpatch to 9.4, where wal_log_hints was added.
2015-06-26 11:38:24 +02:00
* If heap block was backed up, we already restored it and there's
* nothing more to do. (This can only happen with checksums or
* wal_log_hints enabled.)
Fix multiple problems in WAL replay. Most of the replay functions for WAL record types that modify more than one page failed to ensure that those pages were locked correctly to ensure that concurrent queries could not see inconsistent page states. This is a hangover from coding decisions made long before Hot Standby was added, when it was hardly necessary to acquire buffer locks during WAL replay at all, let alone hold them for carefully-chosen periods. The key problem was that RestoreBkpBlocks was written to hold lock on each page restored from a full-page image for only as long as it took to update that page. This was guaranteed to break any WAL replay function in which there was any update-ordering constraint between pages, because even if the nominal order of the pages is the right one, any mixture of full-page and non-full-page updates in the same record would result in out-of-order updates. Moreover, it wouldn't work for situations where there's a requirement to maintain lock on one page while updating another. Failure to honor an update ordering constraint in this way is thought to be the cause of bug #7648 from Daniel Farina: what seems to have happened there is that a btree page being split was rewritten from a full-page image before the new right sibling page was written, and because lock on the original page was not maintained it was possible for hot standby queries to try to traverse the page's right-link to the not-yet-existing sibling page. To fix, get rid of RestoreBkpBlocks as such, and instead create a new function RestoreBackupBlock that restores just one full-page image at a time. This function can be invoked by WAL replay functions at the points where they would otherwise perform non-full-page updates; in this way, the physical order of page updates remains the same no matter which pages are replaced by full-page images. We can then further adjust the logic in individual replay functions if it is necessary to hold buffer locks for overlapping periods. A side benefit is that we can simplify the handling of concurrency conflict resolution by moving that code into the record-type-specfic functions; there's no more need to contort the code layout to keep conflict resolution in front of the RestoreBkpBlocks call. In connection with that, standardize on zero-based numbering rather than one-based numbering for referencing the full-page images. In HEAD, I removed the macros XLR_BKP_BLOCK_1 through XLR_BKP_BLOCK_4. They are still there in the header files in previous branches, but are no longer used by the code. In addition, fix some other bugs identified in the course of making these changes: spgRedoAddNode could fail to update the parent downlink at all, if the parent tuple is in the same page as either the old or new split tuple and we're not doing a full-page image: it would get fooled by the LSN having been advanced already. This would result in permanent index corruption, not just transient failure of concurrent queries. Also, ginHeapTupleFastInsert's "merge lists" case failed to mark the old tail page as a candidate for a full-page image; in the worst case this could result in torn-page corruption. heap_xlog_freeze() was inconsistent about using a cleanup lock or plain exclusive lock: it did the former in the normal path but the latter for a full-page image. A plain exclusive lock seems sufficient, so change to that. Also, remove gistRedoPageDeleteRecord(), which has been dead code since VACUUM FULL was rewritten. Back-patch to 9.0, where hot standby was introduced. Note however that 9.0 had a significantly different WAL-logging scheme for GIST index updates, and it doesn't appear possible to make that scheme safe for concurrent hot standby queries, because it can leave inconsistent states in the index even between WAL records. Given the lack of complaints from the field, we won't work too hard on fixing that branch.
2012-11-13 04:05:08 +01:00
*/
}
Update FSM on WAL replay of page all-visible/frozen We aren't very strict about keeping FSM up to date on WAL replay, because per-page freespace values aren't critical in replicas (can't write to heap in a replica; and if the replica is promoted, the values would be updated by VACUUM anyway). However, VACUUM since 9.6 can skip processing pages marked all-visible or all-frozen, and if such pages are recorded in FSM with wrong values, those values are blindly propagated to FSM's upper layers by VACUUM's FreeSpaceMapVacuum. (This rationale assumes that crashes are not very frequent, because those would cause outdated FSM to occur in the primary.) Even when the FSM is outdated in standby, things are not too bad normally, because, most per-page FSM values will be zero (other than those propagated with the base-backup that created the standby); only once the remaining free space is less than 0.2*BLCKSZ the per-page value is maintained by WAL replay of heap ins/upd/del. However, if wal_log_hints=on causes complete FSM pages to be propagated to a standby via full-page images, many too-optimistic per-page values can end up being registered in the standby. Incorrect per-page values aren't critical in most cases, since an inserter that is given a page that doesn't actually contain the claimed free space will update FSM with the correct value, and retry until it finds a usable page. However, if there are many such updates to do, an inserter can spend a long time doing them before a usable page is found; in a heavily trafficked insert-only table with many concurrent inserters this has been observed to cause several second stalls, causing visible application malfunction. To fix this problem, it seems sufficient to have heap_xlog_visible (replay of setting all-visible and all-frozen VM bits for a heap page) update the FSM value for the page being processed. This fixes the per-page counters together with making the page skippable to vacuum, so when vacuum does FreeSpaceMapVacuum, the values propagated to FSM upper layers are the correct ones, avoiding the problem. While at it, apply the same fix to heap_xlog_clean (replay of tuple removal by HOT pruning and vacuum). This makes any space freed by the cleaning available earlier than the next vacuum in the promoted replica. Backpatch to 9.6, where this problem was diagnosed on an insert-only table with all-frozen pages, which were introduced as a concept in that release. Theoretically it could apply with all-visible pages to older branches, but there's been no report of that and it doesn't backpatch cleanly anyway. Author: Álvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/20180802172857.5skoexsilnjvgruk@alvherre.pgsql
2018-08-15 23:09:29 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (BufferIsValid(buffer))
Update FSM on WAL replay of page all-visible/frozen We aren't very strict about keeping FSM up to date on WAL replay, because per-page freespace values aren't critical in replicas (can't write to heap in a replica; and if the replica is promoted, the values would be updated by VACUUM anyway). However, VACUUM since 9.6 can skip processing pages marked all-visible or all-frozen, and if such pages are recorded in FSM with wrong values, those values are blindly propagated to FSM's upper layers by VACUUM's FreeSpaceMapVacuum. (This rationale assumes that crashes are not very frequent, because those would cause outdated FSM to occur in the primary.) Even when the FSM is outdated in standby, things are not too bad normally, because, most per-page FSM values will be zero (other than those propagated with the base-backup that created the standby); only once the remaining free space is less than 0.2*BLCKSZ the per-page value is maintained by WAL replay of heap ins/upd/del. However, if wal_log_hints=on causes complete FSM pages to be propagated to a standby via full-page images, many too-optimistic per-page values can end up being registered in the standby. Incorrect per-page values aren't critical in most cases, since an inserter that is given a page that doesn't actually contain the claimed free space will update FSM with the correct value, and retry until it finds a usable page. However, if there are many such updates to do, an inserter can spend a long time doing them before a usable page is found; in a heavily trafficked insert-only table with many concurrent inserters this has been observed to cause several second stalls, causing visible application malfunction. To fix this problem, it seems sufficient to have heap_xlog_visible (replay of setting all-visible and all-frozen VM bits for a heap page) update the FSM value for the page being processed. This fixes the per-page counters together with making the page skippable to vacuum, so when vacuum does FreeSpaceMapVacuum, the values propagated to FSM upper layers are the correct ones, avoiding the problem. While at it, apply the same fix to heap_xlog_clean (replay of tuple removal by HOT pruning and vacuum). This makes any space freed by the cleaning available earlier than the next vacuum in the promoted replica. Backpatch to 9.6, where this problem was diagnosed on an insert-only table with all-frozen pages, which were introduced as a concept in that release. Theoretically it could apply with all-visible pages to older branches, but there's been no report of that and it doesn't backpatch cleanly anyway. Author: Álvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/20180802172857.5skoexsilnjvgruk@alvherre.pgsql
2018-08-15 23:09:29 +02:00
{
Size space = PageGetFreeSpace(BufferGetPage(buffer));
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
UnlockReleaseBuffer(buffer);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Update FSM on WAL replay of page all-visible/frozen We aren't very strict about keeping FSM up to date on WAL replay, because per-page freespace values aren't critical in replicas (can't write to heap in a replica; and if the replica is promoted, the values would be updated by VACUUM anyway). However, VACUUM since 9.6 can skip processing pages marked all-visible or all-frozen, and if such pages are recorded in FSM with wrong values, those values are blindly propagated to FSM's upper layers by VACUUM's FreeSpaceMapVacuum. (This rationale assumes that crashes are not very frequent, because those would cause outdated FSM to occur in the primary.) Even when the FSM is outdated in standby, things are not too bad normally, because, most per-page FSM values will be zero (other than those propagated with the base-backup that created the standby); only once the remaining free space is less than 0.2*BLCKSZ the per-page value is maintained by WAL replay of heap ins/upd/del. However, if wal_log_hints=on causes complete FSM pages to be propagated to a standby via full-page images, many too-optimistic per-page values can end up being registered in the standby. Incorrect per-page values aren't critical in most cases, since an inserter that is given a page that doesn't actually contain the claimed free space will update FSM with the correct value, and retry until it finds a usable page. However, if there are many such updates to do, an inserter can spend a long time doing them before a usable page is found; in a heavily trafficked insert-only table with many concurrent inserters this has been observed to cause several second stalls, causing visible application malfunction. To fix this problem, it seems sufficient to have heap_xlog_visible (replay of setting all-visible and all-frozen VM bits for a heap page) update the FSM value for the page being processed. This fixes the per-page counters together with making the page skippable to vacuum, so when vacuum does FreeSpaceMapVacuum, the values propagated to FSM upper layers are the correct ones, avoiding the problem. While at it, apply the same fix to heap_xlog_clean (replay of tuple removal by HOT pruning and vacuum). This makes any space freed by the cleaning available earlier than the next vacuum in the promoted replica. Backpatch to 9.6, where this problem was diagnosed on an insert-only table with all-frozen pages, which were introduced as a concept in that release. Theoretically it could apply with all-visible pages to older branches, but there's been no report of that and it doesn't backpatch cleanly anyway. Author: Álvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/20180802172857.5skoexsilnjvgruk@alvherre.pgsql
2018-08-15 23:09:29 +02:00
/*
* Since FSM is not WAL-logged and only updated heuristically, it
* easily becomes stale in standbys. If the standby is later promoted
* and runs VACUUM, it will skip updating individual free space
* figures for pages that became all-visible (or all-frozen, depending
* on the vacuum mode,) which is troublesome when FreeSpaceMapVacuum
* propagates too optimistic free space values to upper FSM layers;
* later inserters try to use such pages only to find out that they
* are unusable. This can cause long stalls when there are many such
* pages.
*
* Forestall those problems by updating FSM's idea about a page that
* is becoming all-visible or all-frozen.
*
* Do this regardless of a full-page image being applied, since the
* FSM data is not in the page anyway.
*/
if (xlrec->flags & VISIBILITYMAP_VALID_BITS)
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecordPageWithFreeSpace(rlocator, blkno, space);
Update FSM on WAL replay of page all-visible/frozen We aren't very strict about keeping FSM up to date on WAL replay, because per-page freespace values aren't critical in replicas (can't write to heap in a replica; and if the replica is promoted, the values would be updated by VACUUM anyway). However, VACUUM since 9.6 can skip processing pages marked all-visible or all-frozen, and if such pages are recorded in FSM with wrong values, those values are blindly propagated to FSM's upper layers by VACUUM's FreeSpaceMapVacuum. (This rationale assumes that crashes are not very frequent, because those would cause outdated FSM to occur in the primary.) Even when the FSM is outdated in standby, things are not too bad normally, because, most per-page FSM values will be zero (other than those propagated with the base-backup that created the standby); only once the remaining free space is less than 0.2*BLCKSZ the per-page value is maintained by WAL replay of heap ins/upd/del. However, if wal_log_hints=on causes complete FSM pages to be propagated to a standby via full-page images, many too-optimistic per-page values can end up being registered in the standby. Incorrect per-page values aren't critical in most cases, since an inserter that is given a page that doesn't actually contain the claimed free space will update FSM with the correct value, and retry until it finds a usable page. However, if there are many such updates to do, an inserter can spend a long time doing them before a usable page is found; in a heavily trafficked insert-only table with many concurrent inserters this has been observed to cause several second stalls, causing visible application malfunction. To fix this problem, it seems sufficient to have heap_xlog_visible (replay of setting all-visible and all-frozen VM bits for a heap page) update the FSM value for the page being processed. This fixes the per-page counters together with making the page skippable to vacuum, so when vacuum does FreeSpaceMapVacuum, the values propagated to FSM upper layers are the correct ones, avoiding the problem. While at it, apply the same fix to heap_xlog_clean (replay of tuple removal by HOT pruning and vacuum). This makes any space freed by the cleaning available earlier than the next vacuum in the promoted replica. Backpatch to 9.6, where this problem was diagnosed on an insert-only table with all-frozen pages, which were introduced as a concept in that release. Theoretically it could apply with all-visible pages to older branches, but there's been no report of that and it doesn't backpatch cleanly anyway. Author: Álvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/20180802172857.5skoexsilnjvgruk@alvherre.pgsql
2018-08-15 23:09:29 +02:00
}
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
/*
Fix multiple problems in WAL replay. Most of the replay functions for WAL record types that modify more than one page failed to ensure that those pages were locked correctly to ensure that concurrent queries could not see inconsistent page states. This is a hangover from coding decisions made long before Hot Standby was added, when it was hardly necessary to acquire buffer locks during WAL replay at all, let alone hold them for carefully-chosen periods. The key problem was that RestoreBkpBlocks was written to hold lock on each page restored from a full-page image for only as long as it took to update that page. This was guaranteed to break any WAL replay function in which there was any update-ordering constraint between pages, because even if the nominal order of the pages is the right one, any mixture of full-page and non-full-page updates in the same record would result in out-of-order updates. Moreover, it wouldn't work for situations where there's a requirement to maintain lock on one page while updating another. Failure to honor an update ordering constraint in this way is thought to be the cause of bug #7648 from Daniel Farina: what seems to have happened there is that a btree page being split was rewritten from a full-page image before the new right sibling page was written, and because lock on the original page was not maintained it was possible for hot standby queries to try to traverse the page's right-link to the not-yet-existing sibling page. To fix, get rid of RestoreBkpBlocks as such, and instead create a new function RestoreBackupBlock that restores just one full-page image at a time. This function can be invoked by WAL replay functions at the points where they would otherwise perform non-full-page updates; in this way, the physical order of page updates remains the same no matter which pages are replaced by full-page images. We can then further adjust the logic in individual replay functions if it is necessary to hold buffer locks for overlapping periods. A side benefit is that we can simplify the handling of concurrency conflict resolution by moving that code into the record-type-specfic functions; there's no more need to contort the code layout to keep conflict resolution in front of the RestoreBkpBlocks call. In connection with that, standardize on zero-based numbering rather than one-based numbering for referencing the full-page images. In HEAD, I removed the macros XLR_BKP_BLOCK_1 through XLR_BKP_BLOCK_4. They are still there in the header files in previous branches, but are no longer used by the code. In addition, fix some other bugs identified in the course of making these changes: spgRedoAddNode could fail to update the parent downlink at all, if the parent tuple is in the same page as either the old or new split tuple and we're not doing a full-page image: it would get fooled by the LSN having been advanced already. This would result in permanent index corruption, not just transient failure of concurrent queries. Also, ginHeapTupleFastInsert's "merge lists" case failed to mark the old tail page as a candidate for a full-page image; in the worst case this could result in torn-page corruption. heap_xlog_freeze() was inconsistent about using a cleanup lock or plain exclusive lock: it did the former in the normal path but the latter for a full-page image. A plain exclusive lock seems sufficient, so change to that. Also, remove gistRedoPageDeleteRecord(), which has been dead code since VACUUM FULL was rewritten. Back-patch to 9.0, where hot standby was introduced. Note however that 9.0 had a significantly different WAL-logging scheme for GIST index updates, and it doesn't appear possible to make that scheme safe for concurrent hot standby queries, because it can leave inconsistent states in the index even between WAL records. Given the lack of complaints from the field, we won't work too hard on fixing that branch.
2012-11-13 04:05:08 +01:00
* Even if we skipped the heap page update due to the LSN interlock, it's
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
* still safe to update the visibility map. Any WAL record that clears
* the visibility map bit does so before checking the page LSN, so any
* bits that need to be cleared will still be cleared.
*/
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (XLogReadBufferForRedoExtended(record, 0, RBM_ZERO_ON_ERROR, false,
&vmbuffer) == BLK_NEEDS_REDO)
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
{
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
Page vmpage = BufferGetPage(vmbuffer);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Relation reln;
Add info in WAL records in preparation for logical slot conflict handling This commit only implements one prerequisite part for allowing logical decoding. The commit message contains an explanation of the overall design, which later commits will refer back to. Overall design: 1. We want to enable logical decoding on standbys, but replay of WAL from the primary might remove data that is needed by logical decoding, causing error(s) on the standby. To prevent those errors, a new replication conflict scenario needs to be addressed (as much as hot standby does). 2. Our chosen strategy for dealing with this type of replication slot is to invalidate logical slots for which needed data has been removed. 3. To do this we need the latestRemovedXid for each change, just as we do for physical replication conflicts, but we also need to know whether any particular change was to data that logical replication might access. That way, during WAL replay, we know when there is a risk of conflict and, if so, if there is a conflict. 4. We can't rely on the standby's relcache entries for this purpose in any way, because the startup process can't access catalog contents. 5. Therefore every WAL record that potentially removes data from the index or heap must carry a flag indicating whether or not it is one that might be accessed during logical decoding. Why do we need this for logical decoding on standby? First, let's forget about logical decoding on standby and recall that on a primary database, any catalog rows that may be needed by a logical decoding replication slot are not removed. This is done thanks to the catalog_xmin associated with the logical replication slot. But, with logical decoding on standby, in the following cases: - hot_standby_feedback is off - hot_standby_feedback is on but there is no a physical slot between the primary and the standby. Then, hot_standby_feedback will work, but only while the connection is alive (for example a node restart would break it) Then, the primary may delete system catalog rows that could be needed by the logical decoding on the standby (as it does not know about the catalog_xmin on the standby). So, it’s mandatory to identify those rows and invalidate the slots that may need them if any. Identifying those rows is the purpose of this commit. Implementation: When a WAL replay on standby indicates that a catalog table tuple is to be deleted by an xid that is greater than a logical slot's catalog_xmin, then that means the slot's catalog_xmin conflicts with the xid, and we need to handle the conflict. While subsequent commits will do the actual conflict handling, this commit adds a new field isCatalogRel in such WAL records (and a new bit set in the xl_heap_visible flags field), that is true for catalog tables, so as to arrange for conflict handling. The affected WAL records are the ones that already contain the snapshotConflictHorizon field, namely: - gistxlogDelete - gistxlogPageReuse - xl_hash_vacuum_one_page - xl_heap_prune - xl_heap_freeze_page - xl_heap_visible - xl_btree_reuse_page - xl_btree_delete - spgxlogVacuumRedirect Due to this new field being added, xl_hash_vacuum_one_page and gistxlogDelete do now contain the offsets to be deleted as a FLEXIBLE_ARRAY_MEMBER. This is needed to ensure correct alignment. It's not needed on the others struct where isCatalogRel has been added. This commit just introduces the WAL format changes mentioned above. Handling the actual conflicts will follow in future commits. Bumps XLOG_PAGE_MAGIC as the several WAL records are changed. Author: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Author: Andres Freund <andres@anarazel.de> (in an older version) Author: Amit Khandekar <amitdkhan.pg@gmail.com> (in an older version) Reviewed-by: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Reviewed-by: Andres Freund <andres@anarazel.de> Reviewed-by: Robert Haas <robertmhaas@gmail.com> Reviewed-by: Fabrízio de Royes Mello <fabriziomello@gmail.com> Reviewed-by: Melanie Plageman <melanieplageman@gmail.com>
2023-04-02 21:32:19 +02:00
uint8 vmbits;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
/* initialize the page if it was read as zeros */
if (PageIsNew(vmpage))
PageInit(vmpage, BLCKSZ, 0);
Add info in WAL records in preparation for logical slot conflict handling This commit only implements one prerequisite part for allowing logical decoding. The commit message contains an explanation of the overall design, which later commits will refer back to. Overall design: 1. We want to enable logical decoding on standbys, but replay of WAL from the primary might remove data that is needed by logical decoding, causing error(s) on the standby. To prevent those errors, a new replication conflict scenario needs to be addressed (as much as hot standby does). 2. Our chosen strategy for dealing with this type of replication slot is to invalidate logical slots for which needed data has been removed. 3. To do this we need the latestRemovedXid for each change, just as we do for physical replication conflicts, but we also need to know whether any particular change was to data that logical replication might access. That way, during WAL replay, we know when there is a risk of conflict and, if so, if there is a conflict. 4. We can't rely on the standby's relcache entries for this purpose in any way, because the startup process can't access catalog contents. 5. Therefore every WAL record that potentially removes data from the index or heap must carry a flag indicating whether or not it is one that might be accessed during logical decoding. Why do we need this for logical decoding on standby? First, let's forget about logical decoding on standby and recall that on a primary database, any catalog rows that may be needed by a logical decoding replication slot are not removed. This is done thanks to the catalog_xmin associated with the logical replication slot. But, with logical decoding on standby, in the following cases: - hot_standby_feedback is off - hot_standby_feedback is on but there is no a physical slot between the primary and the standby. Then, hot_standby_feedback will work, but only while the connection is alive (for example a node restart would break it) Then, the primary may delete system catalog rows that could be needed by the logical decoding on the standby (as it does not know about the catalog_xmin on the standby). So, it’s mandatory to identify those rows and invalidate the slots that may need them if any. Identifying those rows is the purpose of this commit. Implementation: When a WAL replay on standby indicates that a catalog table tuple is to be deleted by an xid that is greater than a logical slot's catalog_xmin, then that means the slot's catalog_xmin conflicts with the xid, and we need to handle the conflict. While subsequent commits will do the actual conflict handling, this commit adds a new field isCatalogRel in such WAL records (and a new bit set in the xl_heap_visible flags field), that is true for catalog tables, so as to arrange for conflict handling. The affected WAL records are the ones that already contain the snapshotConflictHorizon field, namely: - gistxlogDelete - gistxlogPageReuse - xl_hash_vacuum_one_page - xl_heap_prune - xl_heap_freeze_page - xl_heap_visible - xl_btree_reuse_page - xl_btree_delete - spgxlogVacuumRedirect Due to this new field being added, xl_hash_vacuum_one_page and gistxlogDelete do now contain the offsets to be deleted as a FLEXIBLE_ARRAY_MEMBER. This is needed to ensure correct alignment. It's not needed on the others struct where isCatalogRel has been added. This commit just introduces the WAL format changes mentioned above. Handling the actual conflicts will follow in future commits. Bumps XLOG_PAGE_MAGIC as the several WAL records are changed. Author: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Author: Andres Freund <andres@anarazel.de> (in an older version) Author: Amit Khandekar <amitdkhan.pg@gmail.com> (in an older version) Reviewed-by: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Reviewed-by: Andres Freund <andres@anarazel.de> Reviewed-by: Robert Haas <robertmhaas@gmail.com> Reviewed-by: Fabrízio de Royes Mello <fabriziomello@gmail.com> Reviewed-by: Melanie Plageman <melanieplageman@gmail.com>
2023-04-02 21:32:19 +02:00
/* remove VISIBILITYMAP_XLOG_* */
vmbits = xlrec->flags & VISIBILITYMAP_VALID_BITS;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
/*
Fix bogus comments Some comments mentioned XLogReplayBuffer, but there's no such function: that was an interim name for a function that got renamed to XLogReadBufferForRedo, before commit 2c03216d831160 was pushed.
