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/transam/multixact.c

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

3422 lines
111 KiB
C
Raw Normal View History

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
/*-------------------------------------------------------------------------
*
* multixact.c
* PostgreSQL multi-transaction-log manager
*
Rename "pg_clog" directory to "pg_xact". Names containing the letters "log" sometimes confuse users into believing that only non-critical data is present. It is hoped this renaming will discourage ill-considered removals of transaction status data. Michael Paquier Discussion: http://postgr.es/m/CA+Tgmoa9xFQyjRZupbdEFuwUerFTvC6HjZq1ud6GYragGDFFgA@mail.gmail.com
2017-03-17 14:46:58 +01:00
* The pg_multixact manager is a pg_xact-like manager that stores an array of
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
* MultiXactMember for each MultiXactId. It is a fundamental part of the
* shared-row-lock implementation. Each MultiXactMember is comprised of a
* TransactionId and a set of flag bits. The name is a bit historical:
* originally, a MultiXactId consisted of more than one TransactionId (except
* in rare corner cases), hence "multi". Nowadays, however, it's perfectly
* legitimate to have MultiXactIds that only include a single Xid.
*
* The meaning of the flag bits is opaque to this module, but they are mostly
* used in heapam.c to identify lock modes that each of the member transactions
* is holding on any given tuple. This module just contains support to store
* and retrieve the arrays.
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
*
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
* We use two SLRU areas, one for storing the offsets at which the 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
* starts for each MultiXactId in the other one. This trick allows us to
* store variable length arrays of TransactionIds. (We could alternatively
* use one area containing counts and TransactionIds, with valid MultiXactId
* values pointing at slots containing counts; but that way seems less robust
* since it would get completely confused if someone inquired about a bogus
* MultiXactId that pointed to an intermediate slot containing an XID.)
*
Fix comments about WAL rule "write xlog before data" versus pg_multixact. Recovery does not achieve its goal of zeroing all pg_multixact entries whose accompanying WAL records never reached disk. Remove that claim and justify its expendability. Detail the need for TrimMultiXact(), which has little in common with the TrimCLOG() rationale. Merge two tightly-related comments. Stop presenting pg_multixact as specific to heap_lock_tuple(); PostgreSQL 9.3 extended its use to heap_update(). Noticed while investigating a report from Andres Freund.
2016-01-01 07:46:46 +01:00
* XLOG interactions: this module generates a record whenever a new OFFSETs or
* MEMBERs page is initialized to zeroes, as well as an
* XLOG_MULTIXACT_CREATE_ID record whenever a new MultiXactId is defined.
* This module ignores the WAL rule "write xlog before data," because it
* suffices that actions recording a MultiXactId in a heap xmax do follow that
* rule. The only way for the MXID to be referenced from any data page is for
* heap_lock_tuple() or heap_update() to have put it there, and each generates
* an XLOG record that must follow ours. The normal LSN interlock between the
* data page and that XLOG record will ensure that our XLOG record reaches
* disk first. If the SLRU members/offsets data reaches disk sooner than the
* XLOG records, we do not care; after recovery, no xmax will refer to it. On
* the flip side, to ensure that all referenced entries _do_ reach disk, this
* module's XLOG records completely rebuild the data entered since the last
* checkpoint. We flush and sync all dirty OFFSETs and MEMBERs pages to disk
* before each checkpoint is considered complete.
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
*
* Like clog.c, and unlike subtrans.c, we have to preserve state across
* crashes and ensure that MXID and offset numbering increases monotonically
* across a crash. We do this in the same way as it's done for transaction
* IDs: the WAL record is guaranteed to contain evidence of every MXID we
* could need to worry about, and we just make sure that at the end of
* replay, the next-MXID and next-offset counters are at least as large as
* anything we saw during replay.
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
* We are able to remove segments no longer necessary by carefully tracking
Have multixact be truncated by checkpoint, not vacuum Instead of truncating pg_multixact at vacuum time, do it only at checkpoint time. The reason for doing it this way is twofold: first, we want it to delete only segments that we're certain will not be required if there's a crash immediately after the removal; and second, we want to do it relatively often so that older files are not left behind if there's an untimely crash. Per my proposal in http://www.postgresql.org/message-id/20140626044519.GJ7340@eldon.alvh.no-ip.org we now execute the truncation in the checkpointer process rather than as part of vacuum. Vacuum is in only charge of maintaining in shared memory the value to which it's possible to truncate the files; that value is stored as part of checkpoints also, and so upon recovery we can reuse the same value to re-execute truncate and reset the oldest-value-still-safe-to-use to one known to remain after truncation. Per bug reported by Jeff Janes in the course of his tests involving bug #8673. While at it, update some comments that hadn't been updated since multixacts were changed. Backpatch to 9.3, where persistency of pg_multixact files was introduced by commit 0ac5ad5134f2.
2014-06-27 20:43:53 +02:00
* each table's used values: during vacuum, any multixact older than a certain
* value is removed; the cutoff value is stored in pg_class. The minimum value
* across all tables in each database is stored in pg_database, and the global
* minimum across all databases is part of pg_control and is kept in shared
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* memory. Whenever that minimum is advanced, the SLRUs are truncated.
Have multixact be truncated by checkpoint, not vacuum Instead of truncating pg_multixact at vacuum time, do it only at checkpoint time. The reason for doing it this way is twofold: first, we want it to delete only segments that we're certain will not be required if there's a crash immediately after the removal; and second, we want to do it relatively often so that older files are not left behind if there's an untimely crash. Per my proposal in http://www.postgresql.org/message-id/20140626044519.GJ7340@eldon.alvh.no-ip.org we now execute the truncation in the checkpointer process rather than as part of vacuum. Vacuum is in only charge of maintaining in shared memory the value to which it's possible to truncate the files; that value is stored as part of checkpoints also, and so upon recovery we can reuse the same value to re-execute truncate and reset the oldest-value-still-safe-to-use to one known to remain after truncation. Per bug reported by Jeff Janes in the course of his tests involving bug #8673. While at it, update some comments that hadn't been updated since multixacts were changed. Backpatch to 9.3, where persistency of pg_multixact files was introduced by commit 0ac5ad5134f2.
2014-06-27 20:43:53 +02:00
*
* When new multixactid values are to be created, care is taken that the
* counter does not fall within the wraparound horizon considering the global
* minimum value.
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
*
Update copyright for 2023 Backpatch-through: 11
2023-01-02 21:00:37 +01:00
* Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
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
* Portions Copyright (c) 1994, Regents of the University of California
*
Remove cvs keywords from all files.
2010-09-20 22:08:53 +02:00
* src/backend/access/transam/multixact.c
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 "postgres.h"
#include "access/multixact.h"
#include "access/slru.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"
Fix an old bug in multixact and two-phase commit. Prepared transactions can be part of multixacts, so allocate a slot for each prepared transaction in the "oldest member" array in multixact.c. On PREPARE TRANSACTION, transfer the oldest member value from the current backends slot to the prepared xact slot. Also save and recover the value from the 2pc state file. The symptom of the bug was that after a transaction prepared, a shared lock still held by the prepared transaction was sometimes ignored by other transactions. Fix back to 8.1, where both 2PC and multixact were introduced.
2009-11-23 10:58:36 +01:00
#include "access/twophase.h"
#include "access/twophase_rmgr.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/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/xloginsert.h"
Move InRecovery and standbyState global vars to xlogutils.c. They are used in code that runs both during normal operation and during WAL replay, and needs to behave differently during replay. Move them to xlogutils.c, because that's where we have other helper functions used by redo routines. Reviewed-by: Andres Freund Discussion: https://www.postgresql.org/message-id/b3b71061-4919-e882-4857-27e370ab134a%40iki.fi
2021-07-31 08:50:26 +02:00
#include "access/xlogutils.h"
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
#include "catalog/pg_type.h"
#include "commands/dbcommands.h"
#include "funcapi.h"
Rework MultiXactId cache code The original performs too poorly; in some scenarios it shows way too high while profiling. Try to make it a bit smarter to avoid excessive cosst. In particular, make it have a maximum size, and have entries be sorted in LRU order; once the max size is reached, evict the oldest entry to avoid it from growing too large. Per complaint from Andres Freund in connection with new tuple freezing code.
2013-12-13 21:16:25 +01:00
#include "lib/ilist.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 "miscadmin.h"
Add a few more DTrace probes to the backend. Robert Lor
2008-08-01 15:16:09 +02:00
#include "pg_trace.h"
Replace hardcoded 200000000 with autovacuum_freeze_max_age Parts of the code used autovacuum_freeze_max_age to determine whether anti-multixact-wraparound vacuums are necessary, while others used a hardcoded 200000000 value. This leads to problems when autovacuum_freeze_max_age is set to a non-default value. Use the latter everywhere. Backpatch to 9.3, where vacuuming of multixacts was introduced. Andres Freund
2013-11-28 20:45:29 +01:00
#include "postmaster/autovacuum.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 "storage/lmgr.h"
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
#include "storage/pmsignal.h"
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
#include "storage/proc.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"
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 "utils/builtins.h"
Add a few more DTrace probes to the backend. Robert Lor
2008-08-01 15:16:09 +02:00
#include "utils/memutils.h"
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
#include "utils/snapmgr.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
/*
* Defines for MultiXactOffset page sizes. A page is the same BLCKSZ as is
* used everywhere else in Postgres.
*
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: because MultiXactOffsets are 32 bits and wrap around at 0xFFFFFFFF,
* MultiXact page numbering also wraps around at
* 0xFFFFFFFF/MULTIXACT_OFFSETS_PER_PAGE, and segment numbering at
Have multixact be truncated by checkpoint, not vacuum Instead of truncating pg_multixact at vacuum time, do it only at checkpoint time. The reason for doing it this way is twofold: first, we want it to delete only segments that we're certain will not be required if there's a crash immediately after the removal; and second, we want to do it relatively often so that older files are not left behind if there's an untimely crash. Per my proposal in http://www.postgresql.org/message-id/20140626044519.GJ7340@eldon.alvh.no-ip.org we now execute the truncation in the checkpointer process rather than as part of vacuum. Vacuum is in only charge of maintaining in shared memory the value to which it's possible to truncate the files; that value is stored as part of checkpoints also, and so upon recovery we can reuse the same value to re-execute truncate and reset the oldest-value-still-safe-to-use to one known to remain after truncation. Per bug reported by Jeff Janes in the course of his tests involving bug #8673. While at it, update some comments that hadn't been updated since multixacts were changed. Backpatch to 9.3, where persistency of pg_multixact files was introduced by commit 0ac5ad5134f2.
2014-06-27 20:43:53 +02:00
* 0xFFFFFFFF/MULTIXACT_OFFSETS_PER_PAGE/SLRU_PAGES_PER_SEGMENT. 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
* take no explicit notice of that fact in this module, except when comparing
* segment and page numbers in TruncateMultiXact (see
* MultiXactOffsetPagePrecedes).
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
/* We need four bytes per offset */
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
#define MULTIXACT_OFFSETS_PER_PAGE (BLCKSZ / sizeof(MultiXactOffset))
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
#define MultiXactIdToOffsetPage(xid) \
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
((xid) / (MultiXactOffset) MULTIXACT_OFFSETS_PER_PAGE)
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
#define MultiXactIdToOffsetEntry(xid) \
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
((xid) % (MultiXactOffset) MULTIXACT_OFFSETS_PER_PAGE)
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
#define MultiXactIdToOffsetSegment(xid) (MultiXactIdToOffsetPage(xid) / SLRU_PAGES_PER_SEGMENT)
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
/*
* The situation for members is a bit more complex: we store one byte of
* additional flag bits for each TransactionId. To do this without getting
* into alignment issues, we store four bytes of flags, and then the
* corresponding 4 Xids. Each such 5-word (20-byte) set we call a "group", and
* are stored as a whole in pages. Thus, with 8kB BLCKSZ, we keep 409 groups
* per page. This wastes 12 bytes per page, but that's OK -- simplicity (and
* performance) trumps space efficiency here.
*
* Note that the "offset" macros work with byte offset, not array indexes, so
* arithmetic must be done using "char *" pointers.
*/
/* We need eight bits per xact, so one xact fits in a byte */
#define MXACT_MEMBER_BITS_PER_XACT 8
#define MXACT_MEMBER_FLAGS_PER_BYTE 1
#define MXACT_MEMBER_XACT_BITMASK ((1 << MXACT_MEMBER_BITS_PER_XACT) - 1)
/* how many full bytes of flags are there in a group? */
#define MULTIXACT_FLAGBYTES_PER_GROUP 4
#define MULTIXACT_MEMBERS_PER_MEMBERGROUP \
(MULTIXACT_FLAGBYTES_PER_GROUP * MXACT_MEMBER_FLAGS_PER_BYTE)
/* size in bytes of a complete group */
#define MULTIXACT_MEMBERGROUP_SIZE \
(sizeof(TransactionId) * MULTIXACT_MEMBERS_PER_MEMBERGROUP + MULTIXACT_FLAGBYTES_PER_GROUP)
#define MULTIXACT_MEMBERGROUPS_PER_PAGE (BLCKSZ / MULTIXACT_MEMBERGROUP_SIZE)
#define MULTIXACT_MEMBERS_PER_PAGE \
(MULTIXACT_MEMBERGROUPS_PER_PAGE * MULTIXACT_MEMBERS_PER_MEMBERGROUP)
Wrap multixact/members correctly during extension, take 2 In a50d97625497b7 I already changed this, but got it wrong for the case where the number of members is larger than the number of entries that fit in the last page of the last segment. As reported by Serge Negodyuck in a followup to bug #8673.
2014-06-09 21:17:23 +02:00
/*
* Because the number of items per page is not a divisor of the last item
* number (member 0xFFFFFFFF), the last segment does not use the maximum number
* of pages, and moreover the last used page therein does not use the same
* number of items as previous pages. (Another way to say it is that the
* 0xFFFFFFFF member is somewhere in the middle of the last page, so the page
* has some empty space after that item.)
*
* This constant is the number of members in the last page of the last segment.
*/
#define MAX_MEMBERS_IN_LAST_MEMBERS_PAGE \
((uint32) ((0xFFFFFFFF % MULTIXACT_MEMBERS_PER_PAGE) + 1))
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
/* page in which a member is to be found */
#define MXOffsetToMemberPage(xid) ((xid) / (TransactionId) MULTIXACT_MEMBERS_PER_PAGE)
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
#define MXOffsetToMemberSegment(xid) (MXOffsetToMemberPage(xid) / SLRU_PAGES_PER_SEGMENT)
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
/* Location (byte offset within page) of flag word for a given member */
#define MXOffsetToFlagsOffset(xid) \
((((xid) / (TransactionId) MULTIXACT_MEMBERS_PER_MEMBERGROUP) % \
(TransactionId) MULTIXACT_MEMBERGROUPS_PER_PAGE) * \
(TransactionId) MULTIXACT_MEMBERGROUP_SIZE)
#define MXOffsetToFlagsBitShift(xid) \
(((xid) % (TransactionId) MULTIXACT_MEMBERS_PER_MEMBERGROUP) * \
MXACT_MEMBER_BITS_PER_XACT)
/* Location (byte offset within page) of TransactionId of given member */
#define MXOffsetToMemberOffset(xid) \
(MXOffsetToFlagsOffset(xid) + MULTIXACT_FLAGBYTES_PER_GROUP + \
((xid) % MULTIXACT_MEMBERS_PER_MEMBERGROUP) * sizeof(TransactionId))
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
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
/* Multixact members wraparound thresholds. */
Increase threshold for multixact member emergency autovac to 50%. Analysis by Noah Misch shows that the 25% threshold set by commit 53bb309d2d5a9432d2602c93ed18e58bd2924e15 is lower than any other, similar autovac threshold. While we don't know exactly what value will be optimal for all users, it is better to err a little on the high side than on the low side. A higher value increases the risk that users might exhaust the available space and start seeing errors before autovacuum can clean things up sufficiently, but a user who hits that problem can compensate for it by reducing autovacuum_multixact_freeze_max_age to a value dependent on their average multixact size. On the flip side, if the emergency cap imposed by that patch kicks in too early, the user will experience excessive wraparound scanning and will be unable to mitigate that problem by configuration. The new value will hopefully reduce the risk of such bad experiences while still providing enough headroom to avoid multixact member exhaustion for most users. Along the way, adjust the documentation to reflect the effects of commit 04e6d3b877e060d8445eb653b7ea26b1ee5cec6b, which taught autovacuum to run for multixact wraparound even when autovacuum is configured off.