2016-05-12 21:02:49 +02:00
* XLogReadBufferForRedoExtended locked the buffer. But
* visibilitymap_set will handle locking itself.
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
*/
LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
reln = CreateFakeRelcacheEntry(rlocator);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
visibilitymap_pin(reln, blkno, &vmbuffer);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Remove obsolete comments and code from prior to f8f4227976. XLogReadBufferForRedo() and XLogReadBufferForRedoExtended() only return BLK_NEEDS_REDO if the record LSN is greater than the page LSN, so the redo routine doesn't need to do the LSN check again. Discussion: https://postgr.es/m/0c37b80e62b1f3007d5a6d1292bd8fa0c275627a.camel@j-davis.com
2022-11-11 17:40:01 +01:00
visibilitymap_set(reln, blkno, InvalidBuffer, lsn, vmbuffer,
Add info in WAL records in preparation for logical slot conflict handling This commit only implements one prerequisite part for allowing logical decoding. The commit message contains an explanation of the overall design, which later commits will refer back to. Overall design: 1. We want to enable logical decoding on standbys, but replay of WAL from the primary might remove data that is needed by logical decoding, causing error(s) on the standby. To prevent those errors, a new replication conflict scenario needs to be addressed (as much as hot standby does). 2. Our chosen strategy for dealing with this type of replication slot is to invalidate logical slots for which needed data has been removed. 3. To do this we need the latestRemovedXid for each change, just as we do for physical replication conflicts, but we also need to know whether any particular change was to data that logical replication might access. That way, during WAL replay, we know when there is a risk of conflict and, if so, if there is a conflict. 4. We can't rely on the standby's relcache entries for this purpose in any way, because the startup process can't access catalog contents. 5. Therefore every WAL record that potentially removes data from the index or heap must carry a flag indicating whether or not it is one that might be accessed during logical decoding. Why do we need this for logical decoding on standby? First, let's forget about logical decoding on standby and recall that on a primary database, any catalog rows that may be needed by a logical decoding replication slot are not removed. This is done thanks to the catalog_xmin associated with the logical replication slot. But, with logical decoding on standby, in the following cases: - hot_standby_feedback is off - hot_standby_feedback is on but there is no a physical slot between the primary and the standby. Then, hot_standby_feedback will work, but only while the connection is alive (for example a node restart would break it) Then, the primary may delete system catalog rows that could be needed by the logical decoding on the standby (as it does not know about the catalog_xmin on the standby). So, it’s mandatory to identify those rows and invalidate the slots that may need them if any. Identifying those rows is the purpose of this commit. Implementation: When a WAL replay on standby indicates that a catalog table tuple is to be deleted by an xid that is greater than a logical slot's catalog_xmin, then that means the slot's catalog_xmin conflicts with the xid, and we need to handle the conflict. While subsequent commits will do the actual conflict handling, this commit adds a new field isCatalogRel in such WAL records (and a new bit set in the xl_heap_visible flags field), that is true for catalog tables, so as to arrange for conflict handling. The affected WAL records are the ones that already contain the snapshotConflictHorizon field, namely: - gistxlogDelete - gistxlogPageReuse - xl_hash_vacuum_one_page - xl_heap_prune - xl_heap_freeze_page - xl_heap_visible - xl_btree_reuse_page - xl_btree_delete - spgxlogVacuumRedirect Due to this new field being added, xl_hash_vacuum_one_page and gistxlogDelete do now contain the offsets to be deleted as a FLEXIBLE_ARRAY_MEMBER. This is needed to ensure correct alignment. It's not needed on the others struct where isCatalogRel has been added. This commit just introduces the WAL format changes mentioned above. Handling the actual conflicts will follow in future commits. Bumps XLOG_PAGE_MAGIC as the several WAL records are changed. Author: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Author: Andres Freund <andres@anarazel.de> (in an older version) Author: Amit Khandekar <amitdkhan.pg@gmail.com> (in an older version) Reviewed-by: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Reviewed-by: Andres Freund <andres@anarazel.de> Reviewed-by: Robert Haas <robertmhaas@gmail.com> Reviewed-by: Fabrízio de Royes Mello <fabriziomello@gmail.com> Reviewed-by: Melanie Plageman <melanieplageman@gmail.com>
2023-04-02 21:32:19 +02:00
xlrec->snapshotConflictHorizon, vmbits);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
ReleaseBuffer(vmbuffer);
FreeFakeRelcacheEntry(reln);
}
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
else if (BufferIsValid(vmbuffer))
UnlockReleaseBuffer(vmbuffer);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
}
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
/*
* Replay XLOG_HEAP2_FREEZE_PAGE records
*/
static void
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_freeze_page(XLogReaderState *record)
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRecPtr lsn = record->EndRecPtr;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
xl_heap_freeze_page *xlrec = (xl_heap_freeze_page *) XLogRecGetData(record);
Buffer buffer;
/*
* In Hot Standby mode, ensure that there's no queries running which still
* consider the frozen xids as running.
*/
if (InHotStandby)
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
{
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
RelFileLocator rlocator;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecGetBlockTag(record, 0, &rlocator, NULL, NULL);
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
ResolveRecoveryConflictWithSnapshot(xlrec->snapshotConflictHorizon,
Handle logical slot conflicts on standby During WAL replay on the standby, when a conflict with a logical slot is identified, invalidate such slots. There are two sources of conflicts: 1) Using the information added in 6af1793954e, logical slots are invalidated if required rows are removed 2) wal_level on the primary server is reduced to below logical Uses the infrastructure introduced in the prior commit. FIXME: add commit reference. Change InvalidatePossiblyObsoleteSlot() to use a recovery conflict to interrupt use of a slot, if called in the startup process. The new recovery conflict is added to pg_stat_database_conflicts, as confl_active_logicalslot. See 6af1793954e for an overall design of logical decoding on a standby. Bumps catversion for the addition of the pg_stat_database_conflicts column. Bumps PGSTAT_FILE_FORMAT_ID for the same reason. Author: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Author: Andres Freund <andres@anarazel.de> Author: Amit Khandekar <amitdkhan.pg@gmail.com> (in an older version) Reviewed-by: "Drouvot, Bertrand" <bertranddrouvot.pg@gmail.com> Reviewed-by: Andres Freund <andres@anarazel.de> Reviewed-by: Robert Haas <robertmhaas@gmail.com> Reviewed-by: Fabrízio de Royes Mello <fabriziomello@gmail.com> Reviewed-by: Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com> Reviewed-by: Amit Kapila <amit.kapila16@gmail.com> Reviewed-by: Alvaro Herrera <alvherre@alvh.no-ip.org> Discussion: https://postgr.es/m/20230407075009.igg7be27ha2htkbt@awork3.anarazel.de
2023-04-08 08:11:28 +02:00
xlrec->isCatalogRel,
Standardize rmgrdesc recovery conflict XID output. Standardize on the name snapshotConflictHorizon for all XID fields from WAL records that generate recovery conflicts when in hot standby mode. This supersedes the previous latestRemovedXid naming convention. The new naming convention places emphasis on how the values are actually used by REDO routines. How the values are generated during original execution (details of which vary by record type) is deemphasized. Users of tools like pg_waldump can now grep for snapshotConflictHorizon to see all potential sources of recovery conflicts in a standardized way, without necessarily having to consider which specific record types might be involved. Also bring a couple of WAL record types that didn't follow any kind of naming convention into line. These are heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type. Now every WAL record whose REDO routine calls ResolveRecoveryConflictWithSnapshot() passes through the snapshotConflictHorizon field from its WAL record. This is follow-up work to the refactoring from commit 9e540599 that made FREEZE_PAGE WAL records use a standard snapshotConflictHorizon style XID cutoff. No bump in XLOG_PAGE_MAGIC, since the underlying format of affected WAL records doesn't change. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/CAH2-Wzm2CQUmViUq7Opgk=McVREHSOorYaAjR1ZpLYkRN7_dPw@mail.gmail.com
2022-11-17 23:55:08 +01:00
rlocator);
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
}
if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
{
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
Page page = BufferGetPage(buffer);
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
xl_heap_freeze_plan *plans;
OffsetNumber *offsets;
int curoff = 0;
plans = (xl_heap_freeze_plan *) XLogRecGetBlockData(record, 0, NULL);
offsets = (OffsetNumber *) ((char *) plans +
(xlrec->nplans *
sizeof(xl_heap_freeze_plan)));
for (int p = 0; p < xlrec->nplans; p++)
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
{
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
HeapTupleFreeze frz;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
/*
* Convert freeze plan representation from WAL record into
* per-tuple format used by heap_execute_freeze_tuple
*/
Rename and relocate freeze plan dedup routines. Rename the heapam.c freeze plan deduplication routines added by commit 9e540599 to names that follow conventions for functions in heapam.c. Also relocate the functions so that they're next to their caller, which runs during original execution, when FREEZE_PAGE WAL records are built. The routines were initially placed next to (and followed the naming conventions of) conceptually related REDO routine code, but that scheme turned out to be kind of jarring when considered in a wider context. Author: Peter Geoghegan <pg@bowt.ie> Reported-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/20230109214308.icz26oqvt3k2274c@awork3.anarazel.de
2023-01-12 02:30:42 +01:00
frz.xmax = plans[p].xmax;
frz.t_infomask2 = plans[p].t_infomask2;
frz.t_infomask = plans[p].t_infomask;
frz.frzflags = plans[p].frzflags;
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
frz.offset = InvalidOffsetNumber; /* unused, but be tidy */
Rename and relocate freeze plan dedup routines. Rename the heapam.c freeze plan deduplication routines added by commit 9e540599 to names that follow conventions for functions in heapam.c. Also relocate the functions so that they're next to their caller, which runs during original execution, when FREEZE_PAGE WAL records are built. The routines were initially placed next to (and followed the naming conventions of) conceptually related REDO routine code, but that scheme turned out to be kind of jarring when considered in a wider context. Author: Peter Geoghegan <pg@bowt.ie> Reported-By: Andres Freund <andres@anarazel.de> Discussion: https://postgr.es/m/20230109214308.icz26oqvt3k2274c@awork3.anarazel.de
2023-01-12 02:30:42 +01:00
for (int i = 0; i < plans[p].ntuples; i++)
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
{
OffsetNumber offset = offsets[curoff++];
ItemId lp;
HeapTupleHeader tuple;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
Deduplicate freeze plans in freeze WAL records. Make heapam WAL records that describe freezing performed by VACUUM more space efficient by storing each distinct "freeze plan" once, alongside an array of associated page offset numbers (one per freeze plan). The freeze plans required for most heap pages tend to naturally have a great deal of redundancy, so this technique is very effective in practice. It often leads to freeze WAL records that are less than 20% of the size of equivalent WAL records generated using the previous approach. The freeze plan concept was introduced by commit 3b97e6823b, which fixed bugs in VACUUM's handling of MultiXacts. We retain the concept of freeze plans, but go back to using page offset number arrays. There is no loss of generality here because deduplication is an additive process that gets applied mechanically when FREEZE_PAGE WAL records are built. More than anything else, freeze plan deduplication is an optimization that reduces the marginal cost of freezing additional tuples on pages that will need to have at least one or two tuples frozen in any case. Ongoing work that adds page-level freezing to VACUUM will take full advantage of the improved cost profile through batching. Also refactor some of the details surrounding recovery conflicts needed to REDO freeze records in passing: make original execution responsible for generating a standard latestRemovedXid cutoff, rather than working backwards to get the same cutoff in the REDO routine. Bugfix commit 66fbcb0d2e did it the other way around, which is equivalent but obscures what's going on. Also rename the cutoff field from the WAL record/struct (rename the field cutoff_xid to latestRemovedXid to match similar WAL records). Processing of conflicts by REDO routines is already completely uniform, so tools like pg_waldump should present the information driving the process uniformly. There are two remaining WAL record types that still don't quite follow this convention (heapam's VISIBLE record type and SP-GiST's VACUUM_REDIRECT record type). They can be brought into line by later work that totally standardizes how the cutoffs are presented. Bump XLOG_PAGE_MAGIC. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Reviewed-By: Nathan Bossart <nathandbossart@gmail.com> Reviewed-By: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CAH2-Wz=XytErMnb8FAyFd+OQEbiipB0Q2FmFdXrggPL4VBnRYQ@mail.gmail.com
2022-11-15 16:48:41 +01:00
lp = PageGetItemId(page, offset);
tuple = (HeapTupleHeader) PageGetItem(page, lp);
heap_execute_freeze_tuple(tuple, &frz);
}
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
}
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageSetLSN(page, lsn);
MarkBufferDirty(buffer);
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
}
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (BufferIsValid(buffer))
UnlockReleaseBuffer(buffer);
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
/*
* Given an "infobits" field from an XLog record, set the correct bits in the
* given infomask and infomask2 for the tuple touched by the record.
*
* (This is the reverse of compute_infobits).
*/
static void
fix_infomask_from_infobits(uint8 infobits, uint16 *infomask, uint16 *infomask2)
{
*infomask &= ~(HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY |
HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_EXCL_LOCK);
*infomask2 &= ~HEAP_KEYS_UPDATED;
if (infobits & XLHL_XMAX_IS_MULTI)
*infomask |= HEAP_XMAX_IS_MULTI;
if (infobits & XLHL_XMAX_LOCK_ONLY)
*infomask |= HEAP_XMAX_LOCK_ONLY;
if (infobits & XLHL_XMAX_EXCL_LOCK)
*infomask |= HEAP_XMAX_EXCL_LOCK;
/* note HEAP_XMAX_SHR_LOCK isn't considered here */
if (infobits & XLHL_XMAX_KEYSHR_LOCK)
*infomask |= HEAP_XMAX_KEYSHR_LOCK;
if (infobits & XLHL_KEYS_UPDATED)
*infomask2 |= HEAP_KEYS_UPDATED;
}
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
static void
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_delete(XLogReaderState *record)
unlock buffer before releasing in heap_insert +unlock buffer in heap_fetch under !ItemIdIsUsed().
2000-07-04 03:39:24 +02:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRecPtr lsn = record->EndRecPtr;
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
xl_heap_delete *xlrec = (xl_heap_delete *) XLogRecGetData(record);
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
Buffer buffer;
Page page;
Silence gcc warnings.
2000-11-20 22:14:13 +01:00
ItemId lp = NULL;
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
HeapTupleHeader htup;
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
BlockNumber blkno;
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
RelFileLocator target_locator;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
ItemPointerData target_tid;
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecGetBlockTag(record, 0, &target_locator, NULL, &blkno);
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
ItemPointerSetBlockNumber(&target_tid, blkno);
ItemPointerSetOffsetNumber(&target_tid, xlrec->offnum);
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
/*
Fix a violation of WAL coding rules in the recent patch to include an "all tuples visible" flag in heap page headers. The flag update *must* be applied before calling XLogInsert, but heap_update and the tuple moving routines in VACUUM FULL were ignoring this rule. A crash and replay could therefore leave the flag incorrectly set, causing rows to appear visible in seqscans when they should not be. This might explain recent reports of data corruption from Jeff Ross and others. In passing, do a bit of editorialization on comments in visibilitymap.c.
2009-08-24 04:18:32 +02:00
* The visibility map may need to be fixed even if the heap page is
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
* already up-to-date.
*/
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
if (xlrec->flags & XLH_DELETE_ALL_VISIBLE_CLEARED)
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
{
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
Relation reln = CreateFakeRelcacheEntry(target_locator);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Buffer vmbuffer = InvalidBuffer;
8.4 pgindent run, with new combined Linux/FreeBSD/MinGW typedef list provided by Andrew.