2015-05-11 18:07:13 +02:00
#define MULTIXACT_MEMBER_SAFE_THRESHOLD (MaxMultiXactOffset / 2)
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
#define MULTIXACT_MEMBER_DANGER_THRESHOLD \
(MaxMultiXactOffset - MaxMultiXactOffset / 4)
Fix off-by-one error in calculating subtrans/multixact truncation point. If there were no subtransactions (or multixacts) active, we would calculate the oldestxid == next xid. That's correct, but if next XID happens to be on the next pg_subtrans (pg_multixact) page, the page does not exist yet, and SimpleLruTruncate will produce an "apparent wraparound" warning. The warning is harmless in this case, but looks very alarming to users. Backpatch to all supported versions. Patch and analysis by Thomas Munro.
2015-07-23 00:29:59 +02:00
#define PreviousMultiXactId(xid) \
((xid) == FirstMultiXactId ? MaxMultiXactId : (xid) - 1)
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
/*
* Links to shared-memory data structures for MultiXact control
*/
static SlruCtlData MultiXactOffsetCtlData;
static SlruCtlData MultiXactMemberCtlData;
#define MultiXactOffsetCtl (&MultiXactOffsetCtlData)
#define MultiXactMemberCtl (&MultiXactMemberCtlData)
/*
* MultiXact state shared across all backends. All this state is protected
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
* by MultiXactGenLock. (We also use MultiXactOffsetSLRULock and
* MultiXactMemberSLRULock to guard accesses to the two sets of SLRU
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
* buffers. For concurrency's sake, we avoid holding more than one of these
* locks at a time.)
*/
typedef struct MultiXactStateData
{
/* next-to-be-assigned MultiXactId */
MultiXactId nextMXact;
/* next-to-be-assigned offset */
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
MultiXactOffset nextOffset;
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
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
/* Have we completed multixact startup? */
bool finishedStartup;
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
/*
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
* Oldest multixact that is still potentially referenced by a relation.
* Anything older than this should not be consulted. These values are
* updated by vacuum.
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
*/
MultiXactId oldestMultiXactId;
Oid oldestMultiXactDB;
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
/*
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* Oldest multixact offset that is potentially referenced by a multixact
* referenced by a relation. We don't always know this value, so there's
* a flag here to indicate whether or not we currently do.
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
*/
Even when autovacuum=off, force it for members as we do in other cases. Thomas Munro, with some adjustments by me.
2015-05-11 16:51:14 +02:00
MultiXactOffset oldestOffset;
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
bool oldestOffsetKnown;
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
/* support for anti-wraparound measures */
MultiXactId multiVacLimit;
MultiXactId multiWarnLimit;
MultiXactId multiStopLimit;
MultiXactId multiWrapLimit;
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
/* support for members anti-wraparound measures */
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
MultiXactOffset offsetStopLimit; /* known if oldestOffsetKnown */
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +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
/*
* Per-backend data starts here. We have two arrays stored in the area
* immediately following the MultiXactStateData struct. Each is indexed by
Fix an old bug in multixact and two-phase commit. Prepared transactions can be part of multixacts, so allocate a slot for each prepared transaction in the "oldest member" array in multixact.c. On PREPARE TRANSACTION, transfer the oldest member value from the current backends slot to the prepared xact slot. Also save and recover the value from the 2pc state file. The symptom of the bug was that after a transaction prepared, a shared lock still held by the prepared transaction was sometimes ignored by other transactions. Fix back to 8.1, where both 2PC and multixact were introduced.
2009-11-23 10:58:36 +01:00
* BackendId.
*
* In both arrays, there's a slot for all normal backends (1..MaxBackends)
* followed by a slot for max_prepared_xacts prepared transactions. Valid
* BackendIds start from 1; element zero of each array is never used.
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
*
* OldestMemberMXactId[k] is the oldest MultiXactId each backend's current
* transaction(s) could possibly be a member of, or InvalidMultiXactId
* when the backend has no live transaction that could possibly be a
* member of a MultiXact. Each backend sets its entry to the current
* nextMXact counter just before first acquiring a shared lock in a given
* transaction, and clears it at transaction end. (This works because only
* during or after acquiring a shared lock could an XID possibly become a
* member of a MultiXact, and that MultiXact would have to be created
* during or after the lock acquisition.)
*
* OldestVisibleMXactId[k] is the oldest MultiXactId each backend's
* current transaction(s) think is potentially live, or InvalidMultiXactId
* when not in a transaction or not in a transaction that's paid any
* attention to MultiXacts yet. This is computed when first needed in a
* given transaction, and cleared at transaction end. We can compute it
* as the minimum of the valid OldestMemberMXactId[] entries at the time
* we compute it (using nextMXact if none are valid). Each backend is
* required not to attempt to access any SLRU data for MultiXactIds older
* than its own OldestVisibleMXactId[] setting; this is necessary because
Update more obsolete multixact.c comments. Update some remaining comments in multixact.c that still described SLRU truncation as happening in the checkpointer, rather than during VACUUM. Follow-up to commit 5212d447. Shi yu, with tweaks by me. Author: Shi yu <shiy.fnst@fujitsu.com> Discussion: https://postgr.es/m/OSZPR01MB631066BF246F8F74E83222FCFDC69@OSZPR01MB6310.jpnprd01.prod.outlook.com
2023-01-25 00:15:33 +01:00
* the relevant SLRU data can be concurrently truncated away.
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
*
Have multixact be truncated by checkpoint, not vacuum Instead of truncating pg_multixact at vacuum time, do it only at checkpoint time. The reason for doing it this way is twofold: first, we want it to delete only segments that we're certain will not be required if there's a crash immediately after the removal; and second, we want to do it relatively often so that older files are not left behind if there's an untimely crash. Per my proposal in http://www.postgresql.org/message-id/20140626044519.GJ7340@eldon.alvh.no-ip.org we now execute the truncation in the checkpointer process rather than as part of vacuum. Vacuum is in only charge of maintaining in shared memory the value to which it's possible to truncate the files; that value is stored as part of checkpoints also, and so upon recovery we can reuse the same value to re-execute truncate and reset the oldest-value-still-safe-to-use to one known to remain after truncation. Per bug reported by Jeff Janes in the course of his tests involving bug #8673. While at it, update some comments that hadn't been updated since multixacts were changed. Backpatch to 9.3, where persistency of pg_multixact files was introduced by commit 0ac5ad5134f2.
2014-06-27 20:43:53 +02:00
* The oldest valid value among all of the OldestMemberMXactId[] and
* OldestVisibleMXactId[] entries is considered by vacuum as the earliest
Update obsolete multixact.c comments. Commit 4f627f89 switched SLRU truncation for multixacts back to being a task performed during VACUUM, but missed some comments that continued to reference truncation happening as part of checkpointing. Update those comments now. Also update comments that became obsolete when commit c3ffa731 changed the way that vacuum_multixact_freeze_min_age is applied by VACUUM as it computes its MultiXactCutoff cutoff (which is used by VACUUM to decide what to freeze). Explain the same issues by referencing how OldestMxact is the latest valid value that relminmxid can ever be advanced to at the end of a VACUUM (following the work in commit 0b018fab).
2023-01-04 01:54:35 +01:00
* possible value still having any live member transaction -- OldestMxact.
* Any value older than that is typically removed from tuple headers, or
* "frozen" via being replaced with a new xmax. VACUUM can sometimes even
* remove an individual MultiXact xmax whose value is >= its OldestMxact
* cutoff, though typically only when no individual member XID is still
* running. See FreezeMultiXactId for full details.
*
* Whenever VACUUM advances relminmxid, then either its OldestMxact cutoff
* or the oldest extant Multi remaining in the table is used as the new
* pg_class.relminmxid value (whichever is earlier). The minimum of all
* relminmxid values in each database is stored in pg_database.datminmxid.
* In turn, the minimum of all of those values is stored in pg_control.
* This is used as the truncation point for pg_multixact when unneeded
* segments get removed by vac_truncate_clog() during vacuuming.
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
*/
Use FLEXIBLE_ARRAY_MEMBER in some more places. Fix a batch of structs that are only visible within individual .c files. Michael Paquier
2015-02-20 23:32:01 +01:00
MultiXactId perBackendXactIds[FLEXIBLE_ARRAY_MEMBER];
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
} MultiXactStateData;
Fix an old bug in multixact and two-phase commit. Prepared transactions can be part of multixacts, so allocate a slot for each prepared transaction in the "oldest member" array in multixact.c. On PREPARE TRANSACTION, transfer the oldest member value from the current backends slot to the prepared xact slot. Also save and recover the value from the 2pc state file. The symptom of the bug was that after a transaction prepared, a shared lock still held by the prepared transaction was sometimes ignored by other transactions. Fix back to 8.1, where both 2PC and multixact were introduced.
2009-11-23 10:58:36 +01:00
/*
Revert the addition of GetMaxBackends() and related stuff. This reverts commits 0147fc7, 4567596, aa64f23, and 5ecd018. There is no longer agreement that introducing this function was the right way to address the problem. The consensus now seems to favor trying to make a correct value for MaxBackends available to mdules executing their _PG_init() functions. Nathan Bossart Discussion: http://postgr.es/m/20220323045229.i23skfscdbvrsuxa@jrouhaud
2022-04-12 20:45:23 +02:00
* Last element of OldestMemberMXactId and OldestVisibleMXactId arrays.
* Valid elements are (1..MaxOldestSlot); element 0 is never used.
Fix an old bug in multixact and two-phase commit. Prepared transactions can be part of multixacts, so allocate a slot for each prepared transaction in the "oldest member" array in multixact.c. On PREPARE TRANSACTION, transfer the oldest member value from the current backends slot to the prepared xact slot. Also save and recover the value from the 2pc state file. The symptom of the bug was that after a transaction prepared, a shared lock still held by the prepared transaction was sometimes ignored by other transactions. Fix back to 8.1, where both 2PC and multixact were introduced.
2009-11-23 10:58:36 +01:00
*/
Revert the addition of GetMaxBackends() and related stuff. This reverts commits 0147fc7, 4567596, aa64f23, and 5ecd018. There is no longer agreement that introducing this function was the right way to address the problem. The consensus now seems to favor trying to make a correct value for MaxBackends available to mdules executing their _PG_init() functions. Nathan Bossart Discussion: http://postgr.es/m/20220323045229.i23skfscdbvrsuxa@jrouhaud
2022-04-12 20:45:23 +02:00
#define MaxOldestSlot (MaxBackends + max_prepared_xacts)
/* Pointers to the state data in shared memory */
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 MultiXactStateData *MultiXactState;
static MultiXactId *OldestMemberMXactId;
static MultiXactId *OldestVisibleMXactId;
/*
* Definitions for the backend-local MultiXactId cache.
*
* We use this cache to store known MultiXacts, so we don't need to go to
Lots of doc corrections. Josh Kupershmidt
2012-04-24 04:43:09 +02:00
* SLRU areas every time.
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
*
* The cache lasts for the duration of a single transaction, the rationale
* for this being that most entries will contain our own TransactionId and
* so they will be uninteresting by the time our next transaction starts.
* (XXX not clear that this is correct --- other members of the MultiXact
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
* could hang around longer than we did. However, it's not clear what 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
* better policy for flushing old cache entries would be.) FIXME actually
* this is plain wrong now that multixact's may contain update Xids.
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 allocate the cache entries in a memory context that is deleted at
* transaction end, so we don't need to do retail freeing of entries.
*/
typedef struct mXactCacheEnt
{
MultiXactId 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
int nmembers;
Rework MultiXactId cache code The original performs too poorly; in some scenarios it shows way too high while profiling. Try to make it a bit smarter to avoid excessive cosst. In particular, make it have a maximum size, and have entries be sorted in LRU order; once the max size is reached, evict the oldest entry to avoid it from growing too large. Per complaint from Andres Freund in connection with new tuple freezing code.
2013-12-13 21:16:25 +01:00
dlist_node node;
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
MultiXactMember members[FLEXIBLE_ARRAY_MEMBER];
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
} mXactCacheEnt;
Rework MultiXactId cache code The original performs too poorly; in some scenarios it shows way too high while profiling. Try to make it a bit smarter to avoid excessive cosst. In particular, make it have a maximum size, and have entries be sorted in LRU order; once the max size is reached, evict the oldest entry to avoid it from growing too large. Per complaint from Andres Freund in connection with new tuple freezing code.
2013-12-13 21:16:25 +01:00
#define MAX_CACHE_ENTRIES 256
Add doubly linked count list implementation We have various requirements when using a dlist_head to keep track of the number of items in the list. This, traditionally, has been done by maintaining a counter variable in the calling code. Here we tidy this up by adding "dclist", which is very similar to dlist but also keeps track of the number of items stored in the list. Callers may use the new dclist_count() function when they need to know how many items are stored. Obtaining the count is an O(1) operation. For simplicity reasons, dclist and dlist both use dlist_node as their node type and dlist_iter/dlist_mutable_iter as their iterator type. dclists have all of the same functionality as dlists except there is no function named dclist_delete(). To remove an item from a list dclist_delete_from() must be used. This requires knowing which dclist the given item is stored in. Additionally, here we also convert some dlists where additional code exists to keep track of the number of items stored and to make these use dclists instead. Author: David Rowley Reviewed-by: Bharath Rupireddy, Aleksander Alekseev Discussion: https://postgr.es/m/CAApHDvrtVxr+FXEX0VbViCFKDGxA3tWDgw9oFewNXCJMmwLjLg@mail.gmail.com
2022-11-02 02:06:05 +01:00
static dclist_head MXactCache = DCLIST_STATIC_INIT(MXactCache);
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 MemoryContext MXactContext = NULL;
#ifdef MULTIXACT_DEBUG
#define debug_elog2(a,b) elog(a,b)
#define debug_elog3(a,b,c) elog(a,b,c)
#define debug_elog4(a,b,c,d) elog(a,b,c,d)
#define debug_elog5(a,b,c,d,e) elog(a,b,c,d,e)
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
#define debug_elog6(a,b,c,d,e,f) elog(a,b,c,d,e,f)
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
#define debug_elog2(a,b)
#define debug_elog3(a,b,c)
#define debug_elog4(a,b,c,d)
#define debug_elog5(a,b,c,d,e)
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
#define debug_elog6(a,b,c,d,e,f)
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
#endif
/* internal MultiXactId management */
static void MultiXactIdSetOldestVisible(void);
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
static void RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
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
int nmembers, MultiXactMember *members);
static MultiXactId GetNewMultiXactId(int nmembers, MultiXactOffset *offset);
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
/* MultiXact cache management */
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 int mxactMemberComparator(const void *arg1, const void *arg2);
static MultiXactId mXactCacheGetBySet(int nmembers, MultiXactMember *members);
static int mXactCacheGetById(MultiXactId multi, MultiXactMember **members);
static void mXactCachePut(MultiXactId multi, int nmembers,
MultiXactMember *members);
Standard pgindent run for 8.1.
2005-10-15 04:49:52 +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 char *mxstatus_to_string(MultiXactStatus status);
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
/* management of SLRU infrastructure */
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
static int ZeroMultiXactOffsetPage(int pageno, bool writeXlog);
static int ZeroMultiXactMemberPage(int pageno, bool writeXlog);
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 bool MultiXactOffsetPagePrecedes(int page1, int page2);
static bool MultiXactMemberPagePrecedes(int page1, int page2);
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
static bool MultiXactOffsetPrecedes(MultiXactOffset offset1,
MultiXactOffset offset2);
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 ExtendMultiXactOffset(MultiXactId multi);
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
static void ExtendMultiXactMember(MultiXactOffset offset, int nmembers);
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
static bool MultiXactOffsetWouldWrap(MultiXactOffset boundary,
MultiXactOffset start, uint32 distance);
Make logging about multixact wraparound protection less chatty. The original messaging design, introduced in commit 068cfadf9, seems too chatty now that some time has elapsed since the bug fix; most installations will be in good shape and don't really need a reminder about this on every postmaster start. Hence, arrange to suppress the "wraparound protections are now enabled" message during startup (specifically, during the TrimMultiXact() call). The message will still appear if protection becomes effective at some later point. Discussion: https://postgr.es/m/17211.1489189214@sss.pgh.pa.us
2017-03-14 17:47:46 +01:00
static bool SetOffsetVacuumLimit(bool is_startup);
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
static bool find_multixact_start(MultiXactId multi, MultiXactOffset *result);
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
static void WriteMZeroPageXlogRec(int pageno, uint8 info);
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
static void WriteMTruncateXlogRec(Oid oldestMultiDB,
Fix inconsistencies in the code This addresses a couple of issues in the code: - Typos and inconsistencies in comments and function declarations. - Removal of unreferenced function declarations. - Removal of unnecessary compile flags. - A cleanup error in regressplans.sh. Author: Alexander Lakhin Discussion: https://postgr.es/m/0c991fdf-2670-1997-c027-772a420c4604@gmail.com
2019-07-08 06:15:09 +02:00
MultiXactId startTruncOff,
MultiXactId endTruncOff,
MultiXactOffset startTruncMemb,
MultiXactOffset endTruncMemb);
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 MultiXactIdExpand's API by separating out the case where we are creating a new MultiXactId from two regular XIDs. The original coding was unnecessarily complicated and didn't save any code anyway.