2009-06-11 16:49:15 +02:00
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
visibilitymap_pin(reln, blkno, &vmbuffer);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
visibilitymap_clear(reln, blkno, vmbuffer, VISIBILITYMAP_VALID_BITS);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
ReleaseBuffer(vmbuffer);
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
FreeFakeRelcacheEntry(reln);
}
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
Fix multiple problems in WAL replay. Most of the replay functions for WAL record types that modify more than one page failed to ensure that those pages were locked correctly to ensure that concurrent queries could not see inconsistent page states. This is a hangover from coding decisions made long before Hot Standby was added, when it was hardly necessary to acquire buffer locks during WAL replay at all, let alone hold them for carefully-chosen periods. The key problem was that RestoreBkpBlocks was written to hold lock on each page restored from a full-page image for only as long as it took to update that page. This was guaranteed to break any WAL replay function in which there was any update-ordering constraint between pages, because even if the nominal order of the pages is the right one, any mixture of full-page and non-full-page updates in the same record would result in out-of-order updates. Moreover, it wouldn't work for situations where there's a requirement to maintain lock on one page while updating another. Failure to honor an update ordering constraint in this way is thought to be the cause of bug #7648 from Daniel Farina: what seems to have happened there is that a btree page being split was rewritten from a full-page image before the new right sibling page was written, and because lock on the original page was not maintained it was possible for hot standby queries to try to traverse the page's right-link to the not-yet-existing sibling page. To fix, get rid of RestoreBkpBlocks as such, and instead create a new function RestoreBackupBlock that restores just one full-page image at a time. This function can be invoked by WAL replay functions at the points where they would otherwise perform non-full-page updates; in this way, the physical order of page updates remains the same no matter which pages are replaced by full-page images. We can then further adjust the logic in individual replay functions if it is necessary to hold buffer locks for overlapping periods. A side benefit is that we can simplify the handling of concurrency conflict resolution by moving that code into the record-type-specfic functions; there's no more need to contort the code layout to keep conflict resolution in front of the RestoreBkpBlocks call. In connection with that, standardize on zero-based numbering rather than one-based numbering for referencing the full-page images. In HEAD, I removed the macros XLR_BKP_BLOCK_1 through XLR_BKP_BLOCK_4. They are still there in the header files in previous branches, but are no longer used by the code. In addition, fix some other bugs identified in the course of making these changes: spgRedoAddNode could fail to update the parent downlink at all, if the parent tuple is in the same page as either the old or new split tuple and we're not doing a full-page image: it would get fooled by the LSN having been advanced already. This would result in permanent index corruption, not just transient failure of concurrent queries. Also, ginHeapTupleFastInsert's "merge lists" case failed to mark the old tail page as a candidate for a full-page image; in the worst case this could result in torn-page corruption. heap_xlog_freeze() was inconsistent about using a cleanup lock or plain exclusive lock: it did the former in the normal path but the latter for a full-page image. A plain exclusive lock seems sufficient, so change to that. Also, remove gistRedoPageDeleteRecord(), which has been dead code since VACUUM FULL was rewritten. Back-patch to 9.0, where hot standby was introduced. Note however that 9.0 had a significantly different WAL-logging scheme for GIST index updates, and it doesn't appear possible to make that scheme safe for concurrent hot standby queries, because it can leave inconsistent states in the index even between WAL records. Given the lack of complaints from the field, we won't work too hard on fixing that branch.
2012-11-13 04:05:08 +01:00
{
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(buffer);
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (PageGetMaxOffsetNumber(page) >= xlrec->offnum)
lp = PageGetItemId(page, xlrec->offnum);
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (PageGetMaxOffsetNumber(page) < xlrec->offnum || !ItemIdIsNormal(lp))
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(PANIC, "invalid lp");
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
htup = (HeapTupleHeader) PageGetItem(page, lp);
WAL fixes.
2000-10-29 19:33:41 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
htup->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
htup->t_infomask2 &= ~HEAP_KEYS_UPDATED;
HeapTupleHeaderClearHotUpdated(htup);
fix_infomask_from_infobits(xlrec->infobits_set,
&htup->t_infomask, &htup->t_infomask2);
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
if (!(xlrec->flags & XLH_DELETE_IS_SUPER))
HeapTupleHeaderSetXmax(htup, xlrec->xmax);
else
HeapTupleHeaderSetXmin(htup, InvalidTransactionId);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
HeapTupleHeaderSetCmax(htup, FirstCommandId, false);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
/* Mark the page as a candidate for pruning */
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
PageSetPrunable(page, XLogRecGetXid(record));
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
if (xlrec->flags & XLH_DELETE_ALL_VISIBLE_CLEARED)
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageClearAllVisible(page);
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
Raise error when affecting tuple moved into different partition. When an update moves a row between partitions (supported since 2f178441044b), our normal logic for following update chains in READ COMMITTED mode doesn't work anymore. Cross partition updates are modeled as an delete from the old and insert into the new partition. No ctid chain exists across partitions, and there's no convenient space to introduce that link. Not throwing an error in a partitioned context when one would have been thrown without partitioning is obviously problematic. This commit introduces infrastructure to detect when a tuple has been moved, not just plainly deleted. That allows to throw an error when encountering a deletion that's actually a move, while attempting to following a ctid chain. The row deleted as part of a cross partition update is marked by pointing it's t_ctid to an invalid block, instead of self as a normal update would. That was deemed to be the least invasive and most future proof way to represent the knowledge, given how few infomask bits are there to be recycled (there's also some locking issues with using infomask bits). External code following ctid chains should be updated to check for moved tuples. The most likely consequence of not doing so is a missed error. Author: Amul Sul, editorialized by me Reviewed-By: Amit Kapila, Pavan Deolasee, Andres Freund, Robert Haas Discussion: http://postgr.es/m/CAAJ_b95PkwojoYfz0bzXU8OokcTVGzN6vYGCNVUukeUDrnF3dw@mail.gmail.com
2018-04-07 22:24:10 +02:00
/* Make sure t_ctid is set correctly */
if (xlrec->flags & XLH_DELETE_IS_PARTITION_MOVE)
HeapTupleHeaderSetMovedPartitions(htup);
else
htup->t_ctid = target_tid;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageSetLSN(page, lsn);
MarkBufferDirty(buffer);
}
if (BufferIsValid(buffer))
UnlockReleaseBuffer(buffer);
unlock buffer before releasing in heap_insert +unlock buffer in heap_fetch under !ItemIdIsUsed().
2000-07-04 03:39:24 +02:00
}
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
static void
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_insert(XLogReaderState *record)
unlock buffer before releasing in heap_insert +unlock buffer in heap_fetch under !ItemIdIsUsed().
2000-07-04 03:39:24 +02:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRecPtr lsn = record->EndRecPtr;
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
xl_heap_insert *xlrec = (xl_heap_insert *) XLogRecGetData(record);
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
Buffer buffer;
Page page;
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
union
Remove the mostly-stubbed-out-anyway support routines for WAL UNDO. That code is never going to be used in the foreseeable future, and where it's more than a stub it's making the redo routines harder to read.
2005-06-06 19:01:25 +02:00
{
HeapTupleHeaderData hdr;
Rename MaxTupleSize to MaxHeapTupleSize to clarify that it's not meant to describe the maximum size of index tuples (which is typically AM-dependent anyway); and consequently remove the bogus deduction for "special space" that was built into it. Adjust TOAST_TUPLE_THRESHOLD and TOAST_MAX_CHUNK_SIZE to avoid wasting two bytes per toast chunk, and to ensure that the calculation correctly tracks any future changes in page header size. The computation had been inaccurate in a way that didn't cause any harm except space wastage, but future changes could have broken it more drastically. Fix the calculation of BTMaxItemSize, which was formerly computed as 1 byte more than it could safely be. This didn't cause any harm in practice because it's only compared against maxalign'd lengths, but future changes in the size of page headers or btree special space could have exposed the problem. initdb forced because of change in TOAST_MAX_CHUNK_SIZE, which alters the storage of toast tables.
2007-02-05 05:22:18 +01:00
char data[MaxHeapTupleSize];
Remove the mostly-stubbed-out-anyway support routines for WAL UNDO. That code is never going to be used in the foreseeable future, and where it's more than a stub it's making the redo routines harder to read.
2005-06-06 19:01:25 +02:00
} tbuf;
HeapTupleHeader htup;
xl_heap_header xlhdr;
uint32 newlen;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
Size freespace = 0;
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
RelFileLocator target_locator;
Update FSM on WAL replay. This is a bit limited; the FSM is only updated on non-full-page-image WAL records, and quite arbitrarily, only if there's less than 20% free space on the page after the insert/update (not on HOT updates, though). The 20% cutoff should avoid most of the overhead, when replaying a bulk insertion, for example, while ensuring that pages that are full are marked as full in the FSM. This is mostly to avoid the nasty worst case scenario, where you replay from a PITR archive, and the FSM information in the base backup is really out of date. If there was a lot of pages that the outdated FSM claims to have free space, but don't actually have any, the first unlucky inserter after the recovery would traverse through all those pages, just to find out that they're full. We didn't have this problem with the old FSM implementation, because we simply threw the FSM information away on a non-clean shutdown.
2008-10-31 20:40:27 +01:00
BlockNumber blkno;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
ItemPointerData target_tid;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
XLogRedoAction action;
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecGetBlockTag(record, 0, &target_locator, NULL, &blkno);
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
ItemPointerSetBlockNumber(&target_tid, blkno);
ItemPointerSetOffsetNumber(&target_tid, xlrec->offnum);
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
/*
Fix a violation of WAL coding rules in the recent patch to include an "all tuples visible" flag in heap page headers. The flag update *must* be applied before calling XLogInsert, but heap_update and the tuple moving routines in VACUUM FULL were ignoring this rule. A crash and replay could therefore leave the flag incorrectly set, causing rows to appear visible in seqscans when they should not be. This might explain recent reports of data corruption from Jeff Ross and others. In passing, do a bit of editorialization on comments in visibilitymap.c.
2009-08-24 04:18:32 +02:00
* The visibility map may need to be fixed even if the heap page is
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
* already up-to-date.
*/
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
if (xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED)
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
{
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
Relation reln = CreateFakeRelcacheEntry(target_locator);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Buffer vmbuffer = InvalidBuffer;
8.4 pgindent run, with new combined Linux/FreeBSD/MinGW typedef list provided by Andrew.
2009-06-11 16:49:15 +02:00
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
visibilitymap_pin(reln, blkno, &vmbuffer);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
visibilitymap_clear(reln, blkno, vmbuffer, VISIBILITYMAP_VALID_BITS);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
ReleaseBuffer(vmbuffer);
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
FreeFakeRelcacheEntry(reln);
}
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
/*
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
* If we inserted the first and only tuple on the page, re-initialize the
* page from scratch.
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
*/
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (XLogRecGetInfo(record) & XLOG_HEAP_INIT_PAGE)
Clean up and document the API for XLogOpenRelation and XLogReadBuffer. This commit doesn't make much functional change, but it does eliminate some duplicated code --- for instance, PageIsNew tests are now done inside XLogReadBuffer rather than by each caller. The GIST xlog code still needs a lot of love, but I'll worry about that separately.
2006-03-29 23:17:39 +02:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
buffer = XLogInitBufferForRedo(record, 0);
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(buffer);
Clean up and document the API for XLogOpenRelation and XLogReadBuffer. This commit doesn't make much functional change, but it does eliminate some duplicated code --- for instance, PageIsNew tests are now done inside XLogReadBuffer rather than by each caller. The GIST xlog code still needs a lot of love, but I'll worry about that separately.
2006-03-29 23:17:39 +02:00
PageInit(page, BufferGetPageSize(buffer), 0);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
action = BLK_NEEDS_REDO;
Clean up and document the API for XLogOpenRelation and XLogReadBuffer. This commit doesn't make much functional change, but it does eliminate some duplicated code --- for instance, PageIsNew tests are now done inside XLogReadBuffer rather than by each caller. The GIST xlog code still needs a lot of love, but I'll worry about that separately.
2006-03-29 23:17:39 +02:00
}
else
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
action = XLogReadBufferForRedo(record, 0, &buffer);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (action == BLK_NEEDS_REDO)
Remove the mostly-stubbed-out-anyway support routines for WAL UNDO. That code is never going to be used in the foreseeable future, and where it's more than a stub it's making the redo routines harder to read.
2005-06-06 19:01:25 +02:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
Size datalen;
char *data;
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(buffer);
Clean up and document the API for XLogOpenRelation and XLogReadBuffer. This commit doesn't make much functional change, but it does eliminate some duplicated code --- for instance, PageIsNew tests are now done inside XLogReadBuffer rather than by each caller. The GIST xlog code still needs a lot of love, but I'll worry about that separately.
2006-03-29 23:17:39 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (PageGetMaxOffsetNumber(page) + 1 < xlrec->offnum)
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(PANIC, "invalid max offset number");
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
data = XLogRecGetBlockData(record, 0, &datalen);
newlen = datalen - SizeOfHeapHeader;
Assert(datalen > SizeOfHeapHeader && newlen <= MaxHeapTupleSize);
memcpy((char *) &xlhdr, data, SizeOfHeapHeader);
data += SizeOfHeapHeader;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
htup = &tbuf.hdr;
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
MemSet((char *) htup, 0, SizeofHeapTupleHeader);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
/* PG73FORMAT: get bitmap [+ padding] [+ oid] + data */
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
memcpy((char *) htup + SizeofHeapTupleHeader,
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
data,
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
newlen);
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
newlen += SizeofHeapTupleHeader;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
htup->t_infomask2 = xlhdr.t_infomask2;
htup->t_infomask = xlhdr.t_infomask;
htup->t_hoff = xlhdr.t_hoff;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
HeapTupleHeaderSetXmin(htup, XLogRecGetXid(record));
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
HeapTupleHeaderSetCmin(htup, FirstCommandId);
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
htup->t_ctid = target_tid;
Remove the mostly-stubbed-out-anyway support routines for WAL UNDO. That code is never going to be used in the foreseeable future, and where it's more than a stub it's making the redo routines harder to read.
2005-06-06 19:01:25 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (PageAddItem(page, (Item) htup, newlen, xlrec->offnum,
true, true) == InvalidOffsetNumber)
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(PANIC, "failed to add tuple");
Update FSM on WAL replay. This is a bit limited; the FSM is only updated on non-full-page-image WAL records, and quite arbitrarily, only if there's less than 20% free space on the page after the insert/update (not on HOT updates, though). The 20% cutoff should avoid most of the overhead, when replaying a bulk insertion, for example, while ensuring that pages that are full are marked as full in the FSM. This is mostly to avoid the nasty worst case scenario, where you replay from a PITR archive, and the FSM information in the base backup is really out of date. If there was a lot of pages that the outdated FSM claims to have free space, but don't actually have any, the first unlucky inserter after the recovery would traverse through all those pages, just to find out that they're full. We didn't have this problem with the old FSM implementation, because we simply threw the FSM information away on a non-clean shutdown.
2008-10-31 20:40:27 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
freespace = PageGetHeapFreeSpace(page); /* needed to update FSM below */
Update FSM on WAL replay. This is a bit limited; the FSM is only updated on non-full-page-image WAL records, and quite arbitrarily, only if there's less than 20% free space on the page after the insert/update (not on HOT updates, though). The 20% cutoff should avoid most of the overhead, when replaying a bulk insertion, for example, while ensuring that pages that are full are marked as full in the FSM. This is mostly to avoid the nasty worst case scenario, where you replay from a PITR archive, and the FSM information in the base backup is really out of date. If there was a lot of pages that the outdated FSM claims to have free space, but don't actually have any, the first unlucky inserter after the recovery would traverse through all those pages, just to find out that they're full. We didn't have this problem with the old FSM implementation, because we simply threw the FSM information away on a non-clean shutdown.
2008-10-31 20:40:27 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageSetLSN(page, lsn);
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
if (xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED)
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageClearAllVisible(page);
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
Set PD_ALL_VISIBLE and visibility map bits in COPY FREEZE Make sure COPY FREEZE marks the pages as PD_ALL_VISIBLE and updates the visibility map. Until now we only marked individual tuples as frozen, but page-level flags were not updated, so the first VACUUM after the COPY FREEZE had to rewrite the whole table. This is a fairly old patch, and multiple people worked on it. The first version was written by Jeff Janes, and then reworked by Pavan Deolasee and Anastasia Lubennikova. Author: Anastasia Lubennikova, Pavan Deolasee, Jeff Janes Reviewed-by: Kuntal Ghosh, Jeff Janes, Tomas Vondra, Masahiko Sawada, Andres Freund, Ibrar Ahmed, Robert Haas, Tatsuro Ishii, Darafei Praliaskouski Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com Discussion: https://postgr.es/m/CAMkU%3D1w3osJJ2FneELhhNRLxfZitDgp9FPHee08NT2FQFmz_pQ%40mail.gmail.com
2021-01-17 22:11:39 +01:00
/* XLH_INSERT_ALL_FROZEN_SET implies that all tuples are visible */
if (xlrec->flags & XLH_INSERT_ALL_FROZEN_SET)
PageSetAllVisible(page);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
MarkBufferDirty(buffer);
}
if (BufferIsValid(buffer))
UnlockReleaseBuffer(buffer);
Update FSM on WAL replay. This is a bit limited; the FSM is only updated on non-full-page-image WAL records, and quite arbitrarily, only if there's less than 20% free space on the page after the insert/update (not on HOT updates, though). The 20% cutoff should avoid most of the overhead, when replaying a bulk insertion, for example, while ensuring that pages that are full are marked as full in the FSM. This is mostly to avoid the nasty worst case scenario, where you replay from a PITR archive, and the FSM information in the base backup is really out of date. If there was a lot of pages that the outdated FSM claims to have free space, but don't actually have any, the first unlucky inserter after the recovery would traverse through all those pages, just to find out that they're full. We didn't have this problem with the old FSM implementation, because we simply threw the FSM information away on a non-clean shutdown.
2008-10-31 20:40:27 +01:00
/*
* If the page is running low on free space, update the FSM as well.
* Arbitrarily, our definition of "low" is less than 20%. We can't do much
* better than that without knowing the fill-factor for the table.
*
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
* XXX: Don't do this if the page was restored from full page image. We
* don't bother to update the FSM in that case, it doesn't need to be
Update FSM on WAL replay. This is a bit limited; the FSM is only updated on non-full-page-image WAL records, and quite arbitrarily, only if there's less than 20% free space on the page after the insert/update (not on HOT updates, though). The 20% cutoff should avoid most of the overhead, when replaying a bulk insertion, for example, while ensuring that pages that are full are marked as full in the FSM. This is mostly to avoid the nasty worst case scenario, where you replay from a PITR archive, and the FSM information in the base backup is really out of date. If there was a lot of pages that the outdated FSM claims to have free space, but don't actually have any, the first unlucky inserter after the recovery would traverse through all those pages, just to find out that they're full. We didn't have this problem with the old FSM implementation, because we simply threw the FSM information away on a non-clean shutdown.
2008-10-31 20:40:27 +01:00
* totally accurate anyway.
*/
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (action == BLK_NEEDS_REDO && freespace < BLCKSZ / 5)
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecordPageWithFreeSpace(target_locator, blkno, freespace);
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
}
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
/*
* Handles MULTI_INSERT record type.
*/
static void
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_multi_insert(XLogReaderState *record)
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRecPtr lsn = record->EndRecPtr;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
xl_heap_multi_insert *xlrec;
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
RelFileLocator rlocator;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
BlockNumber blkno;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
Buffer buffer;
Page page;
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
union
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
{
HeapTupleHeaderData hdr;
char data[MaxHeapTupleSize];
} tbuf;
HeapTupleHeader htup;
uint32 newlen;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
Size freespace = 0;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
int i;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
bool isinit = (XLogRecGetInfo(record) & XLOG_HEAP_INIT_PAGE) != 0;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
XLogRedoAction action;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
Fix bugs in the COPY heap-insert batching patch. Forgot to call RestoreBkpBlocks() in the redo-function, as pointed out by Simon Riggs. In redo of a regular heap insert, it's taken care of in heap_redo(), but this new record type uses the heap2 RM, and heap2_redo() does not take care of that for you. Also, failed to reset the vmbuffer and all_visibile_cleared local variables after switching to a new buffer.
2011-11-09 20:24:26 +01:00
/*
* Insertion doesn't overwrite MVCC data, so no conflict processing is
* required.
*/
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
xlrec = (xl_heap_multi_insert *) XLogRecGetData(record);
Fix bugs in the COPY heap-insert batching patch. Forgot to call RestoreBkpBlocks() in the redo-function, as pointed out by Simon Riggs. In redo of a regular heap insert, it's taken care of in heap_redo(), but this new record type uses the heap2 RM, and heap2_redo() does not take care of that for you. Also, failed to reset the vmbuffer and all_visibile_cleared local variables after switching to a new buffer.
2011-11-09 20:24:26 +01:00
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecGetBlockTag(record, 0, &rlocator, NULL, &blkno);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
Set PD_ALL_VISIBLE and visibility map bits in COPY FREEZE Make sure COPY FREEZE marks the pages as PD_ALL_VISIBLE and updates the visibility map. Until now we only marked individual tuples as frozen, but page-level flags were not updated, so the first VACUUM after the COPY FREEZE had to rewrite the whole table. This is a fairly old patch, and multiple people worked on it. The first version was written by Jeff Janes, and then reworked by Pavan Deolasee and Anastasia Lubennikova. Author: Anastasia Lubennikova, Pavan Deolasee, Jeff Janes Reviewed-by: Kuntal Ghosh, Jeff Janes, Tomas Vondra, Masahiko Sawada, Andres Freund, Ibrar Ahmed, Robert Haas, Tatsuro Ishii, Darafei Praliaskouski Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com Discussion: https://postgr.es/m/CAMkU%3D1w3osJJ2FneELhhNRLxfZitDgp9FPHee08NT2FQFmz_pQ%40mail.gmail.com
2021-01-17 22:11:39 +01:00
/* check that the mutually exclusive flags are not both set */
Assert(!((xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED) &&
(xlrec->flags & XLH_INSERT_ALL_FROZEN_SET)));
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
/*
* The visibility map may need to be fixed even if the heap page is
* already up-to-date.
*/
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
if (xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED)
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
{
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
Relation reln = CreateFakeRelcacheEntry(rlocator);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
Buffer vmbuffer = InvalidBuffer;
visibilitymap_pin(reln, blkno, &vmbuffer);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
visibilitymap_clear(reln, blkno, vmbuffer, VISIBILITYMAP_VALID_BITS);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
ReleaseBuffer(vmbuffer);
FreeFakeRelcacheEntry(reln);
}
if (isinit)
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
buffer = XLogInitBufferForRedo(record, 0);
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(buffer);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
PageInit(page, BufferGetPageSize(buffer), 0);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
action = BLK_NEEDS_REDO;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
}
else
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
action = XLogReadBufferForRedo(record, 0, &buffer);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (action == BLK_NEEDS_REDO)
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
char *tupdata;
char *endptr;
Size len;
/* Tuples are stored as block data */
tupdata = XLogRecGetBlockData(record, 0, &len);
endptr = tupdata + len;
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = (Page) BufferGetPage(buffer);
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
for (i = 0; i < xlrec->ntuples; i++)
{
OffsetNumber offnum;
xl_multi_insert_tuple *xlhdr;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
/*
* If we're reinitializing the page, the tuples are stored in
* order from FirstOffsetNumber. Otherwise there's an array of
* offsets in the WAL record, and the tuples come after that.
*/
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (isinit)
offnum = FirstOffsetNumber + i;
else
offnum = xlrec->offsets[i];
if (PageGetMaxOffsetNumber(page) + 1 < offnum)
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(PANIC, "invalid max offset number");
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
xlhdr = (xl_multi_insert_tuple *) SHORTALIGN(tupdata);
tupdata = ((char *) xlhdr) + SizeOfMultiInsertTuple;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
newlen = xlhdr->datalen;
Assert(newlen <= MaxHeapTupleSize);
htup = &tbuf.hdr;
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
MemSet((char *) htup, 0, SizeofHeapTupleHeader);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
/* PG73FORMAT: get bitmap [+ padding] [+ oid] + data */
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
memcpy((char *) htup + SizeofHeapTupleHeader,
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
(char *) tupdata,
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
newlen);
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
tupdata += newlen;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
newlen += SizeofHeapTupleHeader;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
htup->t_infomask2 = xlhdr->t_infomask2;
htup->t_infomask = xlhdr->t_infomask;
htup->t_hoff = xlhdr->t_hoff;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
HeapTupleHeaderSetXmin(htup, XLogRecGetXid(record));
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
HeapTupleHeaderSetCmin(htup, FirstCommandId);
ItemPointerSetBlockNumber(&htup->t_ctid, blkno);
ItemPointerSetOffsetNumber(&htup->t_ctid, offnum);
offnum = PageAddItem(page, (Item) htup, newlen, offnum, true, true);
if (offnum == InvalidOffsetNumber)
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(PANIC, "failed to add tuple");
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
}
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (tupdata != endptr)
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(PANIC, "total tuple length mismatch");
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
freespace = PageGetHeapFreeSpace(page); /* needed to update FSM below */
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageSetLSN(page, lsn);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
if (xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED)
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageClearAllVisible(page);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
Fix COPY FREEZE with CLOBBER_CACHE_ALWAYS This adds code omitted from commit 7db0cd2145 by accident, which had two consequences. Firstly, only rows inserted by heap_multi_insert were frozen as expected when running COPY FREEZE, while heap_insert left rows unfrozen. That however includes rows in TOAST tables, so a lot of data might have been left unfrozen. Secondly, page might have been left partially empty after relcache invalidation. This addresses both of those issues. Discussion: https://postgr.es/m/CABOikdN-ptGv0mZntrK2Q8OtfUuAjqaYMGmkdU1dCKFtUxVLrg@mail.gmail.com
2021-01-24 00:24:50 +01:00
/* XLH_INSERT_ALL_FROZEN_SET implies that all tuples are visible */
if (xlrec->flags & XLH_INSERT_ALL_FROZEN_SET)
PageSetAllVisible(page);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
MarkBufferDirty(buffer);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
}
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (BufferIsValid(buffer))
UnlockReleaseBuffer(buffer);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
/*
* If the page is running low on free space, update the FSM as well.
* Arbitrarily, our definition of "low" is less than 20%. We can't do much
* better than that without knowing the fill-factor for the table.
*
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
* XXX: Don't do this if the page was restored from full page image. We
* don't bother to update the FSM in that case, it doesn't need to be
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
* totally accurate anyway.
*/
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (action == BLK_NEEDS_REDO && freespace < BLCKSZ / 5)
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecordPageWithFreeSpace(rlocator, blkno, freespace);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
}
WAL misc
2000-10-24 11:56:23 +02:00
/*
Remove old-style VACUUM FULL (which was known for a little while as VACUUM FULL INPLACE), along with a boatload of subsidiary code and complexity. Per discussion, the use case for this method of vacuuming is no longer large enough to justify maintaining it; not to mention that we don't wish to invest the work that would be needed to make it play nicely with Hot Standby. Aside from the code directly related to old-style VACUUM FULL, this commit removes support for certain WAL record types that could only be generated within VACUUM FULL, redirect-pointer removal in heap_page_prune, and nontransactional generation of cache invalidation sinval messages (the last being the sticking point for Hot Standby). We still have to retain all code that copes with finding HEAP_MOVED_OFF and HEAP_MOVED_IN flag bits on existing tuples. This can't be removed as long as we want to support in-place update from pre-9.0 databases.
2010-02-08 05:33:55 +01:00
* Handles UPDATE and HOT_UPDATE
WAL misc
2000-10-24 11:56:23 +02:00
*/
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
static void
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_update(XLogReaderState *record, bool hot_update)
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRecPtr lsn = record->EndRecPtr;
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
xl_heap_update *xlrec = (xl_heap_update *) XLogRecGetData(record);
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
RelFileLocator rlocator;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
BlockNumber oldblk;
BlockNumber newblk;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
ItemPointerData newtid;
Fix multiple problems in WAL replay. Most of the replay functions for WAL record types that modify more than one page failed to ensure that those pages were locked correctly to ensure that concurrent queries could not see inconsistent page states. This is a hangover from coding decisions made long before Hot Standby was added, when it was hardly necessary to acquire buffer locks during WAL replay at all, let alone hold them for carefully-chosen periods. The key problem was that RestoreBkpBlocks was written to hold lock on each page restored from a full-page image for only as long as it took to update that page. This was guaranteed to break any WAL replay function in which there was any update-ordering constraint between pages, because even if the nominal order of the pages is the right one, any mixture of full-page and non-full-page updates in the same record would result in out-of-order updates. Moreover, it wouldn't work for situations where there's a requirement to maintain lock on one page while updating another. Failure to honor an update ordering constraint in this way is thought to be the cause of bug #7648 from Daniel Farina: what seems to have happened there is that a btree page being split was rewritten from a full-page image before the new right sibling page was written, and because lock on the original page was not maintained it was possible for hot standby queries to try to traverse the page's right-link to the not-yet-existing sibling page. To fix, get rid of RestoreBkpBlocks as such, and instead create a new function RestoreBackupBlock that restores just one full-page image at a time. This function can be invoked by WAL replay functions at the points where they would otherwise perform non-full-page updates; in this way, the physical order of page updates remains the same no matter which pages are replaced by full-page images. We can then further adjust the logic in individual replay functions if it is necessary to hold buffer locks for overlapping periods. A side benefit is that we can simplify the handling of concurrency conflict resolution by moving that code into the record-type-specfic functions; there's no more need to contort the code layout to keep conflict resolution in front of the RestoreBkpBlocks call. In connection with that, standardize on zero-based numbering rather than one-based numbering for referencing the full-page images. In HEAD, I removed the macros XLR_BKP_BLOCK_1 through XLR_BKP_BLOCK_4. They are still there in the header files in previous branches, but are no longer used by the code. In addition, fix some other bugs identified in the course of making these changes: spgRedoAddNode could fail to update the parent downlink at all, if the parent tuple is in the same page as either the old or new split tuple and we're not doing a full-page image: it would get fooled by the LSN having been advanced already. This would result in permanent index corruption, not just transient failure of concurrent queries. Also, ginHeapTupleFastInsert's "merge lists" case failed to mark the old tail page as a candidate for a full-page image; in the worst case this could result in torn-page corruption. heap_xlog_freeze() was inconsistent about using a cleanup lock or plain exclusive lock: it did the former in the normal path but the latter for a full-page image. A plain exclusive lock seems sufficient, so change to that. Also, remove gistRedoPageDeleteRecord(), which has been dead code since VACUUM FULL was rewritten. Back-patch to 9.0, where hot standby was introduced. Note however that 9.0 had a significantly different WAL-logging scheme for GIST index updates, and it doesn't appear possible to make that scheme safe for concurrent hot standby queries, because it can leave inconsistent states in the index even between WAL records. Given the lack of complaints from the field, we won't work too hard on fixing that branch.
2012-11-13 04:05:08 +01:00
Buffer obuffer,
nbuffer;
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
Page page;
OffsetNumber offnum;
Silence gcc warnings.
2000-11-20 22:14:13 +01:00
ItemId lp = NULL;
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
HeapTupleData oldtup;
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
HeapTupleHeader htup;
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
uint16 prefixlen = 0,
suffixlen = 0;
char *newp;
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
union
Remove the mostly-stubbed-out-anyway support routines for WAL UNDO. That code is never going to be used in the foreseeable future, and where it's more than a stub it's making the redo routines harder to read.
2005-06-06 19:01:25 +02:00
{
HeapTupleHeaderData hdr;
Rename MaxTupleSize to MaxHeapTupleSize to clarify that it's not meant to describe the maximum size of index tuples (which is typically AM-dependent anyway); and consequently remove the bogus deduction for "special space" that was built into it. Adjust TOAST_TUPLE_THRESHOLD and TOAST_MAX_CHUNK_SIZE to avoid wasting two bytes per toast chunk, and to ensure that the calculation correctly tracks any future changes in page header size. The computation had been inaccurate in a way that didn't cause any harm except space wastage, but future changes could have broken it more drastically. Fix the calculation of BTMaxItemSize, which was formerly computed as 1 byte more than it could safely be. This didn't cause any harm in practice because it's only compared against maxalign'd lengths, but future changes in the size of page headers or btree special space could have exposed the problem. initdb forced because of change in TOAST_MAX_CHUNK_SIZE, which alters the storage of toast tables.
2007-02-05 05:22:18 +01:00
char data[MaxHeapTupleSize];
Remove the mostly-stubbed-out-anyway support routines for WAL UNDO. That code is never going to be used in the foreseeable future, and where it's more than a stub it's making the redo routines harder to read.
2005-06-06 19:01:25 +02:00
} tbuf;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
xl_heap_header xlhdr;
Remove the mostly-stubbed-out-anyway support routines for WAL UNDO. That code is never going to be used in the foreseeable future, and where it's more than a stub it's making the redo routines harder to read.
2005-06-06 19:01:25 +02:00
uint32 newlen;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
Size freespace = 0;
XLogRedoAction oldaction;
XLogRedoAction newaction;
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
/* initialize to keep the compiler quiet */
oldtup.t_data = NULL;
oldtup.t_len = 0;
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecGetBlockTag(record, 0, &rlocator, NULL, &newblk);
Make XLogRecGetBlockTag() throw error if there's no such block. All but a few existing callers assume without checking that this function succeeds. While it probably will, that's a poor excuse for not checking. Let's make it return void and instead throw an error if it doesn't find the block reference. Callers that actually need to handle the no-such-block case must now use the underlying function XLogRecGetBlockTagExtended. In addition to being a bit less error-prone, this should also serve to suppress some Coverity complaints about XLogRecGetBlockRefInfo. While at it, clean up some inconsistency about use of the XLogRecHasBlockRef macro: make XLogRecGetBlockTagExtended use that instead of open-coding the same condition, and avoid calling XLogRecHasBlockRef twice in relevant code paths. (That is, calling XLogRecHasBlockRef followed by XLogRecGetBlockTag is now deprecated: use XLogRecGetBlockTagExtended instead.) Patch HEAD only; this doesn't seem to have enough value to consider a back-branch API break. Discussion: https://postgr.es/m/425039.1649701221@sss.pgh.pa.us
2022-04-11 23:43:46 +02:00
if (XLogRecGetBlockTagExtended(record, 1, NULL, NULL, &oldblk, NULL))
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
{
/* HOT updates are never done across pages */
Assert(!hot_update);
}
else
oldblk = newblk;
ItemPointerSet(&newtid, newblk, xlrec->new_offnum);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
/*
Fix a violation of WAL coding rules in the recent patch to include an "all tuples visible" flag in heap page headers. The flag update *must* be applied before calling XLogInsert, but heap_update and the tuple moving routines in VACUUM FULL were ignoring this rule. A crash and replay could therefore leave the flag incorrectly set, causing rows to appear visible in seqscans when they should not be. This might explain recent reports of data corruption from Jeff Ross and others. In passing, do a bit of editorialization on comments in visibilitymap.c.
2009-08-24 04:18:32 +02:00
* The visibility map may need to be fixed even if the heap page is
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
* already up-to-date.
*/
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
if (xlrec->flags & XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED)
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
{
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
Relation reln = CreateFakeRelcacheEntry(rlocator);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Buffer vmbuffer = InvalidBuffer;
8.4 pgindent run, with new combined Linux/FreeBSD/MinGW typedef list provided by Andrew.
2009-06-11 16:49:15 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
visibilitymap_pin(reln, oldblk, &vmbuffer);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
visibilitymap_clear(reln, oldblk, vmbuffer, VISIBILITYMAP_VALID_BITS);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
ReleaseBuffer(vmbuffer);
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
FreeFakeRelcacheEntry(reln);
}
Fix multiple problems in WAL replay. Most of the replay functions for WAL record types that modify more than one page failed to ensure that those pages were locked correctly to ensure that concurrent queries could not see inconsistent page states. This is a hangover from coding decisions made long before Hot Standby was added, when it was hardly necessary to acquire buffer locks during WAL replay at all, let alone hold them for carefully-chosen periods. The key problem was that RestoreBkpBlocks was written to hold lock on each page restored from a full-page image for only as long as it took to update that page. This was guaranteed to break any WAL replay function in which there was any update-ordering constraint between pages, because even if the nominal order of the pages is the right one, any mixture of full-page and non-full-page updates in the same record would result in out-of-order updates. Moreover, it wouldn't work for situations where there's a requirement to maintain lock on one page while updating another. Failure to honor an update ordering constraint in this way is thought to be the cause of bug #7648 from Daniel Farina: what seems to have happened there is that a btree page being split was rewritten from a full-page image before the new right sibling page was written, and because lock on the original page was not maintained it was possible for hot standby queries to try to traverse the page's right-link to the not-yet-existing sibling page. To fix, get rid of RestoreBkpBlocks as such, and instead create a new function RestoreBackupBlock that restores just one full-page image at a time. This function can be invoked by WAL replay functions at the points where they would otherwise perform non-full-page updates; in this way, the physical order of page updates remains the same no matter which pages are replaced by full-page images. We can then further adjust the logic in individual replay functions if it is necessary to hold buffer locks for overlapping periods. A side benefit is that we can simplify the handling of concurrency conflict resolution by moving that code into the record-type-specfic functions; there's no more need to contort the code layout to keep conflict resolution in front of the RestoreBkpBlocks call. In connection with that, standardize on zero-based numbering rather than one-based numbering for referencing the full-page images. In HEAD, I removed the macros XLR_BKP_BLOCK_1 through XLR_BKP_BLOCK_4. They are still there in the header files in previous branches, but are no longer used by the code. In addition, fix some other bugs identified in the course of making these changes: spgRedoAddNode could fail to update the parent downlink at all, if the parent tuple is in the same page as either the old or new split tuple and we're not doing a full-page image: it would get fooled by the LSN having been advanced already. This would result in permanent index corruption, not just transient failure of concurrent queries. Also, ginHeapTupleFastInsert's "merge lists" case failed to mark the old tail page as a candidate for a full-page image; in the worst case this could result in torn-page corruption. heap_xlog_freeze() was inconsistent about using a cleanup lock or plain exclusive lock: it did the former in the normal path but the latter for a full-page image. A plain exclusive lock seems sufficient, so change to that. Also, remove gistRedoPageDeleteRecord(), which has been dead code since VACUUM FULL was rewritten. Back-patch to 9.0, where hot standby was introduced. Note however that 9.0 had a significantly different WAL-logging scheme for GIST index updates, and it doesn't appear possible to make that scheme safe for concurrent hot standby queries, because it can leave inconsistent states in the index even between WAL records. Given the lack of complaints from the field, we won't work too hard on fixing that branch.
2012-11-13 04:05:08 +01:00
/*
* In normal operation, it is important to lock the two pages in
* page-number order, to avoid possible deadlocks against other update
* operations going the other way. However, during WAL replay there can
* be no other update happening, so we don't need to worry about that. But
* we *do* need to worry that we don't expose an inconsistent state to Hot
* Standby queries --- so the original page can't be unlocked before we've
* added the new tuple to the new page.
*/
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
/* Deal with old tuple version */
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
oldaction = XLogReadBufferForRedo(record, (oldblk == newblk) ? 0 : 1,
&obuffer);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (oldaction == BLK_NEEDS_REDO)
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
{
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(obuffer);
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
offnum = xlrec->old_offnum;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (PageGetMaxOffsetNumber(page) >= offnum)
lp = PageGetItemId(page, offnum);
New WAL version - CRC and data blocks backup.
2000-12-28 14:00:29 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(PANIC, "invalid lp");
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
htup = (HeapTupleHeader) PageGetItem(page, lp);
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
oldtup.t_data = htup;
oldtup.t_len = ItemIdGetLength(lp);
pgindent run for 8.2.
2006-10-04 02:30:14 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
htup->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
htup->t_infomask2 &= ~HEAP_KEYS_UPDATED;
if (hot_update)
HeapTupleHeaderSetHotUpdated(htup);
else
HeapTupleHeaderClearHotUpdated(htup);
fix_infomask_from_infobits(xlrec->old_infobits_set, &htup->t_infomask,
&htup->t_infomask2);
HeapTupleHeaderSetXmax(htup, xlrec->old_xmax);
HeapTupleHeaderSetCmax(htup, FirstCommandId, false);
/* Set forward chain link in t_ctid */
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
htup->t_ctid = newtid;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
/* Mark the page as a candidate for pruning */
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
PageSetPrunable(page, XLogRecGetXid(record));
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
if (xlrec->flags & XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED)
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageClearAllVisible(page);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageSetLSN(page, lsn);
MarkBufferDirty(obuffer);
}
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
Clean up and document the API for XLogOpenRelation and XLogReadBuffer. This commit doesn't make much functional change, but it does eliminate some duplicated code --- for instance, PageIsNew tests are now done inside XLogReadBuffer rather than by each caller. The GIST xlog code still needs a lot of love, but I'll worry about that separately.
2006-03-29 23:17:39 +02:00
/*
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
* Read the page the new tuple goes into, if different from old.
Clean up and document the API for XLogOpenRelation and XLogReadBuffer. This commit doesn't make much functional change, but it does eliminate some duplicated code --- for instance, PageIsNew tests are now done inside XLogReadBuffer rather than by each caller. The GIST xlog code still needs a lot of love, but I'll worry about that separately.