2005-05-03 21:42:41 +02:00
* MultiXactIdCreate
* Construct a MultiXactId representing two TransactionIds.
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
* The two XIDs must be different, or be requesting different statuses.
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 MultiXactIdExpand's API by separating out the case where we are creating a new MultiXactId from two regular XIDs. The original coding was unnecessarily complicated and didn't save any code anyway.
2005-05-03 21:42:41 +02:00
* NB - we don't worry about our local MultiXactId cache here, because that
* is handled by the lower-level routines.
*/
MultiXactId
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
MultiXactIdCreate(TransactionId xid1, MultiXactStatus status1,
TransactionId xid2, MultiXactStatus status2)
Clean up MultiXactIdExpand's API by separating out the case where we are creating a new MultiXactId from two regular XIDs. The original coding was unnecessarily complicated and didn't save any code anyway.
2005-05-03 21:42:41 +02:00
{
MultiXactId newMulti;
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
MultiXactMember members[2];
Clean up MultiXactIdExpand's API by separating out the case where we are creating a new MultiXactId from two regular XIDs. The original coding was unnecessarily complicated and didn't save any code anyway.
2005-05-03 21:42:41 +02:00
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(TransactionIdIsValid(xid1));
Assert(TransactionIdIsValid(xid2));
Clean up MultiXactIdExpand's API by separating out the case where we are creating a new MultiXactId from two regular XIDs. The original coding was unnecessarily complicated and didn't save any code anyway.
2005-05-03 21:42:41 +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
Assert(!TransactionIdEquals(xid1, xid2) || (status1 != status2));
Clean up MultiXactIdExpand's API by separating out the case where we are creating a new MultiXactId from two regular XIDs. The original coding was unnecessarily complicated and didn't save any code anyway.
2005-05-03 21:42:41 +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
/* MultiXactIdSetOldestMember() must have been called already. */
Assert(MultiXactIdIsValid(OldestMemberMXactId[MyBackendId]));
Clean up MultiXactIdExpand's API by separating out the case where we are creating a new MultiXactId from two regular XIDs. The original coding was unnecessarily complicated and didn't save any code anyway.
2005-05-03 21:42:41 +02:00
/*
* Note: unlike MultiXactIdExpand, we don't bother to check that both XIDs
* are still running. In typical usage, xid2 will be our own XID and the
* caller just did a check on xid1, so it'd be wasted effort.
*/
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
members[0].xid = xid1;
members[0].status = status1;
members[1].xid = xid2;
members[1].status = status2;
Clean up MultiXactIdExpand's API by separating out the case where we are creating a new MultiXactId from two regular XIDs. The original coding was unnecessarily complicated and didn't save any code anyway.
2005-05-03 21:42:41 +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
newMulti = MultiXactIdCreateFromMembers(2, members);
Clean up MultiXactIdExpand's API by separating out the case where we are creating a new MultiXactId from two regular XIDs. The original coding was unnecessarily complicated and didn't save any code anyway.
2005-05-03 21:42:41 +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
debug_elog3(DEBUG2, "Create: %s",
mxid_to_string(newMulti, 2, members));
Clean up MultiXactIdExpand's API by separating out the case where we are creating a new MultiXactId from two regular XIDs. The original coding was unnecessarily complicated and didn't save any code anyway.
2005-05-03 21:42:41 +02:00
return newMulti;
}
/*
* MultiXactIdExpand
* Add a TransactionId to a pre-existing MultiXactId.
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 the TransactionId is already a member of the passed MultiXactId with the
* same status, just return it as-is.
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
*
* Note that we do NOT actually modify the membership of a pre-existing
* MultiXactId; instead we create a new one. This is necessary to avoid
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
* a race condition against code trying to wait for one MultiXactId to finish;
* see notes in heapam.c.
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
*
* NB - we don't worry about our local MultiXactId cache here, because that
* is handled by the lower-level routines.
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: It is critical that MultiXactIds that come from an old cluster (i.e.
* one upgraded by pg_upgrade from a cluster older than this feature) are not
* passed in.
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
*/
MultiXactId
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
MultiXactIdExpand(MultiXactId multi, TransactionId xid, MultiXactStatus status)
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
{
MultiXactId newMulti;
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
MultiXactMember *members;
MultiXactMember *newMembers;
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
int nmembers;
int i;
int j;
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(MultiXactIdIsValid(multi));
Assert(TransactionIdIsValid(xid));
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
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
/* MultiXactIdSetOldestMember() must have been called already. */
Assert(MultiXactIdIsValid(OldestMemberMXactId[MyBackendId]));
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
debug_elog5(DEBUG2, "Expand: received multi %u, xid %u status %s",
multi, xid, mxstatus_to_string(status));
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: we don't allow for old multis here. The reason is that the only
* caller of this function does a check that the multixact is no longer
* running.
*/
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);
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 (nmembers < 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
MultiXactMember member;
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
/*
* The MultiXactId is obsolete. This can only happen if all the
* MultiXactId members stop running between the caller checking and
* passing it to us. It would be better to return that fact to the
* caller, but it would complicate the API and it's unlikely to happen
* too often, so just deal with it by creating a singleton MultiXact.
*/
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
member.xid = xid;
member.status = 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
newMulti = MultiXactIdCreateFromMembers(1, &member);
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
debug_elog4(DEBUG2, "Expand: %u has no members, create singleton %u",
multi, newMulti);
return newMulti;
}
/*
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 TransactionId is already a member of the MultiXactId with the
* same status, just return the existing MultiXactId.
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
*/
for (i = 0; i < nmembers; i++)
{
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 (TransactionIdEquals(members[i].xid, xid) &&
(members[i].status == status))
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
{
debug_elog4(DEBUG2, "Expand: %u is already a member of %u",
xid, multi);
Fix case in which a debug printout would print already-pfreed data.
2005-05-07 20:14:25 +02:00
pfree(members);
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
return 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
* Determine which of the members of the MultiXactId are still of
* interest. This is any running transaction, and also any transaction
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
* that grabbed something stronger than just a lock and was committed. (An
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
* update that aborted is of no interest here; and having more than one
* update Xid in a multixact would cause errors elsewhere.)
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 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
* Removing dead members is not just an optimization: freezing of tuples
* whose Xmax are multis depends on this behavior.
*
* Note we have the same race condition here as above: j could be 0 at the
* end of the loop.
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
newMembers = (MultiXactMember *)
palloc(sizeof(MultiXactMember) * (nmembers + 1));
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
for (i = 0, j = 0; i < nmembers; i++)
{
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 (TransactionIdIsInProgress(members[i].xid) ||
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
(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
TransactionIdDidCommit(members[i].xid)))
{
newMembers[j].xid = members[i].xid;
newMembers[j++].status = members[i].status;
}
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
newMembers[j].xid = xid;
newMembers[j++].status = 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
newMulti = MultiXactIdCreateFromMembers(j, newMembers);
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
pfree(members);
pfree(newMembers);
debug_elog3(DEBUG2, "Expand: returning new multi %u", newMulti);
return newMulti;
}
/*
* MultiXactIdIsRunning
* Returns whether a MultiXactId is "running".
*
* We return true if at least one member of the given MultiXactId is still
* running. Note that a "false" result is certain not to change,
* because it is not legal to add members to an existing MultiXactId.
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
*
* Caller is expected to have verified that the multixact does not come from
* a pg_upgraded share-locked tuple.
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
*/
bool
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
MultiXactIdIsRunning(MultiXactId multi, bool isLockOnly)
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
MultiXactMember *members;
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
int nmembers;
int i;
debug_elog3(DEBUG2, "IsRunning %u?", 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
/*
* "false" here means we assume our callers have checked that the given
* multi cannot possibly come from a pg_upgraded database.
*/
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, isLockOnly);
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 (nmembers <= 0)
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
{
debug_elog2(DEBUG2, "IsRunning: no members");
return false;
}
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
* Checking for myself is cheap compared to looking in shared memory;
* return true if any live subtransaction of the current top-level
* transaction is a member.
*
* This is not needed for correctness, it's just a fast path.
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
for (i = 0; i < nmembers; i++)
{
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 (TransactionIdIsCurrentTransactionId(members[i].xid))
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
{
debug_elog3(DEBUG2, "IsRunning: I (%d) am running!", i);
Fix case in which a debug printout would print already-pfreed data.
2005-05-07 20:14:25 +02:00
pfree(members);
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
return true;
}
}
/*
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
* This could be made faster by having another entry point in procarray.c,
* walking the PGPROC array only once for all the members. But in most
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
* cases nmembers should be small enough that it doesn't much matter.
*/
for (i = 0; i < nmembers; i++)
{
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 (TransactionIdIsInProgress(members[i].xid))
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
{
debug_elog4(DEBUG2, "IsRunning: member %d (%u) is 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
i, members[i].xid);
Fix case in which a debug printout would print already-pfreed data.
2005-05-07 20:14:25 +02:00
pfree(members);
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
return true;
}
}
pfree(members);
Fix case in which a debug printout would print already-pfreed data.
2005-05-07 20:14:25 +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
debug_elog3(DEBUG2, "IsRunning: %u is not running", multi);
return false;
}
/*
* MultiXactIdSetOldestMember
* Save the oldest MultiXactId this transaction could be a member of.
*
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 set the OldestMemberMXactId for a given transaction the first time it's
* going to do some operation that might require a MultiXactId (tuple lock,
* update or delete). We need to do this even if we end up using a
* TransactionId instead of a MultiXactId, because there is a chance that
* another transaction would add our XID to a MultiXactId.
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
* The value to set is the next-to-be-assigned MultiXactId, so this is meant to
* be called just before doing any such possibly-MultiXactId-able operation.
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
*/
void
MultiXactIdSetOldestMember(void)
{
if (!MultiXactIdIsValid(OldestMemberMXactId[MyBackendId]))
{
MultiXactId nextMXact;
/*
* You might think we don't need to acquire a lock here, since
* fetching and storing of TransactionIds is probably atomic, but in
* fact we do: suppose we pick up nextMXact and then lose the CPU for
* a long time. Someone else could advance nextMXact, and then
* another someone else could compute an OldestVisibleMXactId that
* would be after the value we are going to store when we get control
* back. Which would be wrong.
Handle wraparound during truncation in multixact/members In pg_multixact/members, relying on modulo-2^32 arithmetic for wraparound handling doesn't work all that well. Because we don't explicitely track wraparound of the allocation counter for members, it is possible that the "live" area exceeds 2^31 entries; trying to remove SLRU segments that are "old" according to the original logic might lead to removal of segments still in use. To fix, have the truncation routine use a tailored SlruScanDirectory callback that keeps track of the live area in actual use; that way, when the live range exceeds 2^31 entries, the oldest segments still live will not get removed untimely. This new SlruScanDir callback needs to take care not to remove segments that are "in the future": if new SLRU segments appear while the truncation is ongoing, make sure we don't remove them. This requires examination of shared memory state to recheck for false positives, but testing suggests that this doesn't cause a problem. The original coding didn't suffer from this pitfall because segments created when truncation is running are never considered to be removable. Per Andres Freund's investigation of bug #8673 reported by Serge Negodyuck.
2014-01-02 22:16:54 +01:00
*
* Note that a shared lock is sufficient, because it's enough to stop
* someone from advancing nextMXact; and nobody else could be trying
* to write to our OldestMember entry, only reading (and we assume
* storing it is atomic.)
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
*/
Handle wraparound during truncation in multixact/members In pg_multixact/members, relying on modulo-2^32 arithmetic for wraparound handling doesn't work all that well. Because we don't explicitely track wraparound of the allocation counter for members, it is possible that the "live" area exceeds 2^31 entries; trying to remove SLRU segments that are "old" according to the original logic might lead to removal of segments still in use. To fix, have the truncation routine use a tailored SlruScanDirectory callback that keeps track of the live area in actual use; that way, when the live range exceeds 2^31 entries, the oldest segments still live will not get removed untimely. This new SlruScanDir callback needs to take care not to remove segments that are "in the future": if new SLRU segments appear while the truncation is ongoing, make sure we don't remove them. This requires examination of shared memory state to recheck for false positives, but testing suggests that this doesn't cause a problem. The original coding didn't suffer from this pitfall because segments created when truncation is running are never considered to be removable. Per Andres Freund's investigation of bug #8673 reported by Serge Negodyuck.
2014-01-02 22:16:54 +01:00
LWLockAcquire(MultiXactGenLock, LW_SHARED);
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 have to beware of the possibility that nextMXact is in the
* wrapped-around state. We don't fix the counter itself here, but we
* must be sure to store a valid value in our array entry.
*/
nextMXact = MultiXactState->nextMXact;
if (nextMXact < FirstMultiXactId)
nextMXact = FirstMultiXactId;
OldestMemberMXactId[MyBackendId] = nextMXact;
LWLockRelease(MultiXactGenLock);
debug_elog4(DEBUG2, "MultiXact: setting OldestMember[%d] = %u",
MyBackendId, nextMXact);
}
}
/*
* MultiXactIdSetOldestVisible
* Save the oldest MultiXactId this transaction considers possibly live.
*
* We set the OldestVisibleMXactId for a given transaction the first time
* it's going to inspect any MultiXactId. Once we have set this, we are
Update more obsolete multixact.c comments. Update some remaining comments in multixact.c that still described SLRU truncation as happening in the checkpointer, rather than during VACUUM. Follow-up to commit 5212d447. Shi yu, with tweaks by me. Author: Shi yu <shiy.fnst@fujitsu.com> Discussion: https://postgr.es/m/OSZPR01MB631066BF246F8F74E83222FCFDC69@OSZPR01MB6310.jpnprd01.prod.outlook.com
2023-01-25 00:15:33 +01:00
* guaranteed that SLRU data for MultiXactIds >= our own OldestVisibleMXactId
* won't be truncated away.
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
*
* The value to set is the oldest of nextMXact and all the valid per-backend
* OldestMemberMXactId[] entries. Because of the locking we do, we can be
* certain that no subsequent call to MultiXactIdSetOldestMember can set
* an OldestMemberMXactId[] entry older than what we compute here. Therefore
* there is no live transaction, now or later, that can be a member of any
* MultiXactId older than the OldestVisibleMXactId we compute here.
*/
static void
MultiXactIdSetOldestVisible(void)
{
if (!MultiXactIdIsValid(OldestVisibleMXactId[MyBackendId]))
{
MultiXactId oldestMXact;
int i;
LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
/*
* We have to beware of the possibility that nextMXact is in the
* wrapped-around state. We don't fix the counter itself here, but we
* must be sure to store a valid value in our array entry.
*/
oldestMXact = MultiXactState->nextMXact;
if (oldestMXact < FirstMultiXactId)
oldestMXact = FirstMultiXactId;
Revert the addition of GetMaxBackends() and related stuff. This reverts commits 0147fc7, 4567596, aa64f23, and 5ecd018. There is no longer agreement that introducing this function was the right way to address the problem. The consensus now seems to favor trying to make a correct value for MaxBackends available to mdules executing their _PG_init() functions. Nathan Bossart Discussion: http://postgr.es/m/20220323045229.i23skfscdbvrsuxa@jrouhaud
2022-04-12 20:45:23 +02:00
for (i = 1; i <= MaxOldestSlot; i++)
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
{
MultiXactId thisoldest = OldestMemberMXactId[i];
if (MultiXactIdIsValid(thisoldest) &&
MultiXactIdPrecedes(thisoldest, oldestMXact))
oldestMXact = thisoldest;
}
OldestVisibleMXactId[MyBackendId] = oldestMXact;
LWLockRelease(MultiXactGenLock);
debug_elog4(DEBUG2, "MultiXact: setting OldestVisible[%d] = %u",
MyBackendId, oldestMXact);
}
}
/*
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
* ReadNextMultiXactId
* Return the next MultiXactId to be assigned, but don't allocate it
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
MultiXactId
ReadNextMultiXactId(void)
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
MultiXactId mxid;
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
/* XXX we could presumably do this without a lock. */
LWLockAcquire(MultiXactGenLock, LW_SHARED);
mxid = MultiXactState->nextMXact;
LWLockRelease(MultiXactGenLock);
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
if (mxid < FirstMultiXactId)
mxid = FirstMultiXactId;
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
return mxid;
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
}
Extend amcheck to check heap pages. Mark Dilger, reviewed by Peter Geoghegan, Andres Freund, Álvaro Herrera, Michael Paquier, Amul Sul, and by me. Some last-minute cosmetic revisions by me. Discussion: http://postgr.es/m/12ED3DA8-25F0-4B68-937D-D907CFBF08E7@enterprisedb.com
2020-10-22 14:44:18 +02:00
/*
* ReadMultiXactIdRange
* Get the range of IDs that may still be referenced by a relation.
*/
void
ReadMultiXactIdRange(MultiXactId *oldest, MultiXactId *next)
{
LWLockAcquire(MultiXactGenLock, LW_SHARED);
*oldest = MultiXactState->oldestMultiXactId;
*next = MultiXactState->nextMXact;
LWLockRelease(MultiXactGenLock);
if (*oldest < FirstMultiXactId)
*oldest = FirstMultiXactId;
if (*next < FirstMultiXactId)
*next = FirstMultiXactId;
}
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
/*
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
* MultiXactIdCreateFromMembers
* Make a new MultiXactId from the specified set of members
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
*
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
* Make XLOG, SLRU and cache entries for a new MultiXactId, recording the
* given TransactionIds as members. Returns the newly created MultiXactId.
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
* NB: the passed members[] array will be sorted in-place.
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
*/
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
MultiXactId
MultiXactIdCreateFromMembers(int nmembers, MultiXactMember *members)
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
{
MultiXactId multi;
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
MultiXactOffset offset;
xl_multixact_create xlrec;
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
debug_elog3(DEBUG2, "Create: %s",
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
mxid_to_string(InvalidMultiXactId, nmembers, members));
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
* See if the same set of members already exists in our cache; if so, just
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
* re-use that MultiXactId. (Note: it might seem that looking in our
* cache is insufficient, and we ought to search disk to see if a
* duplicate definition already exists. But since we only ever create
* MultiXacts containing our own XID, in most cases any such MultiXacts
* were in fact created by us, and so will be in our cache. There are
* corner cases where someone else added us to a MultiXact without our
* knowledge, but it's not worth checking for.)
*/
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
multi = mXactCacheGetBySet(nmembers, members);
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 (MultiXactIdIsValid(multi))
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
{
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
debug_elog2(DEBUG2, "Create: in cache!");
return multi;
}
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
/* Verify that there is a single update Xid among the given members. */
{
int i;
bool has_update = false;
for (i = 0; i < nmembers; i++)
{
if (ISUPDATE_from_mxstatus(members[i].status))
{
if (has_update)
Make multixact error message more explicit There are recent reports involving a very old error message that we have no history of hitting -- perhaps a recently introduced bug. Improve the error message in an attempt to improve our chances of investigating the bug. Per reports from Dimos Stamatakis and Bob Krier. Backpatch to 11. Discussion: https://postgr.es/m/CO2PR0801MB2310579F65529380A4E5EDC0E20A9@CO2PR0801MB2310.namprd08.prod.outlook.com Discussion: https://postgr.es/m/17518-04e368df5ad7f2ee@postgresql.org
2022-11-24 10:45:10 +01:00
elog(ERROR, "new multixact has more than one updating member: %s",
mxid_to_string(InvalidMultiXactId, nmembers, members));
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
has_update = true;
}
}
}
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
/*
* Assign the MXID and offsets range to use, and make sure there is space
Reorder code so that we don't have to hold a critical section while reserving SLRU space for a new MultiXact. The original coding would have treated out-of-disk-space as a PANIC condition, which is unnecessary.
2005-10-28 21:00:19 +02:00
* in the OFFSETs and MEMBERs files. NB: this routine does
* START_CRIT_SECTION().
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: unlike MultiXactIdCreate and MultiXactIdExpand, we do not check
* that we've called MultiXactIdSetOldestMember here. This is because
* this routine is used in some places to create new MultiXactIds of which
* the current backend is not a member, notably during freezing of multis
* in vacuum. During vacuum, in particular, it would be unacceptable to
* keep OldestMulti set, in case it runs for long.
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
*/
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
multi = GetNewMultiXactId(nmembers, &offset);
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 comments about WAL rule "write xlog before data" versus pg_multixact. Recovery does not achieve its goal of zeroing all pg_multixact entries whose accompanying WAL records never reached disk. Remove that claim and justify its expendability. Detail the need for TrimMultiXact(), which has little in common with the TrimCLOG() rationale. Merge two tightly-related comments. Stop presenting pg_multixact as specific to heap_lock_tuple(); PostgreSQL 9.3 extended its use to heap_update(). Noticed while investigating a report from Andres Freund.
2016-01-01 07:46:46 +01:00
/* Make an XLOG entry describing the new MXID. */
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
xlrec.mid = multi;
xlrec.moff = offset;
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.nmembers = nmembers;
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
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
/*
* XXX Note: there's a lot of padding space in MultiXactMember. We could
* find a more compact representation of this Xlog record -- perhaps all
* the status flags in one XLogRecData, then all the xids in another one?
* Not clear that it's worth the trouble though.
*/
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), SizeOfMultiXactCreate);
XLogRegisterData((char *) members, nmembers * sizeof(MultiXactMember));
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
(void) XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_CREATE_ID);
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
/* Now enter the information into the OFFSETs and MEMBERs logs */
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
RecordNewMultiXact(multi, offset, nmembers, members);
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
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
/* Done with critical section */
END_CRIT_SECTION();
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
/* Store the new MultiXactId in the local cache, too */
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
mXactCachePut(multi, nmembers, members);
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
debug_elog2(DEBUG2, "Create: all done");
return multi;
}
/*
* RecordNewMultiXact
* Write info about a new multixact into the offsets and members files
*
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
* This is broken out of MultiXactIdCreateFromMembers so that xlog replay can
* use it.
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
*/
static void
RecordNewMultiXact(MultiXactId multi, MultiXactOffset offset,
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
int nmembers, MultiXactMember *members)
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
{
int pageno;
int prev_pageno;
int entryno;
int slotno;
MultiXactOffset *offptr;
int i;
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
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockAcquire(MultiXactOffsetSLRULock, LW_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
pageno = MultiXactIdToOffsetPage(multi);
entryno = MultiXactIdToOffsetEntry(multi);
/*
* Note: we pass the MultiXactId to SimpleLruReadPage as the "transaction"
* to complain about if there's any I/O error. This is kinda bogus, but
* since the errors will always give the full pathname, it should be clear
* enough that a MultiXactId is really involved. Perhaps someday we'll
* take the trouble to generalize the slru.c error reporting code.
*/
Support an optional asynchronous commit mode, in which we don't flush WAL before reporting a transaction committed. Data consistency is still guaranteed (unlike setting fsync = off), but a crash may lose the effects of the last few transactions. Patch by Simon, some editorialization by Tom.
2007-08-02 00:45:09 +02:00
slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
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
offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
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
offptr += entryno;
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
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
*offptr = offset;
Clean up representation of SLRU page state. This is the cleaner fix for the SLRU race condition that I posted a few days ago, but we decided not to use in 8.1 and older branches.
2005-11-05 22:19:47 +01:00
MultiXactOffsetCtl->shared->page_dirty[slotno] = 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
/* Exchange our lock */
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactOffsetSLRULock);
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
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockAcquire(MultiXactMemberSLRULock, LW_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
prev_pageno = -1;
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++, offset++)
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 *memberptr;
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
uint32 *flagsptr;
uint32 flagsval;
int bshift;
int flagsoff;
int memberoff;
Assert(members[i].status <= MultiXactStatusUpdate);
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
pageno = MXOffsetToMemberPage(offset);
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
memberoff = MXOffsetToMemberOffset(offset);
flagsoff = MXOffsetToFlagsOffset(offset);
bshift = MXOffsetToFlagsBitShift(offset);
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 (pageno != prev_pageno)
{
Support an optional asynchronous commit mode, in which we don't flush WAL before reporting a transaction committed. Data consistency is still guaranteed (unlike setting fsync = off), but a crash may lose the effects of the last few transactions. Patch by Simon, some editorialization by Tom.
2007-08-02 00:45:09 +02:00
slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
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
prev_pageno = pageno;
}
memberptr = (TransactionId *)
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
(MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
*memberptr = members[i].xid;
flagsptr = (uint32 *)
(MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
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
flagsval = *flagsptr;
flagsval &= ~(((1 << MXACT_MEMBER_BITS_PER_XACT) - 1) << bshift);
flagsval |= (members[i].status << bshift);
*flagsptr = flagsval;
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
Clean up representation of SLRU page state. This is the cleaner fix for the SLRU race condition that I posted a few days ago, but we decided not to use in 8.1 and older branches.
2005-11-05 22:19:47 +01:00
MultiXactMemberCtl->shared->page_dirty[slotno] = 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
}
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactMemberSLRULock);
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
}
/*
* GetNewMultiXactId
* Get the next MultiXactId.
*
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, reserve the needed amount of space in the "members" area. The
* starting offset of the reserved space is returned in *offset.
*
* This may generate XLOG records for expansion of the offsets and/or members
* files. Unfortunately, we have to do that while holding MultiXactGenLock
* to avoid race conditions --- the XLOG record for zeroing a page must appear
* before any backend can possibly try to store data in that page!
Reorder code so that we don't have to hold a critical section while reserving SLRU space for a new MultiXact. The original coding would have treated out-of-disk-space as a PANIC condition, which is unnecessary.
2005-10-28 21:00:19 +02:00
*
* We start a critical section before advancing the shared counters. The
* caller must end the critical section after writing SLRU 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
*/
static MultiXactId
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
GetNewMultiXactId(int nmembers, MultiXactOffset *offset)
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
{
MultiXactId result;
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
MultiXactOffset nextOffset;
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
debug_elog3(DEBUG2, "GetNew: for %d xids", nmembers);
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
Remove uses of "slave" in replication contexts This affects mostly code comments, some documentation, and tests. Official APIs already used "standby".
2017-08-07 23:42:47 +02:00
/* safety check, we should never get this far in a HS standby */
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 (RecoveryInProgress())
elog(ERROR, "cannot assign MultiXactIds during recovery");
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
LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
/* Handle wraparound of the nextMXact counter */
if (MultiXactState->nextMXact < FirstMultiXactId)
MultiXactState->nextMXact = FirstMultiXactId;
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
/* Assign the MXID */
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
result = MultiXactState->nextMXact;
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 to see if it's safe to assign another MultiXactId. This protects
* against catastrophic data loss due to multixact wraparound. The basic
* rules are:
*
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
* If we're past multiVacLimit or the safe threshold for member storage
* space, or we don't know what the safe threshold for member storage is,
Even when autovacuum=off, force it for members as we do in other cases. Thomas Munro, with some adjustments by me.
2015-05-11 16:51:14 +02:00
* start trying to force autovacuum cycles.
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 past multiWarnLimit, start issuing warnings.
* If we're past multiStopLimit, refuse to create new MultiXactIds.
*
* Note these are pretty much the same protections in GetNewTransactionId.
*----------
*/
Improve multixact emergency autovacuum logic. Previously autovacuum was not necessarily triggered if space in the members slru got tight. The first problem was that the signalling was tied to values in the offsets slru, but members can advance much faster. Thats especially a problem if old sessions had been around that previously prevented the multixact horizon to increase. Secondly the skipping logic doesn't work if the database was restarted after autovacuum was triggered - that knowledge is not preserved across restart. This is especially a problem because it's a common panic-reaction to restart the database if it gets slow to anti-wraparound vacuums. Fix the first problem by separating the logic for members from offsets. Trigger autovacuum whenever a multixact crosses a segment boundary, as the current member offset increases in irregular values, so we can't use a simple modulo logic as for offsets. Add a stopgap for the second problem, by signalling autovacuum whenver ERRORing out because of boundaries. Discussion: 20150608163707.GD20772@alap3.anarazel.de Backpatch into 9.3, where it became more likely that multixacts wrap around.
2015-06-21 18:57:28 +02:00
if (!MultiXactIdPrecedes(result, MultiXactState->multiVacLimit))
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 safety's sake, we release MultiXactGenLock while sending
* signals, warnings, etc. This is not so much because we care about
* preserving concurrency in this situation, as to avoid any
* possibility of deadlock while doing get_database_name(). First,
* copy all the shared values we'll need in this path.
*/
MultiXactId multiWarnLimit = MultiXactState->multiWarnLimit;
MultiXactId multiStopLimit = MultiXactState->multiStopLimit;
MultiXactId multiWrapLimit = MultiXactState->multiWrapLimit;
Oid oldest_datoid = MultiXactState->oldestMultiXactDB;
LWLockRelease(MultiXactGenLock);
if (IsUnderPostmaster &&
!MultiXactIdPrecedes(result, multiStopLimit))
{
char *oldest_datname = get_database_name(oldest_datoid);
Improve multixact emergency autovacuum logic. Previously autovacuum was not necessarily triggered if space in the members slru got tight. The first problem was that the signalling was tied to values in the offsets slru, but members can advance much faster. Thats especially a problem if old sessions had been around that previously prevented the multixact horizon to increase. Secondly the skipping logic doesn't work if the database was restarted after autovacuum was triggered - that knowledge is not preserved across restart. This is especially a problem because it's a common panic-reaction to restart the database if it gets slow to anti-wraparound vacuums. Fix the first problem by separating the logic for members from offsets. Trigger autovacuum whenever a multixact crosses a segment boundary, as the current member offset increases in irregular values, so we can't use a simple modulo logic as for offsets. Add a stopgap for the second problem, by signalling autovacuum whenver ERRORing out because of boundaries. Discussion: 20150608163707.GD20772@alap3.anarazel.de Backpatch into 9.3, where it became more likely that multixacts wrap around.
2015-06-21 18:57:28 +02:00
/*
* Immediately kick autovacuum into action as we're already in
* ERROR territory.
*/
SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
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
/* complain even if that DB has disappeared */
if (oldest_datname)
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("database is not accepting commands that generate new MultiXactIds to avoid wraparound data loss in database \"%s\"",
oldest_datname),
errhint("Execute a database-wide VACUUM in that database.\n"
Extend near-wraparound hints to include replication slots Author: Feike Steenbergen Reviewed-by: Michael Paquier
2017-12-29 15:01:25 +01:00
"You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
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
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("database is not accepting commands that generate new MultiXactIds to avoid wraparound data loss in database with OID %u",
oldest_datoid),
errhint("Execute a database-wide VACUUM in that database.\n"
Extend near-wraparound hints to include replication slots Author: Feike Steenbergen Reviewed-by: Michael Paquier
2017-12-29 15:01:25 +01:00
"You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
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
}
Improve multixact emergency autovacuum logic. Previously autovacuum was not necessarily triggered if space in the members slru got tight. The first problem was that the signalling was tied to values in the offsets slru, but members can advance much faster. Thats especially a problem if old sessions had been around that previously prevented the multixact horizon to increase. Secondly the skipping logic doesn't work if the database was restarted after autovacuum was triggered - that knowledge is not preserved across restart. This is especially a problem because it's a common panic-reaction to restart the database if it gets slow to anti-wraparound vacuums. Fix the first problem by separating the logic for members from offsets. Trigger autovacuum whenever a multixact crosses a segment boundary, as the current member offset increases in irregular values, so we can't use a simple modulo logic as for offsets. Add a stopgap for the second problem, by signalling autovacuum whenver ERRORing out because of boundaries. Discussion: 20150608163707.GD20772@alap3.anarazel.de Backpatch into 9.3, where it became more likely that multixacts wrap around.