2006-03-29 23:17:39 +02:00
*/
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (oldblk == newblk)
Fix multiple problems in WAL replay. Most of the replay functions for WAL record types that modify more than one page failed to ensure that those pages were locked correctly to ensure that concurrent queries could not see inconsistent page states. This is a hangover from coding decisions made long before Hot Standby was added, when it was hardly necessary to acquire buffer locks during WAL replay at all, let alone hold them for carefully-chosen periods. The key problem was that RestoreBkpBlocks was written to hold lock on each page restored from a full-page image for only as long as it took to update that page. This was guaranteed to break any WAL replay function in which there was any update-ordering constraint between pages, because even if the nominal order of the pages is the right one, any mixture of full-page and non-full-page updates in the same record would result in out-of-order updates. Moreover, it wouldn't work for situations where there's a requirement to maintain lock on one page while updating another. Failure to honor an update ordering constraint in this way is thought to be the cause of bug #7648 from Daniel Farina: what seems to have happened there is that a btree page being split was rewritten from a full-page image before the new right sibling page was written, and because lock on the original page was not maintained it was possible for hot standby queries to try to traverse the page's right-link to the not-yet-existing sibling page. To fix, get rid of RestoreBkpBlocks as such, and instead create a new function RestoreBackupBlock that restores just one full-page image at a time. This function can be invoked by WAL replay functions at the points where they would otherwise perform non-full-page updates; in this way, the physical order of page updates remains the same no matter which pages are replaced by full-page images. We can then further adjust the logic in individual replay functions if it is necessary to hold buffer locks for overlapping periods. A side benefit is that we can simplify the handling of concurrency conflict resolution by moving that code into the record-type-specfic functions; there's no more need to contort the code layout to keep conflict resolution in front of the RestoreBkpBlocks call. In connection with that, standardize on zero-based numbering rather than one-based numbering for referencing the full-page images. In HEAD, I removed the macros XLR_BKP_BLOCK_1 through XLR_BKP_BLOCK_4. They are still there in the header files in previous branches, but are no longer used by the code. In addition, fix some other bugs identified in the course of making these changes: spgRedoAddNode could fail to update the parent downlink at all, if the parent tuple is in the same page as either the old or new split tuple and we're not doing a full-page image: it would get fooled by the LSN having been advanced already. This would result in permanent index corruption, not just transient failure of concurrent queries. Also, ginHeapTupleFastInsert's "merge lists" case failed to mark the old tail page as a candidate for a full-page image; in the worst case this could result in torn-page corruption. heap_xlog_freeze() was inconsistent about using a cleanup lock or plain exclusive lock: it did the former in the normal path but the latter for a full-page image. A plain exclusive lock seems sufficient, so change to that. Also, remove gistRedoPageDeleteRecord(), which has been dead code since VACUUM FULL was rewritten. Back-patch to 9.0, where hot standby was introduced. Note however that 9.0 had a significantly different WAL-logging scheme for GIST index updates, and it doesn't appear possible to make that scheme safe for concurrent hot standby queries, because it can leave inconsistent states in the index even between WAL records. Given the lack of complaints from the field, we won't work too hard on fixing that branch.
2012-11-13 04:05:08 +01:00
{
nbuffer = obuffer;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
newaction = oldaction;
Fix multiple problems in WAL replay. Most of the replay functions for WAL record types that modify more than one page failed to ensure that those pages were locked correctly to ensure that concurrent queries could not see inconsistent page states. This is a hangover from coding decisions made long before Hot Standby was added, when it was hardly necessary to acquire buffer locks during WAL replay at all, let alone hold them for carefully-chosen periods. The key problem was that RestoreBkpBlocks was written to hold lock on each page restored from a full-page image for only as long as it took to update that page. This was guaranteed to break any WAL replay function in which there was any update-ordering constraint between pages, because even if the nominal order of the pages is the right one, any mixture of full-page and non-full-page updates in the same record would result in out-of-order updates. Moreover, it wouldn't work for situations where there's a requirement to maintain lock on one page while updating another. Failure to honor an update ordering constraint in this way is thought to be the cause of bug #7648 from Daniel Farina: what seems to have happened there is that a btree page being split was rewritten from a full-page image before the new right sibling page was written, and because lock on the original page was not maintained it was possible for hot standby queries to try to traverse the page's right-link to the not-yet-existing sibling page. To fix, get rid of RestoreBkpBlocks as such, and instead create a new function RestoreBackupBlock that restores just one full-page image at a time. This function can be invoked by WAL replay functions at the points where they would otherwise perform non-full-page updates; in this way, the physical order of page updates remains the same no matter which pages are replaced by full-page images. We can then further adjust the logic in individual replay functions if it is necessary to hold buffer locks for overlapping periods. A side benefit is that we can simplify the handling of concurrency conflict resolution by moving that code into the record-type-specfic functions; there's no more need to contort the code layout to keep conflict resolution in front of the RestoreBkpBlocks call. In connection with that, standardize on zero-based numbering rather than one-based numbering for referencing the full-page images. In HEAD, I removed the macros XLR_BKP_BLOCK_1 through XLR_BKP_BLOCK_4. They are still there in the header files in previous branches, but are no longer used by the code. In addition, fix some other bugs identified in the course of making these changes: spgRedoAddNode could fail to update the parent downlink at all, if the parent tuple is in the same page as either the old or new split tuple and we're not doing a full-page image: it would get fooled by the LSN having been advanced already. This would result in permanent index corruption, not just transient failure of concurrent queries. Also, ginHeapTupleFastInsert's "merge lists" case failed to mark the old tail page as a candidate for a full-page image; in the worst case this could result in torn-page corruption. heap_xlog_freeze() was inconsistent about using a cleanup lock or plain exclusive lock: it did the former in the normal path but the latter for a full-page image. A plain exclusive lock seems sufficient, so change to that. Also, remove gistRedoPageDeleteRecord(), which has been dead code since VACUUM FULL was rewritten. Back-patch to 9.0, where hot standby was introduced. Note however that 9.0 had a significantly different WAL-logging scheme for GIST index updates, and it doesn't appear possible to make that scheme safe for concurrent hot standby queries, because it can leave inconsistent states in the index even between WAL records. Given the lack of complaints from the field, we won't work too hard on fixing that branch.
2012-11-13 04:05:08 +01:00
}
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
else if (XLogRecGetInfo(record) & XLOG_HEAP_INIT_PAGE)
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
nbuffer = XLogInitBufferForRedo(record, 0);
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = (Page) BufferGetPage(nbuffer);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageInit(page, BufferGetPageSize(nbuffer), 0);
newaction = BLK_NEEDS_REDO;
}
else
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
newaction = XLogReadBufferForRedo(record, 0, &nbuffer);
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
/*
Fix a violation of WAL coding rules in the recent patch to include an "all tuples visible" flag in heap page headers. The flag update *must* be applied before calling XLogInsert, but heap_update and the tuple moving routines in VACUUM FULL were ignoring this rule. A crash and replay could therefore leave the flag incorrectly set, causing rows to appear visible in seqscans when they should not be. This might explain recent reports of data corruption from Jeff Ross and others. In passing, do a bit of editorialization on comments in visibilitymap.c.
2009-08-24 04:18:32 +02:00
* The visibility map may need to be fixed even if the heap page is
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
* already up-to-date.
*/
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
if (xlrec->flags & XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED)
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
{
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
Relation reln = CreateFakeRelcacheEntry(rlocator);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
Buffer vmbuffer = InvalidBuffer;
8.4 pgindent run, with new combined Linux/FreeBSD/MinGW typedef list provided by Andrew.
2009-06-11 16:49:15 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
visibilitymap_pin(reln, newblk, &vmbuffer);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
visibilitymap_clear(reln, newblk, vmbuffer, VISIBILITYMAP_VALID_BITS);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
ReleaseBuffer(vmbuffer);
Introduce visibility map. The visibility map is a bitmap with one bit per heap page, where a set bit indicates that all tuples on the page are visible to all transactions, and the page therefore doesn't need vacuuming. It is stored in a new relation fork. Lazy vacuum uses the visibility map to skip pages that don't need vacuuming. Vacuum is also responsible for setting the bits in the map. In the future, this can hopefully be used to implement index-only-scans, but we can't currently guarantee that the visibility map is always 100% up-to-date. In addition to the visibility map, there's a new PD_ALL_VISIBLE flag on each heap page, also indicating that all tuples on the page are visible to all transactions. It's important that this flag is kept up-to-date. It is also used to skip visibility tests in sequential scans, which gives a small performance gain on seqscans.
2008-12-03 14:05:22 +01:00
FreeFakeRelcacheEntry(reln);
}
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
/* Deal with new tuple */
if (newaction == BLK_NEEDS_REDO)
Clean up and document the API for XLogOpenRelation and XLogReadBuffer. This commit doesn't make much functional change, but it does eliminate some duplicated code --- for instance, PageIsNew tests are now done inside XLogReadBuffer rather than by each caller. The GIST xlog code still needs a lot of love, but I'll worry about that separately.
2006-03-29 23:17:39 +02:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
char *recdata;
char *recdata_end;
Size datalen;
Size tuplen;
recdata = XLogRecGetBlockData(record, 0, &datalen);
recdata_end = recdata + datalen;
pgindent run. Make it all clean.
2001-03-22 05:01:46 +01:00
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(nbuffer);
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
offnum = xlrec->new_offnum;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (PageGetMaxOffsetNumber(page) + 1 < offnum)
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(PANIC, "invalid max offset number");
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
if (xlrec->flags & XLH_UPDATE_PREFIX_FROM_OLD)
Fix buffer pin leak in heap update redo routine. In a heap update, if the old and new tuple were on different pages, and the new page no longer existed (because it was subsequently truncated away by vacuum), heap_xlog_update forgot to release the pin on the old buffer. This bug was introduced by the "Fix multiple problems in WAL replay" patch, commit 3bbf668de9f1bc172371681e80a4e769b6d014c8 (on master branch). With full_page_writes=off, this triggered an "incorrect local pin count" error later in replay, if the old page was vacuumed. This fixes bug #7969, reported by Yunong Xiao. Backpatch to 9.0, like the commit that introduced this bug.
2013-03-27 20:51:27 +01:00
{
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
Assert(newblk == oldblk);
memcpy(&prefixlen, recdata, sizeof(uint16));
recdata += sizeof(uint16);
Fix buffer pin leak in heap update redo routine. In a heap update, if the old and new tuple were on different pages, and the new page no longer existed (because it was subsequently truncated away by vacuum), heap_xlog_update forgot to release the pin on the old buffer. This bug was introduced by the "Fix multiple problems in WAL replay" patch, commit 3bbf668de9f1bc172371681e80a4e769b6d014c8 (on master branch). With full_page_writes=off, this triggered an "incorrect local pin count" error later in replay, if the old page was vacuumed. This fixes bug #7969, reported by Yunong Xiao. Backpatch to 9.0, like the commit that introduced this bug.
2013-03-27 20:51:27 +01:00
}
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
if (xlrec->flags & XLH_UPDATE_SUFFIX_FROM_OLD)
Clean up and document the API for XLogOpenRelation and XLogReadBuffer. This commit doesn't make much functional change, but it does eliminate some duplicated code --- for instance, PageIsNew tests are now done inside XLogReadBuffer rather than by each caller. The GIST xlog code still needs a lot of love, but I'll worry about that separately.
2006-03-29 23:17:39 +02:00
{
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
Assert(newblk == oldblk);
memcpy(&suffixlen, recdata, sizeof(uint16));
recdata += sizeof(uint16);
Clean up and document the API for XLogOpenRelation and XLogReadBuffer. This commit doesn't make much functional change, but it does eliminate some duplicated code --- for instance, PageIsNew tests are now done inside XLogReadBuffer rather than by each caller. The GIST xlog code still needs a lot of love, but I'll worry about that separately.
2006-03-29 23:17:39 +02:00
}
Heap redo/undo (except for tuple moving used by vacuum).
2000-09-07 11:58:38 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
memcpy((char *) &xlhdr, recdata, SizeOfHeapHeader);
recdata += SizeOfHeapHeader;
tuplen = recdata_end - recdata;
Assert(tuplen <= MaxHeapTupleSize);
Clean up and document the API for XLogOpenRelation and XLogReadBuffer. This commit doesn't make much functional change, but it does eliminate some duplicated code --- for instance, PageIsNew tests are now done inside XLogReadBuffer rather than by each caller. The GIST xlog code still needs a lot of love, but I'll worry about that separately.
2006-03-29 23:17:39 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
htup = &tbuf.hdr;
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
MemSet((char *) htup, 0, SizeofHeapTupleHeader);
Remove the mostly-stubbed-out-anyway support routines for WAL UNDO. That code is never going to be used in the foreseeable future, and where it's more than a stub it's making the redo routines harder to read.
2005-06-06 19:01:25 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
/*
* Reconstruct the new tuple using the prefix and/or suffix from the
* old tuple, and the data stored in the WAL record.
*/
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
newp = (char *) htup + SizeofHeapTupleHeader;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (prefixlen > 0)
{
int len;
/* copy bitmap [+ padding] [+ oid] from WAL record */
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
len = xlhdr.t_hoff - SizeofHeapTupleHeader;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
memcpy(newp, recdata, len);
recdata += len;
newp += len;
/* copy prefix from old tuple */
memcpy(newp, (char *) oldtup.t_data + oldtup.t_data->t_hoff, prefixlen);
newp += prefixlen;
/* copy new tuple data from WAL record */
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
len = tuplen - (xlhdr.t_hoff - SizeofHeapTupleHeader);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
memcpy(newp, recdata, len);
recdata += len;
newp += len;
}
else
{
/*
* copy bitmap [+ padding] [+ oid] + data from record, all in one
* go
*/
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
memcpy(newp, recdata, tuplen);
recdata += tuplen;
newp += tuplen;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
}
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
Assert(recdata == recdata_end);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
/* copy suffix from old tuple */
if (suffixlen > 0)
memcpy(newp, (char *) oldtup.t_data + oldtup.t_len - suffixlen, suffixlen);
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
Use FLEXIBLE_ARRAY_MEMBER for HeapTupleHeaderData.t_bits[]. This requires changing quite a few places that were depending on sizeof(HeapTupleHeaderData), but it seems for the best. Michael Paquier, some adjustments by me
2015-02-21 21:13:06 +01:00
newlen = SizeofHeapTupleHeader + tuplen + prefixlen + suffixlen;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
htup->t_infomask2 = xlhdr.t_infomask2;
htup->t_infomask = xlhdr.t_infomask;
htup->t_hoff = xlhdr.t_hoff;
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
HeapTupleHeaderSetXmin(htup, XLogRecGetXid(record));
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
HeapTupleHeaderSetCmin(htup, FirstCommandId);
HeapTupleHeaderSetXmax(htup, xlrec->new_xmax);
/* Make sure there is no forward chain link in t_ctid */
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
htup->t_ctid = newtid;
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
offnum = PageAddItem(page, (Item) htup, newlen, offnum, true, true);
if (offnum == InvalidOffsetNumber)
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(PANIC, "failed to add tuple");
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
if (xlrec->flags & XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED)
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageClearAllVisible(page);
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
freespace = PageGetHeapFreeSpace(page); /* needed to update FSM below */
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageSetLSN(page, lsn);
MarkBufferDirty(nbuffer);
Only WAL-log the modified portion in an UPDATE, if possible. When a row is updated, and the new tuple version is put on the same page as the old one, only WAL-log the part of the new tuple that's not identical to the old. This saves significantly on the amount of WAL that needs to be written, in the common case that most fields are not modified. Amit Kapila, with a lot of back and forth with me, Robert Haas, and others.
2014-03-12 21:46:04 +01:00
}
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (BufferIsValid(nbuffer) && nbuffer != obuffer)
UnlockReleaseBuffer(nbuffer);
if (BufferIsValid(obuffer))
Fix multiple problems in WAL replay. Most of the replay functions for WAL record types that modify more than one page failed to ensure that those pages were locked correctly to ensure that concurrent queries could not see inconsistent page states. This is a hangover from coding decisions made long before Hot Standby was added, when it was hardly necessary to acquire buffer locks during WAL replay at all, let alone hold them for carefully-chosen periods. The key problem was that RestoreBkpBlocks was written to hold lock on each page restored from a full-page image for only as long as it took to update that page. This was guaranteed to break any WAL replay function in which there was any update-ordering constraint between pages, because even if the nominal order of the pages is the right one, any mixture of full-page and non-full-page updates in the same record would result in out-of-order updates. Moreover, it wouldn't work for situations where there's a requirement to maintain lock on one page while updating another. Failure to honor an update ordering constraint in this way is thought to be the cause of bug #7648 from Daniel Farina: what seems to have happened there is that a btree page being split was rewritten from a full-page image before the new right sibling page was written, and because lock on the original page was not maintained it was possible for hot standby queries to try to traverse the page's right-link to the not-yet-existing sibling page. To fix, get rid of RestoreBkpBlocks as such, and instead create a new function RestoreBackupBlock that restores just one full-page image at a time. This function can be invoked by WAL replay functions at the points where they would otherwise perform non-full-page updates; in this way, the physical order of page updates remains the same no matter which pages are replaced by full-page images. We can then further adjust the logic in individual replay functions if it is necessary to hold buffer locks for overlapping periods. A side benefit is that we can simplify the handling of concurrency conflict resolution by moving that code into the record-type-specfic functions; there's no more need to contort the code layout to keep conflict resolution in front of the RestoreBkpBlocks call. In connection with that, standardize on zero-based numbering rather than one-based numbering for referencing the full-page images. In HEAD, I removed the macros XLR_BKP_BLOCK_1 through XLR_BKP_BLOCK_4. They are still there in the header files in previous branches, but are no longer used by the code. In addition, fix some other bugs identified in the course of making these changes: spgRedoAddNode could fail to update the parent downlink at all, if the parent tuple is in the same page as either the old or new split tuple and we're not doing a full-page image: it would get fooled by the LSN having been advanced already. This would result in permanent index corruption, not just transient failure of concurrent queries. Also, ginHeapTupleFastInsert's "merge lists" case failed to mark the old tail page as a candidate for a full-page image; in the worst case this could result in torn-page corruption. heap_xlog_freeze() was inconsistent about using a cleanup lock or plain exclusive lock: it did the former in the normal path but the latter for a full-page image. A plain exclusive lock seems sufficient, so change to that. Also, remove gistRedoPageDeleteRecord(), which has been dead code since VACUUM FULL was rewritten. Back-patch to 9.0, where hot standby was introduced. Note however that 9.0 had a significantly different WAL-logging scheme for GIST index updates, and it doesn't appear possible to make that scheme safe for concurrent hot standby queries, because it can leave inconsistent states in the index even between WAL records. Given the lack of complaints from the field, we won't work too hard on fixing that branch.
2012-11-13 04:05:08 +01:00
UnlockReleaseBuffer(obuffer);
Update FSM on WAL replay. This is a bit limited; the FSM is only updated on non-full-page-image WAL records, and quite arbitrarily, only if there's less than 20% free space on the page after the insert/update (not on HOT updates, though). The 20% cutoff should avoid most of the overhead, when replaying a bulk insertion, for example, while ensuring that pages that are full are marked as full in the FSM. This is mostly to avoid the nasty worst case scenario, where you replay from a PITR archive, and the FSM information in the base backup is really out of date. If there was a lot of pages that the outdated FSM claims to have free space, but don't actually have any, the first unlucky inserter after the recovery would traverse through all those pages, just to find out that they're full. We didn't have this problem with the old FSM implementation, because we simply threw the FSM information away on a non-clean shutdown.