2015-06-21 18:57:28 +02:00
/*
* To avoid swamping the postmaster with signals, we issue the autovac
* request only once per 64K multis generated. This still gives
* plenty of chances before we get into real trouble.
*/
if (IsUnderPostmaster && (result % 65536) == 0)
SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
if (!MultiXactIdPrecedes(result, multiWarnLimit))
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
{
char *oldest_datname = get_database_name(oldest_datoid);
/* complain even if that DB has disappeared */
if (oldest_datname)
ereport(WARNING,
Message punctuation and pluralization fixes
2013-08-09 13:59:53 +02:00
(errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
"database \"%s\" must be vacuumed before %u more MultiXactIds are used",
multiWrapLimit - 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
oldest_datname,
multiWrapLimit - result),
errhint("Execute a database-wide VACUUM in that database.\n"
Extend near-wraparound hints to include replication slots Author: Feike Steenbergen Reviewed-by: Michael Paquier
2017-12-29 15:01:25 +01:00
"You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
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
ereport(WARNING,
Message punctuation and pluralization fixes
2013-08-09 13:59:53 +02:00
(errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
"database with OID %u must be vacuumed before %u more MultiXactIds are used",
multiWrapLimit - 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
oldest_datoid,
multiWrapLimit - result),
errhint("Execute a database-wide VACUUM in that database.\n"
Extend near-wraparound hints to include replication slots Author: Feike Steenbergen Reviewed-by: Michael Paquier
2017-12-29 15:01:25 +01:00
"You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
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
}
/* Re-acquire lock and start over */
LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
result = MultiXactState->nextMXact;
if (result < FirstMultiXactId)
result = FirstMultiXactId;
}
/* Make sure there is room for the MXID in the file. */
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
ExtendMultiXactOffset(result);
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
/*
Reorder code so that we don't have to hold a critical section while reserving SLRU space for a new MultiXact. The original coding would have treated out-of-disk-space as a PANIC condition, which is unnecessary.
2005-10-28 21:00:19 +02:00
* Reserve the members space, similarly to above. Also, be careful not to
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
* return zero as the starting offset for any multixact. See
* GetMultiXactIdMembers() for motivation.
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 in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
nextOffset = MultiXactState->nextOffset;
if (nextOffset == 0)
{
*offset = 1;
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
nmembers++; /* allocate member slot 0 too */
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
}
else
*offset = nextOffset;
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
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
/*----------
* Protect against overrun of the members space as well, with the
* following rules:
*
* If we're past offsetStopLimit, refuse to generate more multis.
* If we're close to offsetStopLimit, emit a warning.
*
* Arbitrarily, we start emitting warnings when we're 20 segments or less
* from offsetStopLimit.
*
* Note we haven't updated the shared state yet, so if we fail at this
* point, the multixact ID we grabbed can still be used by the next guy.
*
* Note that there is no point in forcing autovacuum runs here: the
* multixact freeze settings would have to be reduced for that to have any
* effect.
*----------
*/
#define OFFSET_WARN_SEGMENTS 20
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
if (MultiXactState->oldestOffsetKnown &&
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
MultiXactOffsetWouldWrap(MultiXactState->offsetStopLimit, nextOffset,
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
nmembers))
Improve multixact emergency autovacuum logic. Previously autovacuum was not necessarily triggered if space in the members slru got tight. The first problem was that the signalling was tied to values in the offsets slru, but members can advance much faster. Thats especially a problem if old sessions had been around that previously prevented the multixact horizon to increase. Secondly the skipping logic doesn't work if the database was restarted after autovacuum was triggered - that knowledge is not preserved across restart. This is especially a problem because it's a common panic-reaction to restart the database if it gets slow to anti-wraparound vacuums. Fix the first problem by separating the logic for members from offsets. Trigger autovacuum whenever a multixact crosses a segment boundary, as the current member offset increases in irregular values, so we can't use a simple modulo logic as for offsets. Add a stopgap for the second problem, by signalling autovacuum whenver ERRORing out because of boundaries. Discussion: 20150608163707.GD20772@alap3.anarazel.de Backpatch into 9.3, where it became more likely that multixacts wrap around.
2015-06-21 18:57:28 +02:00
{
/* see comment in the corresponding offsets wraparound case */
SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("multixact \"members\" limit exceeded"),
Code review for multixact bugfix Reword messages, rename a confusingly named function. Per Robert Haas.
2015-04-28 19:52:29 +02:00
errdetail_plural("This command would create a multixact with %u members, but the remaining space is only enough for %u member.",
"This command would create a multixact with %u members, but the remaining space is only enough for %u members.",
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
MultiXactState->offsetStopLimit - nextOffset - 1,
nmembers,
MultiXactState->offsetStopLimit - nextOffset - 1),
Code review for multixact bugfix Reword messages, rename a confusingly named function. Per Robert Haas.
2015-04-28 19:52:29 +02:00
errhint("Execute a database-wide VACUUM in database with OID %u with reduced vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age settings.",
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
MultiXactState->oldestMultiXactDB)));
Improve multixact emergency autovacuum logic. Previously autovacuum was not necessarily triggered if space in the members slru got tight. The first problem was that the signalling was tied to values in the offsets slru, but members can advance much faster. Thats especially a problem if old sessions had been around that previously prevented the multixact horizon to increase. Secondly the skipping logic doesn't work if the database was restarted after autovacuum was triggered - that knowledge is not preserved across restart. This is especially a problem because it's a common panic-reaction to restart the database if it gets slow to anti-wraparound vacuums. Fix the first problem by separating the logic for members from offsets. Trigger autovacuum whenever a multixact crosses a segment boundary, as the current member offset increases in irregular values, so we can't use a simple modulo logic as for offsets. Add a stopgap for the second problem, by signalling autovacuum whenver ERRORing out because of boundaries. Discussion: 20150608163707.GD20772@alap3.anarazel.de Backpatch into 9.3, where it became more likely that multixacts wrap around.
2015-06-21 18:57:28 +02:00
}
/*
* Check whether we should kick autovacuum into action, to prevent members
* wraparound. NB we use a much larger window to trigger autovacuum than
* just the warning limit. The warning is just a measure of last resort -
* this is in line with GetNewTransactionId's behaviour.
*/
if (!MultiXactState->oldestOffsetKnown ||
(MultiXactState->nextOffset - MultiXactState->oldestOffset
> MULTIXACT_MEMBER_SAFE_THRESHOLD))
{
/*
* To avoid swamping the postmaster with signals, we issue the autovac
* request only when crossing a segment boundary. With default
Fix typos in comments. CharSyam
2015-10-22 20:51:49 +02:00
* compilation settings that's roughly after 50k members. This still
Improve multixact emergency autovacuum logic. Previously autovacuum was not necessarily triggered if space in the members slru got tight. The first problem was that the signalling was tied to values in the offsets slru, but members can advance much faster. Thats especially a problem if old sessions had been around that previously prevented the multixact horizon to increase. Secondly the skipping logic doesn't work if the database was restarted after autovacuum was triggered - that knowledge is not preserved across restart. This is especially a problem because it's a common panic-reaction to restart the database if it gets slow to anti-wraparound vacuums. Fix the first problem by separating the logic for members from offsets. Trigger autovacuum whenever a multixact crosses a segment boundary, as the current member offset increases in irregular values, so we can't use a simple modulo logic as for offsets. Add a stopgap for the second problem, by signalling autovacuum whenver ERRORing out because of boundaries. Discussion: 20150608163707.GD20772@alap3.anarazel.de Backpatch into 9.3, where it became more likely that multixacts wrap around.
2015-06-21 18:57:28 +02:00
* gives plenty of chances before we get into real trouble.
*/
if ((MXOffsetToMemberPage(nextOffset) / SLRU_PAGES_PER_SEGMENT) !=
(MXOffsetToMemberPage(nextOffset + nmembers) / SLRU_PAGES_PER_SEGMENT))
SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
}
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
if (MultiXactState->oldestOffsetKnown &&
Improve multixact emergency autovacuum logic. Previously autovacuum was not necessarily triggered if space in the members slru got tight. The first problem was that the signalling was tied to values in the offsets slru, but members can advance much faster. Thats especially a problem if old sessions had been around that previously prevented the multixact horizon to increase. Secondly the skipping logic doesn't work if the database was restarted after autovacuum was triggered - that knowledge is not preserved across restart. This is especially a problem because it's a common panic-reaction to restart the database if it gets slow to anti-wraparound vacuums. Fix the first problem by separating the logic for members from offsets. Trigger autovacuum whenever a multixact crosses a segment boundary, as the current member offset increases in irregular values, so we can't use a simple modulo logic as for offsets. Add a stopgap for the second problem, by signalling autovacuum whenver ERRORing out because of boundaries. Discussion: 20150608163707.GD20772@alap3.anarazel.de Backpatch into 9.3, where it became more likely that multixacts wrap around.
2015-06-21 18:57:28 +02:00
MultiXactOffsetWouldWrap(MultiXactState->offsetStopLimit,
nextOffset,
nmembers + MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT * OFFSET_WARN_SEGMENTS))
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
ereport(WARNING,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
Message style improvements Message style, plurals, quoting, spelling, consistency with similar messages
2015-10-29 01:23:53 +01:00
errmsg_plural("database with OID %u must be vacuumed before %d more multixact member is used",
"database with OID %u must be vacuumed before %d more multixact members are used",
MultiXactState->offsetStopLimit - nextOffset + nmembers,
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
MultiXactState->oldestMultiXactDB,
MultiXactState->offsetStopLimit - nextOffset + nmembers),
Code review for multixact bugfix Reword messages, rename a confusingly named function. Per Robert Haas.
2015-04-28 19:52:29 +02:00
errhint("Execute a database-wide VACUUM in that database with reduced vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age settings.")));
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +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
ExtendMultiXactMember(nextOffset, nmembers);
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
Reorder code so that we don't have to hold a critical section while reserving SLRU space for a new MultiXact. The original coding would have treated out-of-disk-space as a PANIC condition, which is unnecessary.
2005-10-28 21:00:19 +02:00
/*
* Critical section from here until caller has written the data into the
* just-reserved SLRU space; we don't want to error out with a partly
* written MultiXact structure. (In particular, failing to write our
* start offset after advancing nextMXact would effectively corrupt the
* previous MultiXact.)
*/
START_CRIT_SECTION();
/*
* Advance counters. As in GetNewTransactionId(), this must not happen
* until after file extension has succeeded!
*
* We don't care about MultiXactId wraparound here; it will be handled by
* the next iteration. But note that nextMXact may be InvalidMultiXactId
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
* or the first value on a segment-beginning page after this routine
* exits, so anyone else looking at the variable must be prepared to deal
* with either case. Similarly, nextOffset may be zero, but we won't use
* that as the actual start offset of the next multixact.
Reorder code so that we don't have to hold a critical section while reserving SLRU space for a new MultiXact. The original coding would have treated out-of-disk-space as a PANIC condition, which is unnecessary.
2005-10-28 21:00:19 +02:00
*/
(MultiXactState->nextMXact)++;
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
MultiXactState->nextOffset += nmembers;
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
LWLockRelease(MultiXactGenLock);
debug_elog4(DEBUG2, "GetNew: returning %u offset %u", result, *offset);
return result;
}
/*
* GetMultiXactIdMembers
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
* Return the set of MultiXactMembers that make up a MultiXactId
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 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
* Return value is the number of members found, or -1 if there are none,
* and *members is set to a newly palloc'ed array of members. It's the
* caller's responsibility to free it when done with it.
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 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
* from_pgupgrade must be passed as true if and only if only the multixact
* corresponds to a value from a tuple that was locked in a 9.2-or-older
* installation and later pg_upgrade'd (that is, the infomask is
* HEAP_LOCKED_UPGRADED). In this case, we know for certain that no members
* can still be running, so we return -1 just like for an empty multixact
* without any further checking. It would be wrong to try to resolve such a
* multixact: either the multixact is within the current valid multixact
* range, in which case the returned result would be bogus, or outside that
* range, in which case an error would be raised.
*
* In all other cases, the passed multixact must be within the known valid
* range, that is, greater to or equal than oldestMultiXactId, and less than
* nextMXact. Otherwise, an error is raised.
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
*
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
* isLockOnly must be set to true if caller is certain that the given multi
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
* is used only to lock tuples; can be false without loss of correctness,
* but passing a true means we can return quickly without checking for
* old updates.
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 GetMultiXactIdMembers() a public function.
2005-08-20 03:29:27 +02:00
int
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
GetMultiXactIdMembers(MultiXactId multi, MultiXactMember **members,
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
bool from_pgupgrade, bool isLockOnly)
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
{
int pageno;
int prev_pageno;
int entryno;
int slotno;
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
MultiXactOffset *offptr;
MultiXactOffset offset;
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
int length;
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
int truelength;
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
int i;
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
MultiXactId oldestMXact;
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
MultiXactId nextMXact;
MultiXactId tmpMXact;
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
MultiXactOffset nextOffset;
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
MultiXactMember *ptr;
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
debug_elog3(DEBUG2, "GetMembers: asked for %u", 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) || from_pgupgrade)
Tidy up GetMultiXactIdMembers()'s behavior on error One of the error paths left *members uninitialized. That's not a live bug, because most callers don't look at *members when the function returns -1, but let's be tidy. One caller, in heap_lock_tuple(), does "if (members != NULL) pfree(members)", but AFAICS it never passes an invalid 'multi' value so it should not reach that error case. The callers are also a bit inconsistent in their expectations. heap_lock_tuple() pfrees the 'members' array if it's not-NULL, others pfree() it if "nmembers >= 0", and others if "nmembers > 0". That's not a live bug either, because the function should never return 0, but add an Assert for that to make it more clear. I left the callers alone for now. I also moved the line where we set *nmembers. It wasn't wrong before, but I like to do that right next to the 'return' statement, to make it clear that it's always set on return. Also remove one unreachable return statement after ereport(ERROR), for brevity and for consistency with the similar if-block right after it. Author: Greg Nancarrow with the additional changes by me Backpatch-through: 9.6, all supported versions
2021-06-17 13:50:42 +02:00
{
*members = NULL;
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
return -1;
Tidy up GetMultiXactIdMembers()'s behavior on error One of the error paths left *members uninitialized. That's not a live bug, because most callers don't look at *members when the function returns -1, but let's be tidy. One caller, in heap_lock_tuple(), does "if (members != NULL) pfree(members)", but AFAICS it never passes an invalid 'multi' value so it should not reach that error case. The callers are also a bit inconsistent in their expectations. heap_lock_tuple() pfrees the 'members' array if it's not-NULL, others pfree() it if "nmembers >= 0", and others if "nmembers > 0". That's not a live bug either, because the function should never return 0, but add an Assert for that to make it more clear. I left the callers alone for now. I also moved the line where we set *nmembers. It wasn't wrong before, but I like to do that right next to the 'return' statement, to make it clear that it's always set on return. Also remove one unreachable return statement after ereport(ERROR), for brevity and for consistency with the similar if-block right after it. Author: Greg Nancarrow with the additional changes by me Backpatch-through: 9.6, all supported versions
2021-06-17 13:50:42 +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
/* See if the MultiXactId is in the local cache */
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
length = mXactCacheGetById(multi, members);
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 (length >= 0)
{
debug_elog3(DEBUG2, "GetMembers: found %s in the cache",
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
mxid_to_string(multi, length, *members));
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
return length;
}
/* Set our OldestVisibleMXactId[] entry if we didn't already */
MultiXactIdSetOldestVisible();
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 we know the multi is used only for locking and not for updates, then
* we can skip checking if the value is older than our oldest visible
* multi. It cannot possibly still be running.
*/
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
if (isLockOnly &&
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
MultiXactIdPrecedes(multi, OldestVisibleMXactId[MyBackendId]))
{
debug_elog2(DEBUG2, "GetMembers: a locker-only multi is too old");
*members = NULL;
return -1;
}
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 check known limits on MultiXact before resorting to the SLRU area.