2008-10-31 20:40:27 +01:00
/*
Fix multiple problems in WAL replay. Most of the replay functions for WAL record types that modify more than one page failed to ensure that those pages were locked correctly to ensure that concurrent queries could not see inconsistent page states. This is a hangover from coding decisions made long before Hot Standby was added, when it was hardly necessary to acquire buffer locks during WAL replay at all, let alone hold them for carefully-chosen periods. The key problem was that RestoreBkpBlocks was written to hold lock on each page restored from a full-page image for only as long as it took to update that page. This was guaranteed to break any WAL replay function in which there was any update-ordering constraint between pages, because even if the nominal order of the pages is the right one, any mixture of full-page and non-full-page updates in the same record would result in out-of-order updates. Moreover, it wouldn't work for situations where there's a requirement to maintain lock on one page while updating another. Failure to honor an update ordering constraint in this way is thought to be the cause of bug #7648 from Daniel Farina: what seems to have happened there is that a btree page being split was rewritten from a full-page image before the new right sibling page was written, and because lock on the original page was not maintained it was possible for hot standby queries to try to traverse the page's right-link to the not-yet-existing sibling page. To fix, get rid of RestoreBkpBlocks as such, and instead create a new function RestoreBackupBlock that restores just one full-page image at a time. This function can be invoked by WAL replay functions at the points where they would otherwise perform non-full-page updates; in this way, the physical order of page updates remains the same no matter which pages are replaced by full-page images. We can then further adjust the logic in individual replay functions if it is necessary to hold buffer locks for overlapping periods. A side benefit is that we can simplify the handling of concurrency conflict resolution by moving that code into the record-type-specfic functions; there's no more need to contort the code layout to keep conflict resolution in front of the RestoreBkpBlocks call. In connection with that, standardize on zero-based numbering rather than one-based numbering for referencing the full-page images. In HEAD, I removed the macros XLR_BKP_BLOCK_1 through XLR_BKP_BLOCK_4. They are still there in the header files in previous branches, but are no longer used by the code. In addition, fix some other bugs identified in the course of making these changes: spgRedoAddNode could fail to update the parent downlink at all, if the parent tuple is in the same page as either the old or new split tuple and we're not doing a full-page image: it would get fooled by the LSN having been advanced already. This would result in permanent index corruption, not just transient failure of concurrent queries. Also, ginHeapTupleFastInsert's "merge lists" case failed to mark the old tail page as a candidate for a full-page image; in the worst case this could result in torn-page corruption. heap_xlog_freeze() was inconsistent about using a cleanup lock or plain exclusive lock: it did the former in the normal path but the latter for a full-page image. A plain exclusive lock seems sufficient, so change to that. Also, remove gistRedoPageDeleteRecord(), which has been dead code since VACUUM FULL was rewritten. Back-patch to 9.0, where hot standby was introduced. Note however that 9.0 had a significantly different WAL-logging scheme for GIST index updates, and it doesn't appear possible to make that scheme safe for concurrent hot standby queries, because it can leave inconsistent states in the index even between WAL records. Given the lack of complaints from the field, we won't work too hard on fixing that branch.
2012-11-13 04:05:08 +01:00
* If the new page is running low on free space, update the FSM as well.
Update FSM on WAL replay. This is a bit limited; the FSM is only updated on non-full-page-image WAL records, and quite arbitrarily, only if there's less than 20% free space on the page after the insert/update (not on HOT updates, though). The 20% cutoff should avoid most of the overhead, when replaying a bulk insertion, for example, while ensuring that pages that are full are marked as full in the FSM. This is mostly to avoid the nasty worst case scenario, where you replay from a PITR archive, and the FSM information in the base backup is really out of date. If there was a lot of pages that the outdated FSM claims to have free space, but don't actually have any, the first unlucky inserter after the recovery would traverse through all those pages, just to find out that they're full. We didn't have this problem with the old FSM implementation, because we simply threw the FSM information away on a non-clean shutdown.
2008-10-31 20:40:27 +01:00
* Arbitrarily, our definition of "low" is less than 20%. We can't do much
* better than that without knowing the fill-factor for the table.
*
* However, don't update the FSM on HOT updates, because after crash
* recovery, either the old or the new tuple will certainly be dead and
* prunable. After pruning, the page will have roughly as much free space
* as it did before the update, assuming the new tuple is about the same
* size as the old one.
*
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
* XXX: Don't do this if the page was restored from full page image. We
* don't bother to update the FSM in that case, it doesn't need to be
Update FSM on WAL replay. This is a bit limited; the FSM is only updated on non-full-page-image WAL records, and quite arbitrarily, only if there's less than 20% free space on the page after the insert/update (not on HOT updates, though). The 20% cutoff should avoid most of the overhead, when replaying a bulk insertion, for example, while ensuring that pages that are full are marked as full in the FSM. This is mostly to avoid the nasty worst case scenario, where you replay from a PITR archive, and the FSM information in the base backup is really out of date. If there was a lot of pages that the outdated FSM claims to have free space, but don't actually have any, the first unlucky inserter after the recovery would traverse through all those pages, just to find out that they're full. We didn't have this problem with the old FSM implementation, because we simply threw the FSM information away on a non-clean shutdown.
2008-10-31 20:40:27 +01:00
* totally accurate anyway.
*/
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (newaction == BLK_NEEDS_REDO && !hot_update && freespace < BLCKSZ / 5)
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecordPageWithFreeSpace(rlocator, newblk, freespace);
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
}
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
static void
heap_xlog_confirm(XLogReaderState *record)
{
XLogRecPtr lsn = record->EndRecPtr;
xl_heap_confirm *xlrec = (xl_heap_confirm *) XLogRecGetData(record);
Buffer buffer;
Page page;
OffsetNumber offnum;
ItemId lp = NULL;
HeapTupleHeader htup;
if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
{
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(buffer);
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
offnum = xlrec->offnum;
if (PageGetMaxOffsetNumber(page) >= offnum)
lp = PageGetItemId(page, offnum);
if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(PANIC, "invalid lp");
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
htup = (HeapTupleHeader) PageGetItem(page, lp);
/*
* Confirm tuple as actually inserted
*/
ItemPointerSet(&htup->t_ctid, BufferGetBlockNumber(buffer), offnum);
PageSetLSN(page, lsn);
MarkBufferDirty(buffer);
}
if (BufferIsValid(buffer))
UnlockReleaseBuffer(buffer);
}
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
static void
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_lock(XLogReaderState *record)
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRecPtr lsn = record->EndRecPtr;
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
xl_heap_lock *xlrec = (xl_heap_lock *) XLogRecGetData(record);
Buffer buffer;
Page page;
OffsetNumber offnum;
ItemId lp = NULL;
HeapTupleHeader htup;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
/*
* The visibility map may need to be fixed even if the heap page is
* already up-to-date.
*/
if (xlrec->flags & XLH_LOCK_ALL_FROZEN_CLEARED)
{
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
RelFileLocator rlocator;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
Buffer vmbuffer = InvalidBuffer;
BlockNumber block;
Relation reln;
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecGetBlockTag(record, 0, &rlocator, NULL, &block);
reln = CreateFakeRelcacheEntry(rlocator);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
visibilitymap_pin(reln, block, &vmbuffer);
visibilitymap_clear(reln, block, vmbuffer, VISIBILITYMAP_ALL_FROZEN);
ReleaseBuffer(vmbuffer);
FreeFakeRelcacheEntry(reln);
}
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
{
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = (Page) BufferGetPage(buffer);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
offnum = xlrec->offnum;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (PageGetMaxOffsetNumber(page) >= offnum)
lp = PageGetItemId(page, offnum);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(PANIC, "invalid lp");
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
htup = (HeapTupleHeader) PageGetItem(page, lp);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
Make HEAP_LOCK/HEAP2_LOCK_UPDATED replay reset HEAP_XMAX_INVALID. 0ac5ad5 started to compress infomask bits in WAL records. Unfortunately the replay routines for XLOG_HEAP_LOCK/XLOG_HEAP2_LOCK_UPDATED forgot to reset the HEAP_XMAX_INVALID (and some other) hint bits. Luckily that's not problematic in the majority of cases, because after a crash/on a standby row locks aren't meaningful. Unfortunately that does not hold true in the presence of prepared transactions. This means that after a crash, or after promotion, row level locks held by a prepared, but not yet committed, prepared transaction might not be enforced. Discussion: 20160715192319.ubfuzim4zv3rqnxv@alap3.anarazel.de Backpatch: 9.3, the oldest branch on which 0ac5ad5 is present.
2016-07-15 23:37:06 +02:00
htup->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
htup->t_infomask2 &= ~HEAP_KEYS_UPDATED;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
fix_infomask_from_infobits(xlrec->infobits_set, &htup->t_infomask,
&htup->t_infomask2);
Fix WAL replay of locking an updated tuple We were resetting the tuple's HEAP_HOT_UPDATED flag as well as t_ctid on WAL replay of a tuple-lock operation, which is incorrect when the tuple is already updated. Back-patch to 9.3. The clearing of both header elements was there previously, but since no update could be present on a tuple that was being locked, it was harmless. Bug reported by Peter Geoghegan and Greg Stark in CAM3SWZTMQiCi5PV5OWHb+bYkUcnCk=O67w0cSswPvV7XfUcU5g@mail.gmail.com and CAM-w4HPTOeMT4KP0OJK+mGgzgcTOtLRTvFZyvD0O4aH-7dxo3Q@mail.gmail.com respectively; diagnosis by Andres Freund.
2014-02-27 15:13:39 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
/*
* Clear relevant update flags, but only if the modified infomask says
* there's no update.
*/
if (HEAP_XMAX_IS_LOCKED_ONLY(htup->t_infomask))
{
HeapTupleHeaderClearHotUpdated(htup);
/* Make sure there is no forward chain link in t_ctid */
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
ItemPointerSet(&htup->t_ctid,
BufferGetBlockNumber(buffer),
offnum);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
}
Fix xl_heap_lock WAL record field's data type. Make xl_heap_lock's infobits_set field of type uint8, not int8. Using int8 isn't appropriate given that the field just holds status bits. This fixes an oversight in commit 0ac5ad5134. In passing rename the nearby TransactionId field to "xmax" to make things consistency with related records, such as xl_heap_lock_updated. Deliberately avoid a bump in XLOG_PAGE_MAGIC. No backpatch, either. Author: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/CAH2-WzkCd3kOS8b7Rfxw7Mh1_6jvX=Nzo-CWR1VBTiOtVZkWHA@mail.gmail.com
2023-04-11 23:07:54 +02:00
HeapTupleHeaderSetXmax(htup, xlrec->xmax);
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
HeapTupleHeaderSetCmax(htup, FirstCommandId, false);
PageSetLSN(page, lsn);
MarkBufferDirty(buffer);
Fix WAL replay of locking an updated tuple We were resetting the tuple's HEAP_HOT_UPDATED flag as well as t_ctid on WAL replay of a tuple-lock operation, which is incorrect when the tuple is already updated. Back-patch to 9.3. The clearing of both header elements was there previously, but since no update could be present on a tuple that was being locked, it was harmless. Bug reported by Peter Geoghegan and Greg Stark in CAM3SWZTMQiCi5PV5OWHb+bYkUcnCk=O67w0cSswPvV7XfUcU5g@mail.gmail.com and CAM-w4HPTOeMT4KP0OJK+mGgzgcTOtLRTvFZyvD0O4aH-7dxo3Q@mail.gmail.com respectively; diagnosis by Andres Freund.
2014-02-27 15:13:39 +01:00
}
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (BufferIsValid(buffer))
UnlockReleaseBuffer(buffer);
Implement sharable row-level locks, and use them for foreign key references to eliminate unnecessary deadlocks. This commit adds SELECT ... FOR SHARE paralleling SELECT ... FOR UPDATE. The implementation uses a new SLRU data structure (managed much like pg_subtrans) to represent multiple- transaction-ID sets. When more than one transaction is holding a shared lock on a particular row, we create a MultiXactId representing that set of transactions and store its ID in the row's XMAX. This scheme allows an effectively unlimited number of row locks, just as we did before, while not costing any extra overhead except when a shared lock actually has to be shared. Still TODO: use the regular lock manager to control the grant order when multiple backends are waiting for a row lock. Alvaro Herrera and Tom Lane.
2005-04-28 23:47:18 +02:00
}
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
static void
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_lock_updated(XLogReaderState *record)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRecPtr lsn = record->EndRecPtr;
xl_heap_lock_updated *xlrec;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Buffer buffer;
Page page;
OffsetNumber offnum;
ItemId lp = NULL;
HeapTupleHeader htup;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
xlrec = (xl_heap_lock_updated *) XLogRecGetData(record);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
/*
* The visibility map may need to be fixed even if the heap page is
* already up-to-date.
*/
if (xlrec->flags & XLH_LOCK_ALL_FROZEN_CLEARED)
{
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
RelFileLocator rlocator;
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
Buffer vmbuffer = InvalidBuffer;
BlockNumber block;
Relation reln;
Change internal RelFileNode references to RelFileNumber or RelFileLocator. We have been using the term RelFileNode to refer to either (1) the integer that is used to name the sequence of files for a certain relation within the directory set aside for that tablespace/database combination; or (2) that value plus the OIDs of the tablespace and database; or occasionally (3) the whole series of files created for a relation based on those values. Using the same name for more than one thing is confusing. Replace RelFileNode with RelFileNumber when we're talking about just the single number, i.e. (1) from above, and with RelFileLocator when we're talking about all the things that are needed to locate a relation's files on disk, i.e. (2) from above. In the places where we refer to (3) as a relfilenode, instead refer to "relation storage". Since there is a ton of SQL code in the world that knows about pg_class.relfilenode, don't change the name of that column, or of other SQL-facing things that derive their name from it. On the other hand, do adjust closely-related internal terminology. For example, the structure member names dbNode and spcNode appear to be derived from the fact that the structure itself was called RelFileNode, so change those to dbOid and spcOid. Likewise, various variables with names like rnode and relnode get renamed appropriately, according to how they're being used in context. Hopefully, this is clearer than before. It is also preparation for future patches that intend to widen the relfilenumber fields from its current width of 32 bits. Variables that store a relfilenumber are now declared as type RelFileNumber rather than type Oid; right now, these are the same, but that can now more easily be changed. Dilip Kumar, per an idea from me. Reviewed also by Andres Freund. I fixed some whitespace issues, changed a couple of words in a comment, and made one other minor correction. Discussion: http://postgr.es/m/CA+TgmoamOtXbVAQf9hWFzonUo6bhhjS6toZQd7HZ-pmojtAmag@mail.gmail.com Discussion: http://postgr.es/m/CA+Tgmobp7+7kmi4gkq7Y+4AM9fTvL+O1oQ4-5gFTT+6Ng-dQ=g@mail.gmail.com Discussion: http://postgr.es/m/CAFiTN-vTe79M8uDH1yprOU64MNFE+R3ODRuA+JWf27JbhY4hJw@mail.gmail.com
2022-07-06 17:39:09 +02:00
XLogRecGetBlockTag(record, 0, &rlocator, NULL, &block);
reln = CreateFakeRelcacheEntry(rlocator);
Clear all-frozen visibilitymap status when locking tuples. Since a892234 & fd31cd265 the visibilitymap's freeze bit is used to avoid vacuuming the whole relation in anti-wraparound vacuums. Doing so correctly relies on not adding xids to the heap without also unsetting the visibilitymap flag. Tuple locking related code has not done so. To allow selectively resetting all-frozen - to avoid pessimizing heap_lock_tuple - allow to selectively reset the all-frozen with visibilitymap_clear(). To avoid having to use visibilitymap_get_status (e.g. via VM_ALL_FROZEN) inside a critical section, have visibilitymap_clear() return whether any bits have been reset. There's a remaining issue (denoted by XXX): After the PageIsAllVisible() check in heap_lock_tuple() and heap_lock_updated_tuple_rec() the page status could theoretically change. Practically that currently seems impossible, because updaters will hold a page level pin already. Due to the next beta coming up, it seems better to get the required WAL magic bump done before resolving this issue. The added flags field fields to xl_heap_lock and xl_heap_lock_updated require bumping the WAL magic. Since there's already been a catversion bump since the last beta, that's not an issue. Reviewed-By: Robert Haas, Amit Kapila and Andres Freund Author: Masahiko Sawada, heavily revised by Andres Freund Discussion: CAEepm=3fWAbWryVW9swHyLTY4sXVf0xbLvXqOwUoDiNCx9mBjQ@mail.gmail.com Backpatch: -
2016-07-18 11:01:13 +02:00
visibilitymap_pin(reln, block, &vmbuffer);
visibilitymap_clear(reln, block, vmbuffer, VISIBILITYMAP_ALL_FROZEN);
ReleaseBuffer(vmbuffer);
FreeFakeRelcacheEntry(reln);
}
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
{
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(buffer);
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
offnum = xlrec->offnum;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (PageGetMaxOffsetNumber(page) >= offnum)
lp = PageGetItemId(page, offnum);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(PANIC, "invalid lp");
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
htup = (HeapTupleHeader) PageGetItem(page, lp);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Make HEAP_LOCK/HEAP2_LOCK_UPDATED replay reset HEAP_XMAX_INVALID. 0ac5ad5 started to compress infomask bits in WAL records. Unfortunately the replay routines for XLOG_HEAP_LOCK/XLOG_HEAP2_LOCK_UPDATED forgot to reset the HEAP_XMAX_INVALID (and some other) hint bits. Luckily that's not problematic in the majority of cases, because after a crash/on a standby row locks aren't meaningful. Unfortunately that does not hold true in the presence of prepared transactions. This means that after a crash, or after promotion, row level locks held by a prepared, but not yet committed, prepared transaction might not be enforced. Discussion: 20160715192319.ubfuzim4zv3rqnxv@alap3.anarazel.de Backpatch: 9.3, the oldest branch on which 0ac5ad5 is present.
2016-07-15 23:37:06 +02:00
htup->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
htup->t_infomask2 &= ~HEAP_KEYS_UPDATED;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
fix_infomask_from_infobits(xlrec->infobits_set, &htup->t_infomask,
&htup->t_infomask2);
HeapTupleHeaderSetXmax(htup, xlrec->xmax);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageSetLSN(page, lsn);
MarkBufferDirty(buffer);
}
if (BufferIsValid(buffer))
UnlockReleaseBuffer(buffer);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
}
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
static void
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_inplace(XLogReaderState *record)
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
XLogRecPtr lsn = record->EndRecPtr;
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
xl_heap_inplace *xlrec = (xl_heap_inplace *) XLogRecGetData(record);
Buffer buffer;
Page page;
OffsetNumber offnum;
ItemId lp = NULL;
HeapTupleHeader htup;
uint32 oldlen;
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
Size newlen;
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
Fix multiple problems in WAL replay. Most of the replay functions for WAL record types that modify more than one page failed to ensure that those pages were locked correctly to ensure that concurrent queries could not see inconsistent page states. This is a hangover from coding decisions made long before Hot Standby was added, when it was hardly necessary to acquire buffer locks during WAL replay at all, let alone hold them for carefully-chosen periods. The key problem was that RestoreBkpBlocks was written to hold lock on each page restored from a full-page image for only as long as it took to update that page. This was guaranteed to break any WAL replay function in which there was any update-ordering constraint between pages, because even if the nominal order of the pages is the right one, any mixture of full-page and non-full-page updates in the same record would result in out-of-order updates. Moreover, it wouldn't work for situations where there's a requirement to maintain lock on one page while updating another. Failure to honor an update ordering constraint in this way is thought to be the cause of bug #7648 from Daniel Farina: what seems to have happened there is that a btree page being split was rewritten from a full-page image before the new right sibling page was written, and because lock on the original page was not maintained it was possible for hot standby queries to try to traverse the page's right-link to the not-yet-existing sibling page. To fix, get rid of RestoreBkpBlocks as such, and instead create a new function RestoreBackupBlock that restores just one full-page image at a time. This function can be invoked by WAL replay functions at the points where they would otherwise perform non-full-page updates; in this way, the physical order of page updates remains the same no matter which pages are replaced by full-page images. We can then further adjust the logic in individual replay functions if it is necessary to hold buffer locks for overlapping periods. A side benefit is that we can simplify the handling of concurrency conflict resolution by moving that code into the record-type-specfic functions; there's no more need to contort the code layout to keep conflict resolution in front of the RestoreBkpBlocks call. In connection with that, standardize on zero-based numbering rather than one-based numbering for referencing the full-page images. In HEAD, I removed the macros XLR_BKP_BLOCK_1 through XLR_BKP_BLOCK_4. They are still there in the header files in previous branches, but are no longer used by the code. In addition, fix some other bugs identified in the course of making these changes: spgRedoAddNode could fail to update the parent downlink at all, if the parent tuple is in the same page as either the old or new split tuple and we're not doing a full-page image: it would get fooled by the LSN having been advanced already. This would result in permanent index corruption, not just transient failure of concurrent queries. Also, ginHeapTupleFastInsert's "merge lists" case failed to mark the old tail page as a candidate for a full-page image; in the worst case this could result in torn-page corruption. heap_xlog_freeze() was inconsistent about using a cleanup lock or plain exclusive lock: it did the former in the normal path but the latter for a full-page image. A plain exclusive lock seems sufficient, so change to that. Also, remove gistRedoPageDeleteRecord(), which has been dead code since VACUUM FULL was rewritten. Back-patch to 9.0, where hot standby was introduced. Note however that 9.0 had a significantly different WAL-logging scheme for GIST index updates, and it doesn't appear possible to make that scheme safe for concurrent hot standby queries, because it can leave inconsistent states in the index even between WAL records. Given the lack of complaints from the field, we won't work too hard on fixing that branch.