*
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
* An ID older than MultiXactState->oldestMultiXactId cannot possibly be
Have multixact be truncated by checkpoint, not vacuum Instead of truncating pg_multixact at vacuum time, do it only at checkpoint time. The reason for doing it this way is twofold: first, we want it to delete only segments that we're certain will not be required if there's a crash immediately after the removal; and second, we want to do it relatively often so that older files are not left behind if there's an untimely crash. Per my proposal in http://www.postgresql.org/message-id/20140626044519.GJ7340@eldon.alvh.no-ip.org we now execute the truncation in the checkpointer process rather than as part of vacuum. Vacuum is in only charge of maintaining in shared memory the value to which it's possible to truncate the files; that value is stored as part of checkpoints also, and so upon recovery we can reuse the same value to re-execute truncate and reset the oldest-value-still-safe-to-use to one known to remain after truncation. Per bug reported by Jeff Janes in the course of his tests involving bug #8673. While at it, update some comments that hadn't been updated since multixacts were changed. Backpatch to 9.3, where persistency of pg_multixact files was introduced by commit 0ac5ad5134f2.
2014-06-27 20:43:53 +02:00
* useful; it has already been removed, or will be removed shortly, by
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
* truncation. If one is passed, an error is raised.
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 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
* Also, an ID >= nextMXact shouldn't ever be seen here; if it is seen, it
* implies undetected ID wraparound has occurred. This raises a hard
* error.
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
*
* Shared lock is enough here since we aren't modifying any global state.
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
* Acquire it just long enough to grab the current counter values. We may
* need both nextMXact and nextOffset; see below.
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +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
LWLockAcquire(MultiXactGenLock, LW_SHARED);
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
oldestMXact = MultiXactState->oldestMultiXactId;
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
nextMXact = MultiXactState->nextMXact;
nextOffset = MultiXactState->nextOffset;
LWLockRelease(MultiXactGenLock);
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 (MultiXactIdPrecedes(multi, oldestMXact))
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
ereport(ERROR,
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
(errcode(ERRCODE_INTERNAL_ERROR),
errmsg("MultiXactId %u does no longer exist -- apparent wraparound",
multi)));
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 (!MultiXactIdPrecedes(multi, nextMXact))
ereport(ERROR,
(errcode(ERRCODE_INTERNAL_ERROR),
errmsg("MultiXactId %u has not been created yet -- apparent wraparound",
multi)));
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 in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
* Find out the offset at which we need to start reading MultiXactMembers
* and the number of members in the multixact. We determine the latter as
* the difference between this multixact's starting offset and the next
* one's. However, there are some corner cases to worry about:
*
* 1. This multixact may be the latest one created, in which case there is
* no next one to look at. In this case the nextOffset value we just
* saved is the correct endpoint.
*
* 2. The next multixact may still be in process of being filled in: that
* is, another process may have done GetNewMultiXactId but not yet written
* the offset entry for that ID. In that scenario, it is guaranteed that
* the offset entry for that multixact exists (because GetNewMultiXactId
* won't release MultiXactGenLock until it does) but contains zero
* (because we are careful to pre-zero offset pages). Because
* GetNewMultiXactId will never return zero as the starting offset for a
* multixact, when we read zero as the next multixact's offset, we know we
* have this case. We sleep for a bit and try again.
*
* 3. Because GetNewMultiXactId increments offset zero to offset one to
* handle case #2, there is an ambiguity near the point of offset
* wraparound. If we see next multixact's offset is one, is that our
* multixact's actual endpoint, or did it end at zero with a subsequent
* increment? We handle this using the knowledge that if the zero'th
* member slot wasn't filled, it'll contain zero, and zero isn't a valid
* transaction ID so it can't be a multixact member. Therefore, if we
* read a zero from the members array, just ignore it.
*
* This is all pretty messy, but the mess occurs only in infrequent corner
* cases, so it seems better than holding the MultiXactGenLock for a long
* time on every multixact creation.
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 in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
retry:
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockAcquire(MultiXactOffsetSLRULock, LW_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
pageno = MultiXactIdToOffsetPage(multi);
entryno = MultiXactIdToOffsetEntry(multi);
Support an optional asynchronous commit mode, in which we don't flush WAL before reporting a transaction committed. Data consistency is still guaranteed (unlike setting fsync = off), but a crash may lose the effects of the last few transactions. Patch by Simon, some editorialization by Tom.
2007-08-02 00:45:09 +02:00
slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, multi);
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
offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
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
offptr += entryno;
offset = *offptr;
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
Assert(offset != 0);
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 in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
* Use the same increment rule as GetNewMultiXactId(), that is, don't
* handle wraparound explicitly until needed.
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
*/
tmpMXact = multi + 1;
if (nextMXact == tmpMXact)
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
{
/* Corner case 1: there is no next multixact */
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
length = nextOffset - offset;
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +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
else
{
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
MultiXactOffset nextMXOffset;
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
/* handle wraparound if needed */
if (tmpMXact < FirstMultiXactId)
tmpMXact = FirstMultiXactId;
prev_pageno = pageno;
pageno = MultiXactIdToOffsetPage(tmpMXact);
entryno = MultiXactIdToOffsetEntry(tmpMXact);
if (pageno != prev_pageno)
Support an optional asynchronous commit mode, in which we don't flush WAL before reporting a transaction committed. Data consistency is still guaranteed (unlike setting fsync = off), but a crash may lose the effects of the last few transactions. Patch by Simon, some editorialization by Tom.
2007-08-02 00:45:09 +02:00
slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, tmpMXact);
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
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
offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
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
offptr += entryno;
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
nextMXOffset = *offptr;
if (nextMXOffset == 0)
{
/* Corner case 2: next multixact is still being filled in */
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactOffsetSLRULock);
Allow interrupting GetMultiXactIdMembers This function has a loop which can lead to uninterruptible process "stalls" (actually infinite loops) when some bugs are triggered. Avoid that unpleasant situation by adding a check for interrupts in a place that shouldn't degrade performance in the normal case. Backpatch to 9.3. Older branches have an identical loop here, but the aforementioned bugs are only a problem starting in 9.3 so there doesn't seem to be any point in backpatching any further.
2014-11-14 19:14:01 +01:00
CHECK_FOR_INTERRUPTS();
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
pg_usleep(1000L);
goto retry;
}
length = nextMXOffset - offset;
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
}
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactOffsetSLRULock);
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
ptr = (MultiXactMember *) palloc(length * sizeof(MultiXactMember));
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
/* Now get the members themselves. */
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockAcquire(MultiXactMemberSLRULock, LW_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
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
truelength = 0;
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
prev_pageno = -1;
for (i = 0; i < length; i++, offset++)
{
TransactionId *xactptr;
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
uint32 *flagsptr;
int flagsoff;
int bshift;
int memberoff;
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
pageno = MXOffsetToMemberPage(offset);
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
memberoff = MXOffsetToMemberOffset(offset);
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 (pageno != prev_pageno)
{
Support an optional asynchronous commit mode, in which we don't flush WAL before reporting a transaction committed. Data consistency is still guaranteed (unlike setting fsync = off), but a crash may lose the effects of the last few transactions. Patch by Simon, some editorialization by Tom.
2007-08-02 00:45:09 +02:00
slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, multi);
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
prev_pageno = pageno;
}
xactptr = (TransactionId *)
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
(MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
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 in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
if (!TransactionIdIsValid(*xactptr))
{
/* Corner case 3: we must be looking at unused slot zero */
Assert(offset == 0);
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
flagsoff = MXOffsetToFlagsOffset(offset);
bshift = MXOffsetToFlagsBitShift(offset);
flagsptr = (uint32 *) (MultiXactMemberCtl->shared->page_buffer[slotno] + flagsoff);
ptr[truelength].xid = *xactptr;
ptr[truelength].status = (*flagsptr >> bshift) & MXACT_MEMBER_XACT_BITMASK;
truelength++;
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
}
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactMemberSLRULock);
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
Tidy up GetMultiXactIdMembers()'s behavior on error One of the error paths left *members uninitialized. That's not a live bug, because most callers don't look at *members when the function returns -1, but let's be tidy. One caller, in heap_lock_tuple(), does "if (members != NULL) pfree(members)", but AFAICS it never passes an invalid 'multi' value so it should not reach that error case. The callers are also a bit inconsistent in their expectations. heap_lock_tuple() pfrees the 'members' array if it's not-NULL, others pfree() it if "nmembers >= 0", and others if "nmembers > 0". That's not a live bug either, because the function should never return 0, but add an Assert for that to make it more clear. I left the callers alone for now. I also moved the line where we set *nmembers. It wasn't wrong before, but I like to do that right next to the 'return' statement, to make it clear that it's always set on return. Also remove one unreachable return statement after ereport(ERROR), for brevity and for consistency with the similar if-block right after it. Author: Greg Nancarrow with the additional changes by me Backpatch-through: 9.6, all supported versions
2021-06-17 13:50:42 +02:00
/* A multixid with zero members should not happen */
Assert(truelength > 0);
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
/*
* Copy the result into the local cache.
*/
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
mXactCachePut(multi, truelength, ptr);
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
debug_elog3(DEBUG2, "GetMembers: no cache for %s",
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
mxid_to_string(multi, truelength, ptr));
Tidy up GetMultiXactIdMembers()'s behavior on error One of the error paths left *members uninitialized. That's not a live bug, because most callers don't look at *members when the function returns -1, but let's be tidy. One caller, in heap_lock_tuple(), does "if (members != NULL) pfree(members)", but AFAICS it never passes an invalid 'multi' value so it should not reach that error case. The callers are also a bit inconsistent in their expectations. heap_lock_tuple() pfrees the 'members' array if it's not-NULL, others pfree() it if "nmembers >= 0", and others if "nmembers > 0". That's not a live bug either, because the function should never return 0, but add an Assert for that to make it more clear. I left the callers alone for now. I also moved the line where we set *nmembers. It wasn't wrong before, but I like to do that right next to the 'return' statement, to make it clear that it's always set on return. Also remove one unreachable return statement after ereport(ERROR), for brevity and for consistency with the similar if-block right after it. Author: Greg Nancarrow with the additional changes by me Backpatch-through: 9.6, all supported versions
2021-06-17 13:50:42 +02:00
*members = ptr;
Fix race condition in multixact code: it's possible to try to read a multixact's starting offset before the offset has been stored into the SLRU file. A simple fix would be to hold the MultiXactGenLock until the offset has been stored, but that looks like a big concurrency hit. Instead rely on knowledge that unset offsets will be zero, and loop when we see a zero. This requires a little extra hacking to ensure that zero is never a valid value for the offset. Problem reported by Matteo Beccati, fix ideas from Martijn van Oosterhout, Alvaro Herrera, and Tom Lane.
2005-10-28 19:27:29 +02:00
return truelength;
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
/*
* mxactMemberComparator
* qsort comparison function for MultiXactMember
*
* We can't use wraparound comparison for XIDs because that does not respect
* the triangle inequality! Any old sort order will do.
*/
static int
mxactMemberComparator(const void *arg1, const void *arg2)
{
MultiXactMember member1 = *(const MultiXactMember *) arg1;
MultiXactMember member2 = *(const MultiXactMember *) arg2;
if (member1.xid > member2.xid)
return 1;
if (member1.xid < member2.xid)
return -1;
if (member1.status > member2.status)
return 1;
if (member1.status < member2.status)
return -1;
return 0;
}
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
/*
* mXactCacheGetBySet
* returns a MultiXactId from the cache based on the set of
* TransactionIds that compose it, or InvalidMultiXactId if
* none matches.
*
* This is helpful, for example, if two transactions want to lock a huge
* table. By using the cache, the second will use the same MultiXactId
* for the majority of tuples, thus keeping MultiXactId usage low (saving
* both I/O and wraparound issues).
*
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
* NB: the passed members array will be sorted in-place.
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 MultiXactId
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
mXactCacheGetBySet(int nmembers, MultiXactMember *members)
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
{
Rework MultiXactId cache code The original performs too poorly; in some scenarios it shows way too high while profiling. Try to make it a bit smarter to avoid excessive cosst. In particular, make it have a maximum size, and have entries be sorted in LRU order; once the max size is reached, evict the oldest entry to avoid it from growing too large. Per complaint from Andres Freund in connection with new tuple freezing code.
2013-12-13 21:16:25 +01:00
dlist_iter iter;
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
debug_elog3(DEBUG2, "CacheGet: looking for %s",
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
mxid_to_string(InvalidMultiXactId, nmembers, members));
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
/* sort the array so comparison is easy */
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
qsort(members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
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
Add doubly linked count list implementation We have various requirements when using a dlist_head to keep track of the number of items in the list. This, traditionally, has been done by maintaining a counter variable in the calling code. Here we tidy this up by adding "dclist", which is very similar to dlist but also keeps track of the number of items stored in the list. Callers may use the new dclist_count() function when they need to know how many items are stored. Obtaining the count is an O(1) operation. For simplicity reasons, dclist and dlist both use dlist_node as their node type and dlist_iter/dlist_mutable_iter as their iterator type. dclists have all of the same functionality as dlists except there is no function named dclist_delete(). To remove an item from a list dclist_delete_from() must be used. This requires knowing which dclist the given item is stored in. Additionally, here we also convert some dlists where additional code exists to keep track of the number of items stored and to make these use dclists instead. Author: David Rowley Reviewed-by: Bharath Rupireddy, Aleksander Alekseev Discussion: https://postgr.es/m/CAApHDvrtVxr+FXEX0VbViCFKDGxA3tWDgw9oFewNXCJMmwLjLg@mail.gmail.com
2022-11-02 02:06:05 +01:00
dclist_foreach(iter, &MXactCache)
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
{
Add doubly linked count list implementation We have various requirements when using a dlist_head to keep track of the number of items in the list. This, traditionally, has been done by maintaining a counter variable in the calling code. Here we tidy this up by adding "dclist", which is very similar to dlist but also keeps track of the number of items stored in the list. Callers may use the new dclist_count() function when they need to know how many items are stored. Obtaining the count is an O(1) operation. For simplicity reasons, dclist and dlist both use dlist_node as their node type and dlist_iter/dlist_mutable_iter as their iterator type. dclists have all of the same functionality as dlists except there is no function named dclist_delete(). To remove an item from a list dclist_delete_from() must be used. This requires knowing which dclist the given item is stored in. Additionally, here we also convert some dlists where additional code exists to keep track of the number of items stored and to make these use dclists instead. Author: David Rowley Reviewed-by: Bharath Rupireddy, Aleksander Alekseev Discussion: https://postgr.es/m/CAApHDvrtVxr+FXEX0VbViCFKDGxA3tWDgw9oFewNXCJMmwLjLg@mail.gmail.com
2022-11-02 02:06:05 +01:00
mXactCacheEnt *entry = dclist_container(mXactCacheEnt, node,
iter.cur);
Rework MultiXactId cache code The original performs too poorly; in some scenarios it shows way too high while profiling. Try to make it a bit smarter to avoid excessive cosst. In particular, make it have a maximum size, and have entries be sorted in LRU order; once the max size is reached, evict the oldest entry to avoid it from growing too large. Per complaint from Andres Freund in connection with new tuple freezing code.
2013-12-13 21:16:25 +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
if (entry->nmembers != nmembers)
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
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
/*
* We assume the cache entries are sorted, and that the unused bits in
* "status" are zeroed.
*/
if (memcmp(members, entry->members, nmembers * sizeof(MultiXactMember)) == 0)
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
{
debug_elog3(DEBUG2, "CacheGet: found %u", entry->multi);
Add doubly linked count list implementation We have various requirements when using a dlist_head to keep track of the number of items in the list. This, traditionally, has been done by maintaining a counter variable in the calling code. Here we tidy this up by adding "dclist", which is very similar to dlist but also keeps track of the number of items stored in the list. Callers may use the new dclist_count() function when they need to know how many items are stored. Obtaining the count is an O(1) operation. For simplicity reasons, dclist and dlist both use dlist_node as their node type and dlist_iter/dlist_mutable_iter as their iterator type. dclists have all of the same functionality as dlists except there is no function named dclist_delete(). To remove an item from a list dclist_delete_from() must be used. This requires knowing which dclist the given item is stored in. Additionally, here we also convert some dlists where additional code exists to keep track of the number of items stored and to make these use dclists instead. Author: David Rowley Reviewed-by: Bharath Rupireddy, Aleksander Alekseev Discussion: https://postgr.es/m/CAApHDvrtVxr+FXEX0VbViCFKDGxA3tWDgw9oFewNXCJMmwLjLg@mail.gmail.com
2022-11-02 02:06:05 +01:00
dclist_move_head(&MXactCache, iter.cur);
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
return entry->multi;
}
}
debug_elog2(DEBUG2, "CacheGet: not found :-(");
return InvalidMultiXactId;
}
/*
* mXactCacheGetById
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
* returns the composing MultiXactMember set from the cache for a
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
* given MultiXactId, if present.
*
* If successful, *xids is set to the address of a palloc'd copy of 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
* MultiXactMember set. Return value is number of members, or -1 on failure.
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 int
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
mXactCacheGetById(MultiXactId multi, MultiXactMember **members)
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
{
Rework MultiXactId cache code The original performs too poorly; in some scenarios it shows way too high while profiling. Try to make it a bit smarter to avoid excessive cosst. In particular, make it have a maximum size, and have entries be sorted in LRU order; once the max size is reached, evict the oldest entry to avoid it from growing too large. Per complaint from Andres Freund in connection with new tuple freezing code.
2013-12-13 21:16:25 +01:00
dlist_iter iter;
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
debug_elog3(DEBUG2, "CacheGet: looking for %u", multi);
Add doubly linked count list implementation We have various requirements when using a dlist_head to keep track of the number of items in the list. This, traditionally, has been done by maintaining a counter variable in the calling code. Here we tidy this up by adding "dclist", which is very similar to dlist but also keeps track of the number of items stored in the list. Callers may use the new dclist_count() function when they need to know how many items are stored. Obtaining the count is an O(1) operation. For simplicity reasons, dclist and dlist both use dlist_node as their node type and dlist_iter/dlist_mutable_iter as their iterator type. dclists have all of the same functionality as dlists except there is no function named dclist_delete(). To remove an item from a list dclist_delete_from() must be used. This requires knowing which dclist the given item is stored in. Additionally, here we also convert some dlists where additional code exists to keep track of the number of items stored and to make these use dclists instead. Author: David Rowley Reviewed-by: Bharath Rupireddy, Aleksander Alekseev Discussion: https://postgr.es/m/CAApHDvrtVxr+FXEX0VbViCFKDGxA3tWDgw9oFewNXCJMmwLjLg@mail.gmail.com
2022-11-02 02:06:05 +01:00
dclist_foreach(iter, &MXactCache)
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
{
Add doubly linked count list implementation We have various requirements when using a dlist_head to keep track of the number of items in the list. This, traditionally, has been done by maintaining a counter variable in the calling code. Here we tidy this up by adding "dclist", which is very similar to dlist but also keeps track of the number of items stored in the list. Callers may use the new dclist_count() function when they need to know how many items are stored. Obtaining the count is an O(1) operation. For simplicity reasons, dclist and dlist both use dlist_node as their node type and dlist_iter/dlist_mutable_iter as their iterator type. dclists have all of the same functionality as dlists except there is no function named dclist_delete(). To remove an item from a list dclist_delete_from() must be used. This requires knowing which dclist the given item is stored in. Additionally, here we also convert some dlists where additional code exists to keep track of the number of items stored and to make these use dclists instead. Author: David Rowley Reviewed-by: Bharath Rupireddy, Aleksander Alekseev Discussion: https://postgr.es/m/CAApHDvrtVxr+FXEX0VbViCFKDGxA3tWDgw9oFewNXCJMmwLjLg@mail.gmail.com
2022-11-02 02:06:05 +01:00
mXactCacheEnt *entry = dclist_container(mXactCacheEnt, node,
iter.cur);
Rework MultiXactId cache code The original performs too poorly; in some scenarios it shows way too high while profiling. Try to make it a bit smarter to avoid excessive cosst. In particular, make it have a maximum size, and have entries be sorted in LRU order; once the max size is reached, evict the oldest entry to avoid it from growing too large. Per complaint from Andres Freund in connection with new tuple freezing code.
2013-12-13 21:16:25 +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
if (entry->multi == 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
MultiXactMember *ptr;
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
Size size;
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
size = sizeof(MultiXactMember) * entry->nmembers;
ptr = (MultiXactMember *) palloc(size);
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
memcpy(ptr, entry->members, size);
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
debug_elog3(DEBUG2, "CacheGet: found %s",
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
mxid_to_string(multi,
entry->nmembers,
entry->members));
Rework MultiXactId cache code The original performs too poorly; in some scenarios it shows way too high while profiling. Try to make it a bit smarter to avoid excessive cosst. In particular, make it have a maximum size, and have entries be sorted in LRU order; once the max size is reached, evict the oldest entry to avoid it from growing too large. Per complaint from Andres Freund in connection with new tuple freezing code.
2013-12-13 21:16:25 +01:00
/*
Fix typo
2013-12-13 21:26:58 +01:00
* Note we modify the list while not using a modifiable iterator.
Rework MultiXactId cache code The original performs too poorly; in some scenarios it shows way too high while profiling. Try to make it a bit smarter to avoid excessive cosst. In particular, make it have a maximum size, and have entries be sorted in LRU order; once the max size is reached, evict the oldest entry to avoid it from growing too large. Per complaint from Andres Freund in connection with new tuple freezing code.
2013-12-13 21:16:25 +01:00
* This is acceptable only because we exit the iteration
* immediately afterwards.
*/
Add doubly linked count list implementation We have various requirements when using a dlist_head to keep track of the number of items in the list. This, traditionally, has been done by maintaining a counter variable in the calling code. Here we tidy this up by adding "dclist", which is very similar to dlist but also keeps track of the number of items stored in the list. Callers may use the new dclist_count() function when they need to know how many items are stored. Obtaining the count is an O(1) operation. For simplicity reasons, dclist and dlist both use dlist_node as their node type and dlist_iter/dlist_mutable_iter as their iterator type. dclists have all of the same functionality as dlists except there is no function named dclist_delete(). To remove an item from a list dclist_delete_from() must be used. This requires knowing which dclist the given item is stored in. Additionally, here we also convert some dlists where additional code exists to keep track of the number of items stored and to make these use dclists instead. Author: David Rowley Reviewed-by: Bharath Rupireddy, Aleksander Alekseev Discussion: https://postgr.es/m/CAApHDvrtVxr+FXEX0VbViCFKDGxA3tWDgw9oFewNXCJMmwLjLg@mail.gmail.com
2022-11-02 02:06:05 +01:00
dclist_move_head(&MXactCache, iter.cur);
Rework MultiXactId cache code The original performs too poorly; in some scenarios it shows way too high while profiling. Try to make it a bit smarter to avoid excessive cosst. In particular, make it have a maximum size, and have entries be sorted in LRU order; once the max size is reached, evict the oldest entry to avoid it from growing too large. Per complaint from Andres Freund in connection with new tuple freezing code.
2013-12-13 21:16:25 +01:00
Tidy up GetMultiXactIdMembers()'s behavior on error One of the error paths left *members uninitialized. That's not a live bug, because most callers don't look at *members when the function returns -1, but let's be tidy. One caller, in heap_lock_tuple(), does "if (members != NULL) pfree(members)", but AFAICS it never passes an invalid 'multi' value so it should not reach that error case. The callers are also a bit inconsistent in their expectations. heap_lock_tuple() pfrees the 'members' array if it's not-NULL, others pfree() it if "nmembers >= 0", and others if "nmembers > 0". That's not a live bug either, because the function should never return 0, but add an Assert for that to make it more clear. I left the callers alone for now. I also moved the line where we set *nmembers. It wasn't wrong before, but I like to do that right next to the 'return' statement, to make it clear that it's always set on return. Also remove one unreachable return statement after ereport(ERROR), for brevity and for consistency with the similar if-block right after it. Author: Greg Nancarrow with the additional changes by me Backpatch-through: 9.6, all supported versions
2021-06-17 13:50:42 +02:00
*members = ptr;
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 entry->nmembers;
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
}
}
debug_elog2(DEBUG2, "CacheGet: not found");
return -1;
}
/*
* mXactCachePut
* Add a new MultiXactId and its composing set into the local cache.
*/
static void
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
mXactCachePut(MultiXactId multi, int nmembers, MultiXactMember *members)
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
{
mXactCacheEnt *entry;
debug_elog3(DEBUG2, "CachePut: storing %s",
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
mxid_to_string(multi, nmembers, members));
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 (MXactContext == NULL)
{
/* The cache only lives as long as the current transaction */
debug_elog2(DEBUG2, "CachePut: initializing memory context");
MXactContext = AllocSetContextCreate(TopTransactionContext,
Add macros to make AllocSetContextCreate() calls simpler and safer. I found that half a dozen (nearly 5%) of our AllocSetContextCreate calls had typos in the context-sizing parameters. While none of these led to especially significant problems, they did create minor inefficiencies, and it's now clear that expecting people to copy-and-paste those calls accurately is not a great idea. Let's reduce the risk of future errors by introducing single macros that encapsulate the common use-cases. Three such macros are enough to cover all but two special-purpose contexts; those two calls can be left as-is, I think. While this patch doesn't in itself improve matters for third-party extensions, it doesn't break anything for them either, and they can gradually adopt the simplified notation over time. In passing, change TopMemoryContext to use the default allocation parameters. Formerly it could only be extended 8K at a time. That was probably reasonable when this code was written; but nowadays we create many more contexts than we did then, so that it's not unusual to have a couple hundred K in TopMemoryContext, even without considering various dubious code that sticks other things there. There seems no good reason not to let it use growing blocks like most other contexts. Back-patch to 9.6, mostly because that's still close enough to HEAD that it's easy to do so, and keeping the branches in sync can be expected to avoid some future back-patching pain. The bugs fixed by these changes don't seem to be significant enough to justify fixing them further back. Discussion: <21072.1472321324@sss.pgh.pa.us>
2016-08-27 23:50:38 +02:00
"MultiXact cache context",
ALLOCSET_SMALL_SIZES);
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
}
entry = (mXactCacheEnt *)
MemoryContextAlloc(MXactContext,
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
offsetof(mXactCacheEnt, members) +
nmembers * sizeof(MultiXactMember));
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
entry->multi = 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
entry->nmembers = nmembers;
memcpy(entry->members, members, nmembers * sizeof(MultiXactMember));
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
/* mXactCacheGetBySet assumes the entries are sorted, so sort 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
qsort(entry->members, nmembers, sizeof(MultiXactMember), mxactMemberComparator);
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
Add doubly linked count list implementation We have various requirements when using a dlist_head to keep track of the number of items in the list. This, traditionally, has been done by maintaining a counter variable in the calling code. Here we tidy this up by adding "dclist", which is very similar to dlist but also keeps track of the number of items stored in the list. Callers may use the new dclist_count() function when they need to know how many items are stored. Obtaining the count is an O(1) operation. For simplicity reasons, dclist and dlist both use dlist_node as their node type and dlist_iter/dlist_mutable_iter as their iterator type. dclists have all of the same functionality as dlists except there is no function named dclist_delete(). To remove an item from a list dclist_delete_from() must be used. This requires knowing which dclist the given item is stored in. Additionally, here we also convert some dlists where additional code exists to keep track of the number of items stored and to make these use dclists instead. Author: David Rowley Reviewed-by: Bharath Rupireddy, Aleksander Alekseev Discussion: https://postgr.es/m/CAApHDvrtVxr+FXEX0VbViCFKDGxA3tWDgw9oFewNXCJMmwLjLg@mail.gmail.com
2022-11-02 02:06:05 +01:00
dclist_push_head(&MXactCache, &entry->node);
if (dclist_count(&MXactCache) > MAX_CACHE_ENTRIES)
Rework MultiXactId cache code The original performs too poorly; in some scenarios it shows way too high while profiling. Try to make it a bit smarter to avoid excessive cosst. In particular, make it have a maximum size, and have entries be sorted in LRU order; once the max size is reached, evict the oldest entry to avoid it from growing too large. Per complaint from Andres Freund in connection with new tuple freezing code.
2013-12-13 21:16:25 +01:00
{
dlist_node *node;
Add doubly linked count list implementation We have various requirements when using a dlist_head to keep track of the number of items in the list. This, traditionally, has been done by maintaining a counter variable in the calling code. Here we tidy this up by adding "dclist", which is very similar to dlist but also keeps track of the number of items stored in the list. Callers may use the new dclist_count() function when they need to know how many items are stored. Obtaining the count is an O(1) operation. For simplicity reasons, dclist and dlist both use dlist_node as their node type and dlist_iter/dlist_mutable_iter as their iterator type. dclists have all of the same functionality as dlists except there is no function named dclist_delete(). To remove an item from a list dclist_delete_from() must be used. This requires knowing which dclist the given item is stored in. Additionally, here we also convert some dlists where additional code exists to keep track of the number of items stored and to make these use dclists instead. Author: David Rowley Reviewed-by: Bharath Rupireddy, Aleksander Alekseev Discussion: https://postgr.es/m/CAApHDvrtVxr+FXEX0VbViCFKDGxA3tWDgw9oFewNXCJMmwLjLg@mail.gmail.com
2022-11-02 02:06:05 +01:00
node = dclist_tail_node(&MXactCache);
dclist_delete_from(&MXactCache, node);
Rework MultiXactId cache code The original performs too poorly; in some scenarios it shows way too high while profiling. Try to make it a bit smarter to avoid excessive cosst. In particular, make it have a maximum size, and have entries be sorted in LRU order; once the max size is reached, evict the oldest entry to avoid it from growing too large. Per complaint from Andres Freund in connection with new tuple freezing code.
2013-12-13 21:16:25 +01:00
Add doubly linked count list implementation We have various requirements when using a dlist_head to keep track of the number of items in the list. This, traditionally, has been done by maintaining a counter variable in the calling code. Here we tidy this up by adding "dclist", which is very similar to dlist but also keeps track of the number of items stored in the list. Callers may use the new dclist_count() function when they need to know how many items are stored. Obtaining the count is an O(1) operation. For simplicity reasons, dclist and dlist both use dlist_node as their node type and dlist_iter/dlist_mutable_iter as their iterator type. dclists have all of the same functionality as dlists except there is no function named dclist_delete(). To remove an item from a list dclist_delete_from() must be used. This requires knowing which dclist the given item is stored in. Additionally, here we also convert some dlists where additional code exists to keep track of the number of items stored and to make these use dclists instead. Author: David Rowley Reviewed-by: Bharath Rupireddy, Aleksander Alekseev Discussion: https://postgr.es/m/CAApHDvrtVxr+FXEX0VbViCFKDGxA3tWDgw9oFewNXCJMmwLjLg@mail.gmail.com
2022-11-02 02:06:05 +01:00
entry = dclist_container(mXactCacheEnt, node, node);
Rework MultiXactId cache code The original performs too poorly; in some scenarios it shows way too high while profiling. Try to make it a bit smarter to avoid excessive cosst. In particular, make it have a maximum size, and have entries be sorted in LRU order; once the max size is reached, evict the oldest entry to avoid it from growing too large. Per complaint from Andres Freund in connection with new tuple freezing code.
2013-12-13 21:16:25 +01:00
debug_elog3(DEBUG2, "CachePut: pruning cached multi %u",
entry->multi);
pfree(entry);
}
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 char *
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
mxstatus_to_string(MultiXactStatus status)
{
switch (status)
{
case MultiXactStatusForKeyShare:
return "keysh";
case MultiXactStatusForShare:
return "sh";
case MultiXactStatusForNoKeyUpdate:
return "fornokeyupd";
case MultiXactStatusForUpdate:
return "forupd";
case MultiXactStatusNoKeyUpdate:
return "nokeyupd";
case MultiXactStatusUpdate:
return "upd";
default:
elog(ERROR, "unrecognized multixact status %d", status);
return "";
}
}
char *
mxid_to_string(MultiXactId multi, int nmembers, MultiXactMember *members)
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 char *str = NULL;
StringInfoData buf;
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
int i;
Standard pgindent run for 8.1.
2005-10-15 04:49:52 +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 (str != NULL)
pfree(str);
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
initStringInfo(&buf);
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
appendStringInfo(&buf, "%u %d[%u (%s)", multi, nmembers, members[0].xid,
mxstatus_to_string(members[0].status));
for (i = 1; i < nmembers; i++)
appendStringInfo(&buf, ", %u (%s)", members[i].xid,
mxstatus_to_string(members[i].status));
appendStringInfoChar(&buf, ']');
str = MemoryContextStrdup(TopMemoryContext, buf.data);
pfree(buf.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
return str;
}
/*
* AtEOXact_MultiXact
* Handle transaction end for MultiXact
*
* This is called at top transaction commit or abort (we don't care which).