2012-11-13 04:05:08 +01:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
char *newtup = XLogRecGetBlockData(record, 0, &newlen);
Revert no-op changes to BufferGetPage() The reverted changes were intended to force a choice of whether any newly-added BufferGetPage() calls needed to be accompanied by a test of the snapshot age, to support the "snapshot too old" feature. Such an accompanying test is needed in about 7% of the cases, where the page is being used as part of a scan rather than positioning for other purposes (such as DML or vacuuming). The additional effort required for back-patching, and the doubt whether the intended benefit would really be there, have indicated it is best just to rely on developers to do the right thing based on comments and existing usage, as we do with many other conventions. This change should have little or no effect on generated executable code. Motivated by the back-patching pain of Tom Lane and Robert Haas
2016-04-20 15:31:19 +02:00
page = BufferGetPage(buffer);
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
offnum = xlrec->offnum;
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (PageGetMaxOffsetNumber(page) >= offnum)
lp = PageGetItemId(page, offnum);
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(PANIC, "invalid lp");
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
htup = (HeapTupleHeader) PageGetItem(page, lp);
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
oldlen = ItemIdGetLength(lp) - htup->t_hoff;
if (oldlen != newlen)
Remove function names from some elog() calls in heapam.c. At least one of the names was, due to a function renaming late in the development of ON CONFLICT, wrong. Since including function names in error messages is against the message style guide anyway, remove them from the messages. Discussion: CAM3SWZT8paz=usgMVHm0XOETkQvzjRtAUthATnmaHQQY0obnGw@mail.gmail.com Backpatch: 9.5, where ON CONFLICT was introduced
2015-11-19 01:25:58 +01:00
elog(PANIC, "wrong tuple length");
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
memcpy((char *) htup + htup->t_hoff, newtup, newlen);
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
Refactor per-page logic common to all redo routines to a new function. Every redo routine uses the same idiom to determine what to do to a page: check if there's a backup block for it, and if not read, the buffer if the block exists, and check its LSN. Refactor that into a common function, XLogReadBufferForRedo, making all the redo routines shorter and more readable. This has no user-visible effect, and makes no changes to the WAL format. Reviewed by Andres Freund, Alvaro Herrera, Michael Paquier.
2014-08-13 14:39:08 +02:00
PageSetLSN(page, lsn);
MarkBufferDirty(buffer);
}
if (BufferIsValid(buffer))
UnlockReleaseBuffer(buffer);
Clean up code associated with updating pg_class statistics columns (relpages/reltuples). To do this, create formal support in heapam.c for "overwrite" tuple updates (including xlog replay capability) and use that instead of the ad-hoc overwrites we'd been using in VACUUM and CREATE INDEX. Take the responsibility for updating stats during CREATE INDEX out of the individual index AMs, and do it where it belongs, in catalog/index.c. Aside from being more modular, this avoids having to update the same tuple twice in some paths through CREATE INDEX. It's probably not measurably faster, but for sure it's a lot cleaner than before.
2006-05-11 01:18:39 +02:00
}
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
void
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_redo(XLogReaderState *record)
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
Allow read only connections during recovery, known as Hot Standby. Enabled by recovery_connections = on (default) and forcing archive recovery using a recovery.conf. Recovery processing now emulates the original transactions as they are replayed, providing full locking and MVCC behaviour for read only queries. Recovery must enter consistent state before connections are allowed, so there is a delay, typically short, before connections succeed. Replay of recovering transactions can conflict and in some cases deadlock with queries during recovery; these result in query cancellation after max_standby_delay seconds have expired. Infrastructure changes have minor effects on normal running, though introduce four new types of WAL record. New test mode "make standbycheck" allows regression tests of static command behaviour on a standby server while in recovery. Typical and extreme dynamic behaviours have been checked via code inspection and manual testing. Few port specific behaviours have been utilised, though primary testing has been on Linux only so far. This commit is the basic patch. Additional changes will follow in this release to enhance some aspects of behaviour, notably improved handling of conflicts, deadlock detection and query cancellation. Changes to VACUUM FULL are also required. Simon Riggs, with significant and lengthy review by Heikki Linnakangas, including streamlined redesign of snapshot creation and two-phase commit. Important contributions from Florian Pflug, Mark Kirkwood, Merlin Moncure, Greg Stark, Gianni Ciolli, Gabriele Bartolini, Hannu Krosing, Robert Haas, Tatsuo Ishii, Hiroyuki Yamada plus support and feedback from many other community members.
2009-12-19 02:32:45 +01:00
/*
* These operations don't overwrite MVCC data so no conflict processing is
* required. The ones in heap2 rmgr do.
*/
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
switch (info & XLOG_HEAP_OPMASK)
{
case XLOG_HEAP_INSERT:
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_insert(record);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
break;
case XLOG_HEAP_DELETE:
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_delete(record);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
break;
case XLOG_HEAP_UPDATE:
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_update(record, false);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
break;
Logical decoding of TRUNCATE Add a new WAL record type for TRUNCATE, which is only used when wal_level >= logical. (For physical replication, TRUNCATE is already replicated via SMGR records.) Add new callback for logical decoding output plugins to receive TRUNCATE actions. Author: Simon Riggs <simon@2ndquadrant.com> Author: Marco Nenciarini <marco.nenciarini@2ndquadrant.it> Author: Peter Eisentraut <peter.eisentraut@2ndquadrant.com> Reviewed-by: Petr Jelinek <petr.jelinek@2ndquadrant.com> Reviewed-by: Andres Freund <andres@anarazel.de> Reviewed-by: Alvaro Herrera <alvherre@alvh.no-ip.org>
2018-04-07 17:17:56 +02:00
case XLOG_HEAP_TRUNCATE:
Post-feature-freeze pgindent run. Discussion: https://postgr.es/m/15719.1523984266@sss.pgh.pa.us
2018-04-26 20:47:16 +02:00
Logical decoding of TRUNCATE Add a new WAL record type for TRUNCATE, which is only used when wal_level >= logical. (For physical replication, TRUNCATE is already replicated via SMGR records.) Add new callback for logical decoding output plugins to receive TRUNCATE actions. Author: Simon Riggs <simon@2ndquadrant.com> Author: Marco Nenciarini <marco.nenciarini@2ndquadrant.it> Author: Peter Eisentraut <peter.eisentraut@2ndquadrant.com> Reviewed-by: Petr Jelinek <petr.jelinek@2ndquadrant.com> Reviewed-by: Andres Freund <andres@anarazel.de> Reviewed-by: Alvaro Herrera <alvherre@alvh.no-ip.org>
2018-04-07 17:17:56 +02:00
/*
* TRUNCATE is a no-op because the actions are already logged as
* SMGR WAL records. TRUNCATE WAL record only exists for logical
* decoding.
*/
break;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
case XLOG_HEAP_HOT_UPDATE:
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_update(record, true);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
break;
Add support for INSERT ... ON CONFLICT DO NOTHING/UPDATE. The newly added ON CONFLICT clause allows to specify an alternative to raising a unique or exclusion constraint violation error when inserting. ON CONFLICT refers to constraints that can either be specified using a inference clause (by specifying the columns of a unique constraint) or by naming a unique or exclusion constraint. DO NOTHING avoids the constraint violation, without touching the pre-existing row. DO UPDATE SET ... [WHERE ...] updates the pre-existing tuple, and has access to both the tuple proposed for insertion and the existing tuple; the optional WHERE clause can be used to prevent an update from being executed. The UPDATE SET and WHERE clauses have access to the tuple proposed for insertion using the "magic" EXCLUDED alias, and to the pre-existing tuple using the table name or its alias. This feature is often referred to as upsert. This is implemented using a new infrastructure called "speculative insertion". It is an optimistic variant of regular insertion that first does a pre-check for existing tuples and then attempts an insert. If a violating tuple was inserted concurrently, the speculatively inserted tuple is deleted and a new attempt is made. If the pre-check finds a matching tuple the alternative DO NOTHING or DO UPDATE action is taken. If the insertion succeeds without detecting a conflict, the tuple is deemed inserted. To handle the possible ambiguity between the excluded alias and a table named excluded, and for convenience with long relation names, INSERT INTO now can alias its target table. Bumps catversion as stored rules change. Author: Peter Geoghegan, with significant contributions from Heikki Linnakangas and Andres Freund. Testing infrastructure by Jeff Janes. Reviewed-By: Heikki Linnakangas, Andres Freund, Robert Haas, Simon Riggs, Dean Rasheed, Stephen Frost and many others.
2015-05-08 05:31:36 +02:00
case XLOG_HEAP_CONFIRM:
heap_xlog_confirm(record);
break;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
case XLOG_HEAP_LOCK:
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_lock(record);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
break;
case XLOG_HEAP_INPLACE:
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_inplace(record);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
break;
default:
elog(PANIC, "heap_redo: unknown op code %u", info);
}
redo/undo support functions and cleanups.
2000-10-20 13:01:21 +02:00
}
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
void
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap2_redo(XLogReaderState *record)
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
{
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
switch (info & XLOG_HEAP_OPMASK)
{
Remove tupgone special case from vacuumlazy.c. Retry the call to heap_prune_page() in rare cases where there is disagreement between the heap_prune_page() call and the call to HeapTupleSatisfiesVacuum() that immediately follows. Disagreement is possible when a concurrently-aborted transaction makes a tuple DEAD during the tiny window between each step. This was the only case where a tuple considered DEAD by VACUUM still had storage following pruning. VACUUM's definition of dead tuples is now uniformly simple and unambiguous: dead tuples from each page are always LP_DEAD line pointers that were encountered just after we performed pruning (and just before we considered freezing remaining items with tuple storage). Eliminating the tupgone=true special case enables INDEX_CLEANUP=off style skipping of index vacuuming that takes place based on flexible, dynamic criteria. The INDEX_CLEANUP=off case had to know about skipping indexes up-front before now, due to a subtle interaction with the special case (see commit dd695979) -- this was a special case unto itself. Now there are no special cases. And so now it won't matter when or how we decide to skip index vacuuming: it won't affect how pruning behaves, and it won't be affected by any of the implementation details of pruning or freezing. Also remove XLOG_HEAP2_CLEANUP_INFO records. These are no longer necessary because we now rely entirely on heap pruning taking care of recovery conflicts. There is no longer any need to generate recovery conflicts for DEAD tuples that pruning just missed. This also means that heap vacuuming now uses exactly the same strategy for recovery conflicts as index vacuuming always has: REDO routines never need to process a latestRemovedXid from the WAL record, since earlier REDO of the WAL record from pruning is sufficient in all cases. The generic XLOG_HEAP2_CLEAN record type is now split into two new record types to reflect this new division (these are called XLOG_HEAP2_PRUNE and XLOG_HEAP2_VACUUM). Also stop acquiring a super-exclusive lock for heap pages when they're vacuumed during VACUUM's second heap pass. A regular exclusive lock is enough. This is correct because heap page vacuuming is now strictly a matter of setting the LP_DEAD line pointers to LP_UNUSED. No other backend can have a pointer to a tuple located in a pinned buffer that can be invalidated by a concurrent heap page vacuum operation. Heap vacuuming can now be thought of as conceptually similar to index vacuuming and conceptually dissimilar to heap pruning. Heap pruning now has sole responsibility for anything involving the logical contents of the database (e.g., managing transaction status information, recovery conflicts, considering what to do with HOT chains). Index vacuuming and heap vacuuming are now only concerned with recycling garbage items from physical data structures that back the logical database. Bump XLOG_PAGE_MAGIC due to pruning and heap page vacuum WAL record changes. Credit for the idea of retrying pruning a page to avoid the tupgone case goes to Andres Freund. Author: Peter Geoghegan <pg@bowt.ie> Reviewed-By: Andres Freund <andres@anarazel.de> Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com> Discussion: https://postgr.es/m/CAH2-WznneCXTzuFmcwx_EyRQgfsfJAAsu+CsqRFmFXCAar=nJw@mail.gmail.com
2021-04-06 17:49:22 +02:00
case XLOG_HEAP2_PRUNE:
heap_xlog_prune(record);
break;
case XLOG_HEAP2_VACUUM:
heap_xlog_vacuum(record);
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
break;
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
case XLOG_HEAP2_FREEZE_PAGE:
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_freeze_page(record);
Rework tuple freezing protocol Tuple freezing was broken in connection to MultiXactIds; commit 8e53ae025de9 tried to fix it, but didn't go far enough. As noted by Noah Misch, freezing a tuple whose Xmax is a multi containing an aborted update might cause locks in the multi to go ignored by later transactions. This is because the code depended on a multixact above their cutoff point not having any lock-only member older than the cutoff point for Xids, which is easily defeated in READ COMMITTED transactions. The fix for this involves creating a new MultiXactId when necessary. But this cannot be done during WAL replay, and moreover multixact examination requires using CLOG access routines which are not supposed to be used during WAL replay either; so tuple freezing cannot be done with the old freeze WAL record. Therefore, separate the freezing computation from its execution, and change the WAL record to carry all necessary information. At WAL replay time, it's easy to re-execute freezing because we don't need to re-compute the new infomask/Xmax values but just take them from the WAL record. While at it, restructure the coding to ensure all page changes occur in a single critical section without much room for failures. The previous coding wasn't using a critical section, without any explanation as to why this was acceptable. In replication scenarios using the 9.3 branch, standby servers must be upgraded before their master, so that they are prepared to deal with the new WAL record once the master is upgraded; failure to do so will cause WAL replay to die with a PANIC message. Later upgrade of the standby will allow the process to continue where it left off, so there's no disruption of the data in the standby in any case. Standbys know how to deal with the old WAL record, so it's okay to keep the master running the old code for a while. In master, the old freeze WAL record is gone, for cleanliness' sake; there's no compatibility concern there. Backpatch to 9.3, where the original bug was introduced and where the previous fix was backpatched. Álvaro Herrera and Andres Freund
2013-12-16 15:29:50 +01:00
break;
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
case XLOG_HEAP2_VISIBLE:
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_visible(record);
Make the visibility map crash-safe. This involves two main changes from the previous behavior. First, when we set a bit in the visibility map, emit a new WAL record of type XLOG_HEAP2_VISIBLE. Replay sets the page-level PD_ALL_VISIBLE bit and the visibility map bit. Second, when inserting, updating, or deleting a tuple, we can no longer get away with clearing the visibility map bit after releasing the lock on the corresponding heap page, because an intervening crash might leave the visibility map bit set and the page-level bit clear. Making this work requires a bit of interface refactoring. In passing, a few minor but related cleanups: change the test in visibilitymap_set and visibilitymap_clear to throw an error if the wrong page (or no page) is pinned, rather than silently doing nothing; this case should never occur. Also, remove duplicate definitions of InvalidXLogRecPtr. Patch by me, review by Noah Misch.
2011-06-22 05:04:40 +02:00
break;
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
case XLOG_HEAP2_MULTI_INSERT:
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_multi_insert(record);
In COPY, insert tuples to the heap in batches. This greatly reduces the WAL volume, especially when the table is narrow. The overhead of locking the heap page is also reduced. Reduced WAL traffic also makes it scale a lot better, if you run multiple COPY processes at the same time.
2011-11-09 09:54:41 +01:00
break;
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
case XLOG_HEAP2_LOCK_UPDATED:
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_lock_updated(record);
Improve concurrency of foreign key locking This patch introduces two additional lock modes for tuples: "SELECT FOR KEY SHARE" and "SELECT FOR NO KEY UPDATE". These don't block each other, in contrast with already existing "SELECT FOR SHARE" and "SELECT FOR UPDATE". UPDATE commands that do not modify the values stored in the columns that are part of the key of the tuple now grab a SELECT FOR NO KEY UPDATE lock on the tuple, allowing them to proceed concurrently with tuple locks of the FOR KEY SHARE variety. Foreign key triggers now use FOR KEY SHARE instead of FOR SHARE; this means the concurrency improvement applies to them, which is the whole point of this patch. The added tuple lock semantics require some rejiggering of the multixact module, so that the locking level that each transaction is holding can be stored alongside its Xid. Also, multixacts now need to persist across server restarts and crashes, because they can now represent not only tuple locks, but also tuple updates. This means we need more careful tracking of lifetime of pg_multixact SLRU files; since they now persist longer, we require more infrastructure to figure out when they can be removed. pg_upgrade also needs to be careful to copy pg_multixact files over from the old server to the new, or at least part of multixact.c state, depending on the versions of the old and new servers. Tuple time qualification rules (HeapTupleSatisfies routines) need to be careful not to consider tuples with the "is multi" infomask bit set as being only locked; they might need to look up MultiXact values (i.e. possibly do pg_multixact I/O) to find out the Xid that updated a tuple, whereas they previously were assured to only use information readily available from the tuple header. This is considered acceptable, because the extra I/O would involve cases that would previously cause some commands to block waiting for concurrent transactions to finish. Another important change is the fact that locking tuples that have previously been updated causes the future versions to be marked as locked, too; this is essential for correctness of foreign key checks. This causes additional WAL-logging, also (there was previously a single WAL record for a locked tuple; now there are as many as updated copies of the tuple there exist.) With all this in place, contention related to tuples being checked by foreign key rules should be much reduced. As a bonus, the old behavior that a subtransaction grabbing a stronger tuple lock than the parent (sub)transaction held on a given tuple and later aborting caused the weaker lock to be lost, has been fixed. Many new spec files were added for isolation tester framework, to ensure overall behavior is sane. There's probably room for several more tests. There were several reviewers of this patch; in particular, Noah Misch and Andres Freund spent considerable time in it. Original idea for the patch came from Simon Riggs, after a problem report by Joel Jacobson. Most code is from me, with contributions from Marti Raudsepp, Alexander Shulgin, Noah Misch and Andres Freund. This patch was discussed in several pgsql-hackers threads; the most important start at the following message-ids: AANLkTimo9XVcEzfiBR-ut3KVNDkjm2Vxh+t8kAmWjPuv@mail.gmail.com 1290721684-sup-3951@alvh.no-ip.org 1294953201-sup-2099@alvh.no-ip.org 1320343602-sup-2290@alvh.no-ip.org 1339690386-sup-8927@alvh.no-ip.org 4FE5FF020200002500048A3D@gw.wicourts.gov 4FEAB90A0200002500048B7D@gw.wicourts.gov
2013-01-23 16:04:59 +01:00
break;
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
case XLOG_HEAP2_NEW_CID:
pgindent run for 9.4 This includes removing tabs after periods in C comments, which was applied to back branches, so this change should not effect backpatching.
2014-05-06 18:12:18 +02:00
Add new wal_level, logical, sufficient for logical decoding. When wal_level=logical, we'll log columns from the old tuple as configured by the REPLICA IDENTITY facility added in commit 07cacba983ef79be4a84fcd0e0ca3b5fcb85dd65. This makes it possible a properly-configured logical replication solution to correctly follow table updates even if they change the chosen key columns, or, with REPLICA IDENTITY FULL, even if the table has no key at all. Note that updates which do not modify the replica identity column won't log anything extra, making the choice of a good key (i.e. one that will rarely be changed) important to performance when wal_level=logical is configured. Each insert, update, or delete to a catalog table will also log the CMIN and/or CMAX values of stamped by the current transaction. This is necessary because logical decoding will require access to historical snapshots of the catalog in order to decode some data types, and the CMIN/CMAX values that we may need in order to judge row visibility may have been overwritten by the time we need them. Andres Freund, reviewed in various versions by myself, Heikki Linnakangas, KONDO Mitsumasa, and many others.
2013-12-11 00:33:45 +01:00
/*
* Nothing to do on a real replay, only used during logical
* decoding.
*/
break;
Introduce logical decoding. This feature, building on previous commits, allows the write-ahead log stream to be decoded into a series of logical changes; that is, inserts, updates, and deletes and the transactions which contain them. It is capable of handling decoding even across changes to the schema of the effected tables. The output format is controlled by a so-called "output plugin"; an example is included. To make use of this in a real replication system, the output plugin will need to be modified to produce output in the format appropriate to that system, and to perform filtering. Currently, information can be extracted from the logical decoding system only via SQL; future commits will add the ability to stream changes via walsender. Andres Freund, with review and other contributions from many other people, including Álvaro Herrera, Abhijit Menon-Sen, Peter Gheogegan, Kevin Grittner, Robert Haas, Heikki Linnakangas, Fujii Masao, Abhijit Menon-Sen, Michael Paquier, Simon Riggs, Craig Ringer, and Steve Singer.
2014-03-03 22:32:18 +01:00
case XLOG_HEAP2_REWRITE:
Revamp the WAL record format. Each WAL record now carries information about the modified relation and block(s) in a standardized format. That makes it easier to write tools that need that information, like pg_rewind, prefetching the blocks to speed up recovery, etc. There's a whole new API for building WAL records, replacing the XLogRecData chains used previously. The new API consists of XLogRegister* functions, which are called for each buffer and chunk of data that is added to the record. The new API also gives more control over when a full-page image is written, by passing flags to the XLogRegisterBuffer function. This also simplifies the XLogReadBufferForRedo() calls. The function can dig the relation and block number from the WAL record, so they no longer need to be passed as arguments. For the convenience of redo routines, XLogReader now disects each WAL record after reading it, copying the main data part and the per-block data into MAXALIGNed buffers. The data chunks are not aligned within the WAL record, but the redo routines can assume that the pointers returned by XLogRecGet* functions are. Redo routines are now passed the XLogReaderState, which contains the record in the already-disected format, instead of the plain XLogRecord. The new record format also makes the fixed size XLogRecord header smaller, by removing the xl_len field. The length of the "main data" portion is now stored at the end of the WAL record, and there's a separate header after XLogRecord for it. The alignment padding at the end of XLogRecord is also removed. This compansates for the fact that the new format would otherwise be more bulky than the old format. Reviewed by Andres Freund, Amit Kapila, Michael Paquier, Alvaro Herrera, Fujii Masao.
2014-11-20 16:56:26 +01:00
heap_xlog_logical_rewrite(record);
Introduce logical decoding. This feature, building on previous commits, allows the write-ahead log stream to be decoded into a series of logical changes; that is, inserts, updates, and deletes and the transactions which contain them. It is capable of handling decoding even across changes to the schema of the effected tables. The output format is controlled by a so-called "output plugin"; an example is included. To make use of this in a real replication system, the output plugin will need to be modified to produce output in the format appropriate to that system, and to perform filtering. Currently, information can be extracted from the logical decoding system only via SQL; future commits will add the ability to stream changes via walsender. Andres Freund, with review and other contributions from many other people, including Álvaro Herrera, Abhijit Menon-Sen, Peter Gheogegan, Kevin Grittner, Robert Haas, Heikki Linnakangas, Fujii Masao, Abhijit Menon-Sen, Michael Paquier, Simon Riggs, Craig Ringer, and Steve Singer.
2014-03-03 22:32:18 +01:00
break;
HOT updates. When we update a tuple without changing any of its indexed columns, and the new version can be stored on the same heap page, we no longer generate extra index entries for the new version. Instead, index searches follow the HOT-chain links to ensure they find the correct tuple version. In addition, this patch introduces the ability to "prune" dead tuples on a per-page basis, without having to do a complete VACUUM pass to recover space. VACUUM is still needed to clean up dead index entries, however. Pavan Deolasee, with help from a bunch of other people.
2007-09-20 19:56:33 +02:00
default:
elog(PANIC, "heap2_redo: unknown op code %u", info);
}
Fix recently-understood problems with handling of XID freezing, particularly in PITR scenarios. We now WAL-log the replacement of old XIDs with FrozenTransactionId, so that such replacement is guaranteed to propagate to PITR slave databases. Also, rather than relying on hint-bit updates to be preserved, pg_clog is not truncated until all instances of an XID are known to have been replaced by FrozenTransactionId. Add new GUC variables and pg_autovacuum columns to allow management of the freezing policy, so that users can trade off the size of pg_clog against the amount of freezing work done. Revise the already-existing code that forces autovacuum of tables approaching the wraparound point to make it more bulletproof; also, revise the autovacuum logic so that anti-wraparound vacuuming is done per-table rather than per-database. initdb forced because of changes in pg_class, pg_database, and pg_autovacuum catalogs. Heikki Linnakangas, Simon Riggs, and Tom Lane.
2006-11-05 23:42:10 +01:00
}
Add WAL consistency checking facility. When the new GUC wal_consistency_checking is set to a non-empty value, it triggers recording of additional full-page images, which are compared on the standby against the results of applying the WAL record (without regard to those full-page images). Allowable differences such as hints are masked out, and the resulting pages are compared; any difference results in a FATAL error on the standby. Kuntal Ghosh, based on earlier patches by Michael Paquier and Heikki Linnakangas. Extensively reviewed and revised by Michael Paquier and by me, with additional reviews and comments from Amit Kapila, Álvaro Herrera, Simon Riggs, and Peter Eisentraut.
2017-02-08 21:45:30 +01:00
/*
* Mask a heap page before performing consistency checks on it.
*/
void
heap_mask(char *pagedata, BlockNumber blkno)
{
Page page = (Page) pagedata;
OffsetNumber off;
For wal_consistency_checking, mask page checksum as well as page LSN. If the LSN is different, the checksum will be different, too. Ashwin Agrawal, reviewed by Michael Paquier and Kuntal Ghosh Discussion: http://postgr.es/m/CALfoeis5iqrAU-+JAN+ZzXkpPr7+-0OAGv7QUHwFn=-wDy4o4Q@mail.gmail.com
2017-09-22 20:28:22 +02:00
mask_page_lsn_and_checksum(page);
Add WAL consistency checking facility. When the new GUC wal_consistency_checking is set to a non-empty value, it triggers recording of additional full-page images, which are compared on the standby against the results of applying the WAL record (without regard to those full-page images). Allowable differences such as hints are masked out, and the resulting pages are compared; any difference results in a FATAL error on the standby. Kuntal Ghosh, based on earlier patches by Michael Paquier and Heikki Linnakangas. Extensively reviewed and revised by Michael Paquier and by me, with additional reviews and comments from Amit Kapila, Álvaro Herrera, Simon Riggs, and Peter Eisentraut.
2017-02-08 21:45:30 +01:00
mask_page_hint_bits(page);
mask_unused_space(page);
for (off = 1; off <= PageGetMaxOffsetNumber(page); off++)
{
ItemId iid = PageGetItemId(page, off);
char *page_item;
page_item = (char *) (page + ItemIdGetOffset(iid));
if (ItemIdIsNormal(iid))
{
HeapTupleHeader page_htup = (HeapTupleHeader) page_item;
/*
* If xmin of a tuple is not yet frozen, we should ignore
* differences in hint bits, since they can be set without
* emitting WAL.
*/
if (!HeapTupleHeaderXminFrozen(page_htup))
page_htup->t_infomask &= ~HEAP_XACT_MASK;
else
{
/* Still we need to mask xmax hint bits. */
page_htup->t_infomask &= ~HEAP_XMAX_INVALID;
page_htup->t_infomask &= ~HEAP_XMAX_COMMITTED;
}
/*
* During replay, we set Command Id to FirstCommandId. Hence, mask
* it. See heap_xlog_insert() for details.
*/
page_htup->t_choice.t_heap.t_field3.t_cid = MASK_MARKER;
/*
* For a speculative tuple, heap_insert() does not set ctid in the
* caller-passed heap tuple itself, leaving the ctid field to
* contain a speculative token value - a per-backend monotonically
* increasing identifier. Besides, it does not WAL-log ctid under
* any circumstances.
*
* During redo, heap_xlog_insert() sets t_ctid to current block
* number and self offset number. It doesn't care about any
code: replace 'master' with 'primary' where appropriate. Also changed "in the primary" to "on the primary", and added a few "the" before "primary". Author: Andres Freund Reviewed-By: David Steele Discussion: https://postgr.es/m/20200615182235.x7lch5n6kcjq4aue@alap3.anarazel.de
2020-06-14 23:05:18 +02:00
* speculative insertions on the primary. Hence, we set t_ctid to
Add WAL consistency checking facility. When the new GUC wal_consistency_checking is set to a non-empty value, it triggers recording of additional full-page images, which are compared on the standby against the results of applying the WAL record (without regard to those full-page images). Allowable differences such as hints are masked out, and the resulting pages are compared; any difference results in a FATAL error on the standby. Kuntal Ghosh, based on earlier patches by Michael Paquier and Heikki Linnakangas. Extensively reviewed and revised by Michael Paquier and by me, with additional reviews and comments from Amit Kapila, Álvaro Herrera, Simon Riggs, and Peter Eisentraut.
2017-02-08 21:45:30 +01:00
* current block number and self offset number to ignore any
* inconsistency.
*/
if (HeapTupleHeaderIsSpeculative(page_htup))
ItemPointerSet(&page_htup->t_ctid, blkno, off);
Raise error when affecting tuple moved into different partition. When an update moves a row between partitions (supported since 2f178441044b), our normal logic for following update chains in READ COMMITTED mode doesn't work anymore. Cross partition updates are modeled as an delete from the old and insert into the new partition. No ctid chain exists across partitions, and there's no convenient space to introduce that link. Not throwing an error in a partitioned context when one would have been thrown without partitioning is obviously problematic. This commit introduces infrastructure to detect when a tuple has been moved, not just plainly deleted. That allows to throw an error when encountering a deletion that's actually a move, while attempting to following a ctid chain. The row deleted as part of a cross partition update is marked by pointing it's t_ctid to an invalid block, instead of self as a normal update would. That was deemed to be the least invasive and most future proof way to represent the knowledge, given how few infomask bits are there to be recycled (there's also some locking issues with using infomask bits). External code following ctid chains should be updated to check for moved tuples. The most likely consequence of not doing so is a missed error. Author: Amul Sul, editorialized by me Reviewed-By: Amit Kapila, Pavan Deolasee, Andres Freund, Robert Haas Discussion: http://postgr.es/m/CAAJ_b95PkwojoYfz0bzXU8OokcTVGzN6vYGCNVUukeUDrnF3dw@mail.gmail.com
2018-04-07 22:24:10 +02:00
/*
* NB: Not ignoring ctid changes due to the tuple having moved
* (i.e. HeapTupleHeaderIndicatesMovedPartitions), because that's
* important information that needs to be in-sync between primary
* and standby, and thus is WAL logged.
*/
Add WAL consistency checking facility. When the new GUC wal_consistency_checking is set to a non-empty value, it triggers recording of additional full-page images, which are compared on the standby against the results of applying the WAL record (without regard to those full-page images). Allowable differences such as hints are masked out, and the resulting pages are compared; any difference results in a FATAL error on the standby. Kuntal Ghosh, based on earlier patches by Michael Paquier and Heikki Linnakangas. Extensively reviewed and revised by Michael Paquier and by me, with additional reviews and comments from Amit Kapila, Álvaro Herrera, Simon Riggs, and Peter Eisentraut.
2017-02-08 21:45:30 +01:00
}
/*
* Ignore any padding bytes after the tuple, when the length of the
* item is not MAXALIGNed.
*/
if (ItemIdHasStorage(iid))
{
int len = ItemIdGetLength(iid);
int padlen = MAXALIGN(len) - len;
if (padlen > 0)
memset(page_item + len, MASK_MARKER, padlen);
}
}
}
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
/*
* HeapCheckForSerializableConflictOut
Improve comments for [Heap]CheckForSerializableConflictOut(). Rewrite the documentation of these functions, in light of recent bug fix commit 5940ffb2. Back-patch to 13 where the check-for-conflict-out code was split up into AM-specific and generic parts, and new documentation was added that now looked wrong. Reviewed-by: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/db7b729d-0226-d162-a126-8a8ab2dc4443%40jepsen.io
2020-06-12 00:44:32 +02:00
* We are reading a tuple. If it's not visible, there may be a
* rw-conflict out with the inserter. Otherwise, if it is visible to us
* but has been deleted, there may be a rw-conflict out with the deleter.
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
*
* We will determine the top level xid of the writing transaction with which
Improve comments for [Heap]CheckForSerializableConflictOut(). Rewrite the documentation of these functions, in light of recent bug fix commit 5940ffb2. Back-patch to 13 where the check-for-conflict-out code was split up into AM-specific and generic parts, and new documentation was added that now looked wrong. Reviewed-by: Peter Geoghegan <pg@bowt.ie> Discussion: https://postgr.es/m/db7b729d-0226-d162-a126-8a8ab2dc4443%40jepsen.io
2020-06-12 00:44:32 +02:00
* we may be in conflict, and ask CheckForSerializableConflictOut() to check
* for overlap with our own transaction.
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
*
* This function should be called just about anywhere in heapam.c where a
* tuple has been read. The caller must hold at least a shared lock on the
* buffer, because this function might set hint bits on the tuple. There is
* currently no known reason to call this function from an index AM.
*/
void
HeapCheckForSerializableConflictOut(bool visible, Relation relation,
HeapTuple tuple, Buffer buffer,
Snapshot snapshot)
{
TransactionId xid;
HTSV_Result htsvResult;
if (!CheckForSerializableConflictOutNeeded(relation, snapshot))
return;
/*
* Check to see whether the tuple has been written to by a concurrent
* transaction, either to create it not visible to us, or to delete it
* while it is visible to us. The "visible" bool indicates whether the
* tuple is visible to us, while HeapTupleSatisfiesVacuum checks what else
* is going on with it.
Avoid update conflict out serialization anomalies. SSI's HeapCheckForSerializableConflictOut() test failed to correctly handle conditions involving a concurrently inserted tuple which is later concurrently updated by a separate transaction . A SELECT statement that called HeapCheckForSerializableConflictOut() could end up using the same XID (updater's XID) for both the original tuple, and the successor tuple, missing the XID of the xact that created the original tuple entirely. This only happened when neither tuple from the chain was visible to the transaction's MVCC snapshot. The observable symptoms of this bug were subtle. A pair of transactions could commit, with the later transaction failing to observe the effects of the earlier transaction (because of the confusion created by the update to the non-visible row). This bug dates all the way back to commit dafaa3ef, which added SSI. To fix, make sure that we check the xmin of concurrently inserted tuples that happen to also have been updated concurrently. Author: Peter Geoghegan Reported-By: Kyle Kingsbury Reviewed-By: Thomas Munro Discussion: https://postgr.es/m/db7b729d-0226-d162-a126-8a8ab2dc4443@jepsen.io Backpatch: All supported versions
2020-06-11 19:09:47 +02:00
*
* In the event of a concurrently inserted tuple that also happens to have
* been concurrently updated (by a separate transaction), the xmin of the
* tuple will be used -- not the updater's xid.
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
*/
htsvResult = HeapTupleSatisfiesVacuum(tuple, TransactionXmin, buffer);
switch (htsvResult)
{
case HEAPTUPLE_LIVE:
if (visible)
return;
xid = HeapTupleHeaderGetXmin(tuple->t_data);
break;
case HEAPTUPLE_RECENTLY_DEAD:
case HEAPTUPLE_DELETE_IN_PROGRESS:
Avoid update conflict out serialization anomalies. SSI's HeapCheckForSerializableConflictOut() test failed to correctly handle conditions involving a concurrently inserted tuple which is later concurrently updated by a separate transaction . A SELECT statement that called HeapCheckForSerializableConflictOut() could end up using the same XID (updater's XID) for both the original tuple, and the successor tuple, missing the XID of the xact that created the original tuple entirely. This only happened when neither tuple from the chain was visible to the transaction's MVCC snapshot. The observable symptoms of this bug were subtle. A pair of transactions could commit, with the later transaction failing to observe the effects of the earlier transaction (because of the confusion created by the update to the non-visible row). This bug dates all the way back to commit dafaa3ef, which added SSI. To fix, make sure that we check the xmin of concurrently inserted tuples that happen to also have been updated concurrently. Author: Peter Geoghegan Reported-By: Kyle Kingsbury Reviewed-By: Thomas Munro Discussion: https://postgr.es/m/db7b729d-0226-d162-a126-8a8ab2dc4443@jepsen.io Backpatch: All supported versions
2020-06-11 19:09:47 +02:00
if (visible)
xid = HeapTupleHeaderGetUpdateXid(tuple->t_data);
else
xid = HeapTupleHeaderGetXmin(tuple->t_data);
if (TransactionIdPrecedes(xid, TransactionXmin))
{
/* This is like the HEAPTUPLE_DEAD case */
Assert(!visible);
return;
}
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
break;
case HEAPTUPLE_INSERT_IN_PROGRESS:
xid = HeapTupleHeaderGetXmin(tuple->t_data);
break;
case HEAPTUPLE_DEAD:
Avoid update conflict out serialization anomalies. SSI's HeapCheckForSerializableConflictOut() test failed to correctly handle conditions involving a concurrently inserted tuple which is later concurrently updated by a separate transaction . A SELECT statement that called HeapCheckForSerializableConflictOut() could end up using the same XID (updater's XID) for both the original tuple, and the successor tuple, missing the XID of the xact that created the original tuple entirely. This only happened when neither tuple from the chain was visible to the transaction's MVCC snapshot. The observable symptoms of this bug were subtle. A pair of transactions could commit, with the later transaction failing to observe the effects of the earlier transaction (because of the confusion created by the update to the non-visible row). This bug dates all the way back to commit dafaa3ef, which added SSI. To fix, make sure that we check the xmin of concurrently inserted tuples that happen to also have been updated concurrently. Author: Peter Geoghegan Reported-By: Kyle Kingsbury Reviewed-By: Thomas Munro Discussion: https://postgr.es/m/db7b729d-0226-d162-a126-8a8ab2dc4443@jepsen.io Backpatch: All supported versions
2020-06-11 19:09:47 +02:00
Assert(!visible);
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
return;
default:
/*
* The only way to get to this default clause is if a new value is
* added to the enum type without adding it to this switch
* statement. That's a bug, so elog.
*/
elog(ERROR, "unrecognized return value from HeapTupleSatisfiesVacuum: %u", htsvResult);
/*
* In spite of having all enum values covered and calling elog on
* this default, some compilers think this is a code path which
* allows xid to be used below without initialization. Silence
* that warning.
*/
xid = InvalidTransactionId;
}
Assert(TransactionIdIsValid(xid));
Assert(TransactionIdFollowsOrEquals(xid, TransactionXmin));
/*
* Find top level xid. Bail out if xid is too early to be a conflict, or
* if it's our own xid.
*/
if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
return;
xid = SubTransGetTopmostTransaction(xid);
if (TransactionIdPrecedes(xid, TransactionXmin))
return;
Fix compile error on HP C. Per build farm animal anole, after commit 6f38d4dac3.
2020-01-28 08:30:40 +01:00
CheckForSerializableConflictOut(relation, xid, snapshot);
Remove dependency on HeapTuple from predicate locking functions. The following changes make the predicate locking functions more generic and suitable for use by future access methods: - PredicateLockTuple() is renamed to PredicateLockTID(). It takes ItemPointer and inserting transaction ID instead of HeapTuple. - CheckForSerializableConflictIn() takes blocknum instead of buffer. - CheckForSerializableConflictOut() no longer takes HeapTuple or buffer. Author: Ashwin Agrawal Reviewed-by: Andres Freund, Kuntal Ghosh, Thomas Munro Discussion: https://postgr.es/m/CALfoeiv0k3hkEb3Oqk%3DziWqtyk2Jys1UOK5hwRBNeANT_yX%2Bng%40mail.gmail.com
2020-01-28 01:13:04 +01:00
}