*/
void
AtEOXact_MultiXact(void)
{
/*
* Reset our OldestMemberMXactId and OldestVisibleMXactId values, both of
* which should only be valid while within a transaction.
*
* We assume that storing a MultiXactId is atomic and so we need not take
* MultiXactGenLock to do this.
*/
OldestMemberMXactId[MyBackendId] = InvalidMultiXactId;
OldestVisibleMXactId[MyBackendId] = InvalidMultiXactId;
/*
* Discard the local MultiXactId cache. Since MXactContext was created as
* a child of TopTransactionContext, we needn't delete it explicitly.
*/
MXactContext = NULL;
Add doubly linked count list implementation We have various requirements when using a dlist_head to keep track of the number of items in the list. This, traditionally, has been done by maintaining a counter variable in the calling code. Here we tidy this up by adding "dclist", which is very similar to dlist but also keeps track of the number of items stored in the list. Callers may use the new dclist_count() function when they need to know how many items are stored. Obtaining the count is an O(1) operation. For simplicity reasons, dclist and dlist both use dlist_node as their node type and dlist_iter/dlist_mutable_iter as their iterator type. dclists have all of the same functionality as dlists except there is no function named dclist_delete(). To remove an item from a list dclist_delete_from() must be used. This requires knowing which dclist the given item is stored in. Additionally, here we also convert some dlists where additional code exists to keep track of the number of items stored and to make these use dclists instead. Author: David Rowley Reviewed-by: Bharath Rupireddy, Aleksander Alekseev Discussion: https://postgr.es/m/CAApHDvrtVxr+FXEX0VbViCFKDGxA3tWDgw9oFewNXCJMmwLjLg@mail.gmail.com
2022-11-02 02:06:05 +01:00
dclist_init(&MXactCache);
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 an old bug in multixact and two-phase commit. Prepared transactions can be part of multixacts, so allocate a slot for each prepared transaction in the "oldest member" array in multixact.c. On PREPARE TRANSACTION, transfer the oldest member value from the current backends slot to the prepared xact slot. Also save and recover the value from the 2pc state file. The symptom of the bug was that after a transaction prepared, a shared lock still held by the prepared transaction was sometimes ignored by other transactions. Fix back to 8.1, where both 2PC and multixact were introduced.
2009-11-23 10:58:36 +01:00
/*
* AtPrepare_MultiXact
Handle wraparound during truncation in multixact/members In pg_multixact/members, relying on modulo-2^32 arithmetic for wraparound handling doesn't work all that well. Because we don't explicitely track wraparound of the allocation counter for members, it is possible that the "live" area exceeds 2^31 entries; trying to remove SLRU segments that are "old" according to the original logic might lead to removal of segments still in use. To fix, have the truncation routine use a tailored SlruScanDirectory callback that keeps track of the live area in actual use; that way, when the live range exceeds 2^31 entries, the oldest segments still live will not get removed untimely. This new SlruScanDir callback needs to take care not to remove segments that are "in the future": if new SLRU segments appear while the truncation is ongoing, make sure we don't remove them. This requires examination of shared memory state to recheck for false positives, but testing suggests that this doesn't cause a problem. The original coding didn't suffer from this pitfall because segments created when truncation is running are never considered to be removable. Per Andres Freund's investigation of bug #8673 reported by Serge Negodyuck.
2014-01-02 22:16:54 +01:00
* Save multixact state at 2PC transaction prepare
Fix an old bug in multixact and two-phase commit. Prepared transactions can be part of multixacts, so allocate a slot for each prepared transaction in the "oldest member" array in multixact.c. On PREPARE TRANSACTION, transfer the oldest member value from the current backends slot to the prepared xact slot. Also save and recover the value from the 2pc state file. The symptom of the bug was that after a transaction prepared, a shared lock still held by the prepared transaction was sometimes ignored by other transactions. Fix back to 8.1, where both 2PC and multixact were introduced.
2009-11-23 10:58:36 +01:00
*
* In this phase, we only store our OldestMemberMXactId value in the two-phase
* state file.
*/
void
AtPrepare_MultiXact(void)
{
MultiXactId myOldestMember = OldestMemberMXactId[MyBackendId];
if (MultiXactIdIsValid(myOldestMember))
RegisterTwoPhaseRecord(TWOPHASE_RM_MULTIXACT_ID, 0,
&myOldestMember, sizeof(MultiXactId));
}
/*
* PostPrepare_MultiXact
* Clean up after successful PREPARE TRANSACTION
*/
void
PostPrepare_MultiXact(TransactionId xid)
{
MultiXactId myOldestMember;
/*
* Transfer our OldestMemberMXactId value to the slot reserved for the
* prepared transaction.
*/
myOldestMember = OldestMemberMXactId[MyBackendId];
if (MultiXactIdIsValid(myOldestMember))
{
Make release of 2PC identifier and locks consistent in COMMIT PREPARED When preparing a transaction in two-phase commit, a dummy PGPROC entry holding the GID used for the transaction is registered, which gets released once COMMIT PREPARED is run. Prior releasing its shared memory state, all the locks taken in the prepared transaction are released using a dedicated set of callbacks (pgstat and multixact having similar callbacks), which may cause the locks to be released before the GID is set free. Hence, there is a small window where lock conflicts could happen, for example: - Transaction A releases its locks, still holding its GID in shared memory. - Transaction B held a lock which conflicted with locks of transaction A. - Transaction B continues its processing, reusing the same GID as transaction A. - Transaction B fails because of a conflicting GID, already in use by transaction A. This commit changes the shared memory state release so as post-commit callbacks and predicate lock cleanup happen consistently with the shared memory state cleanup for the dummy PGPROC entry. The race window is small and 2PC had this issue from the start, so no backpatch is done. On top if that fixes discussed involved ABI breakages, which are not welcome in stable branches. Reported-by: Oleksii Kliukin, Ildar Musin Diagnosed-by: Oleksii Kliukin, Ildar Musin Author: Michael Paquier Reviewed-by: Masahiko Sawada, Oleksii Kliukin Discussion: https://postgr.es/m/BF9B38A4-2BFF-46E8-BA87-A2D00A8047A6@hintbits.com
2019-02-25 06:19:34 +01:00
BackendId dummyBackendId = TwoPhaseGetDummyBackendId(xid, false);
Fix an old bug in multixact and two-phase commit. Prepared transactions can be part of multixacts, so allocate a slot for each prepared transaction in the "oldest member" array in multixact.c. On PREPARE TRANSACTION, transfer the oldest member value from the current backends slot to the prepared xact slot. Also save and recover the value from the 2pc state file. The symptom of the bug was that after a transaction prepared, a shared lock still held by the prepared transaction was sometimes ignored by other transactions. Fix back to 8.1, where both 2PC and multixact were introduced.
2009-11-23 10:58:36 +01:00
/*
* Even though storing MultiXactId is atomic, acquire lock to make
* sure others see both changes, not just the reset of the slot of the
* current backend. Using a volatile pointer might suffice, but this
* isn't a hot spot.
*/
LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
OldestMemberMXactId[dummyBackendId] = myOldestMember;
OldestMemberMXactId[MyBackendId] = InvalidMultiXactId;
LWLockRelease(MultiXactGenLock);
}
/*
* We don't need to transfer OldestVisibleMXactId value, because the
* transaction is not going to be looking at any more multixacts once it's
* prepared.
*
* We assume that storing a MultiXactId is atomic and so we need not take
* MultiXactGenLock to do this.
*/
OldestVisibleMXactId[MyBackendId] = InvalidMultiXactId;
/*
Fix typo in multixact.c AtEOXact_MultiXact() was referenced in two places with an incorrect routine name. Author: Hou Zhijie Discussion: https://postgr.es/m/1b41e9311e8f474cb5a360292f0b3cb1@G08CNEXMBPEKD05.g08.fujitsu.local
2020-10-08 07:06:12 +02:00
* Discard the local MultiXactId cache like in AtEOXact_MultiXact.
Fix an old bug in multixact and two-phase commit. Prepared transactions can be part of multixacts, so allocate a slot for each prepared transaction in the "oldest member" array in multixact.c. On PREPARE TRANSACTION, transfer the oldest member value from the current backends slot to the prepared xact slot. Also save and recover the value from the 2pc state file. The symptom of the bug was that after a transaction prepared, a shared lock still held by the prepared transaction was sometimes ignored by other transactions. Fix back to 8.1, where both 2PC and multixact were introduced.
2009-11-23 10:58:36 +01:00
*/
MXactContext = NULL;
Add doubly linked count list implementation We have various requirements when using a dlist_head to keep track of the number of items in the list. This, traditionally, has been done by maintaining a counter variable in the calling code. Here we tidy this up by adding "dclist", which is very similar to dlist but also keeps track of the number of items stored in the list. Callers may use the new dclist_count() function when they need to know how many items are stored. Obtaining the count is an O(1) operation. For simplicity reasons, dclist and dlist both use dlist_node as their node type and dlist_iter/dlist_mutable_iter as their iterator type. dclists have all of the same functionality as dlists except there is no function named dclist_delete(). To remove an item from a list dclist_delete_from() must be used. This requires knowing which dclist the given item is stored in. Additionally, here we also convert some dlists where additional code exists to keep track of the number of items stored and to make these use dclists instead. Author: David Rowley Reviewed-by: Bharath Rupireddy, Aleksander Alekseev Discussion: https://postgr.es/m/CAApHDvrtVxr+FXEX0VbViCFKDGxA3tWDgw9oFewNXCJMmwLjLg@mail.gmail.com
2022-11-02 02:06:05 +01:00
dclist_init(&MXactCache);
Fix an old bug in multixact and two-phase commit. Prepared transactions can be part of multixacts, so allocate a slot for each prepared transaction in the "oldest member" array in multixact.c. On PREPARE TRANSACTION, transfer the oldest member value from the current backends slot to the prepared xact slot. Also save and recover the value from the 2pc state file. The symptom of the bug was that after a transaction prepared, a shared lock still held by the prepared transaction was sometimes ignored by other transactions. Fix back to 8.1, where both 2PC and multixact were introduced.
2009-11-23 10:58:36 +01:00
}
/*
* multixact_twophase_recover
* Recover the state of a prepared transaction at startup
*/
void
multixact_twophase_recover(TransactionId xid, uint16 info,
void *recdata, uint32 len)
{
Make release of 2PC identifier and locks consistent in COMMIT PREPARED When preparing a transaction in two-phase commit, a dummy PGPROC entry holding the GID used for the transaction is registered, which gets released once COMMIT PREPARED is run. Prior releasing its shared memory state, all the locks taken in the prepared transaction are released using a dedicated set of callbacks (pgstat and multixact having similar callbacks), which may cause the locks to be released before the GID is set free. Hence, there is a small window where lock conflicts could happen, for example: - Transaction A releases its locks, still holding its GID in shared memory. - Transaction B held a lock which conflicted with locks of transaction A. - Transaction B continues its processing, reusing the same GID as transaction A. - Transaction B fails because of a conflicting GID, already in use by transaction A. This commit changes the shared memory state release so as post-commit callbacks and predicate lock cleanup happen consistently with the shared memory state cleanup for the dummy PGPROC entry. The race window is small and 2PC had this issue from the start, so no backpatch is done. On top if that fixes discussed involved ABI breakages, which are not welcome in stable branches. Reported-by: Oleksii Kliukin, Ildar Musin Diagnosed-by: Oleksii Kliukin, Ildar Musin Author: Michael Paquier Reviewed-by: Masahiko Sawada, Oleksii Kliukin Discussion: https://postgr.es/m/BF9B38A4-2BFF-46E8-BA87-A2D00A8047A6@hintbits.com
2019-02-25 06:19:34 +01:00
BackendId dummyBackendId = TwoPhaseGetDummyBackendId(xid, false);
Fix an old bug in multixact and two-phase commit. Prepared transactions can be part of multixacts, so allocate a slot for each prepared transaction in the "oldest member" array in multixact.c. On PREPARE TRANSACTION, transfer the oldest member value from the current backends slot to the prepared xact slot. Also save and recover the value from the 2pc state file. The symptom of the bug was that after a transaction prepared, a shared lock still held by the prepared transaction was sometimes ignored by other transactions. Fix back to 8.1, where both 2PC and multixact were introduced.
2009-11-23 10:58:36 +01:00
MultiXactId oldestMember;
/*
* Get the oldest member XID from the state file record, and set it in the
* OldestMemberMXactId slot reserved for this prepared transaction.
*/
Assert(len == sizeof(MultiXactId));
oldestMember = *((MultiXactId *) recdata);
OldestMemberMXactId[dummyBackendId] = oldestMember;
}
/*
* multixact_twophase_postcommit
Fix typo in multixact.c AtEOXact_MultiXact() was referenced in two places with an incorrect routine name. Author: Hou Zhijie Discussion: https://postgr.es/m/1b41e9311e8f474cb5a360292f0b3cb1@G08CNEXMBPEKD05.g08.fujitsu.local
2020-10-08 07:06:12 +02:00
* Similar to AtEOXact_MultiXact but for COMMIT PREPARED
Fix an old bug in multixact and two-phase commit. Prepared transactions can be part of multixacts, so allocate a slot for each prepared transaction in the "oldest member" array in multixact.c. On PREPARE TRANSACTION, transfer the oldest member value from the current backends slot to the prepared xact slot. Also save and recover the value from the 2pc state file. The symptom of the bug was that after a transaction prepared, a shared lock still held by the prepared transaction was sometimes ignored by other transactions. Fix back to 8.1, where both 2PC and multixact were introduced.
2009-11-23 10:58:36 +01:00
*/
void
multixact_twophase_postcommit(TransactionId xid, uint16 info,
void *recdata, uint32 len)
{
Make release of 2PC identifier and locks consistent in COMMIT PREPARED When preparing a transaction in two-phase commit, a dummy PGPROC entry holding the GID used for the transaction is registered, which gets released once COMMIT PREPARED is run. Prior releasing its shared memory state, all the locks taken in the prepared transaction are released using a dedicated set of callbacks (pgstat and multixact having similar callbacks), which may cause the locks to be released before the GID is set free. Hence, there is a small window where lock conflicts could happen, for example: - Transaction A releases its locks, still holding its GID in shared memory. - Transaction B held a lock which conflicted with locks of transaction A. - Transaction B continues its processing, reusing the same GID as transaction A. - Transaction B fails because of a conflicting GID, already in use by transaction A. This commit changes the shared memory state release so as post-commit callbacks and predicate lock cleanup happen consistently with the shared memory state cleanup for the dummy PGPROC entry. The race window is small and 2PC had this issue from the start, so no backpatch is done. On top if that fixes discussed involved ABI breakages, which are not welcome in stable branches. Reported-by: Oleksii Kliukin, Ildar Musin Diagnosed-by: Oleksii Kliukin, Ildar Musin Author: Michael Paquier Reviewed-by: Masahiko Sawada, Oleksii Kliukin Discussion: https://postgr.es/m/BF9B38A4-2BFF-46E8-BA87-A2D00A8047A6@hintbits.com
2019-02-25 06:19:34 +01:00
BackendId dummyBackendId = TwoPhaseGetDummyBackendId(xid, true);
Fix an old bug in multixact and two-phase commit. Prepared transactions can be part of multixacts, so allocate a slot for each prepared transaction in the "oldest member" array in multixact.c. On PREPARE TRANSACTION, transfer the oldest member value from the current backends slot to the prepared xact slot. Also save and recover the value from the 2pc state file. The symptom of the bug was that after a transaction prepared, a shared lock still held by the prepared transaction was sometimes ignored by other transactions. Fix back to 8.1, where both 2PC and multixact were introduced.
2009-11-23 10:58:36 +01:00
Assert(len == sizeof(MultiXactId));
OldestMemberMXactId[dummyBackendId] = InvalidMultiXactId;
}
/*
* multixact_twophase_postabort
* This is actually just the same as the COMMIT case.
*/
void
multixact_twophase_postabort(TransactionId xid, uint16 info,
void *recdata, uint32 len)
{
multixact_twophase_postcommit(xid, info, recdata, len);
}
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
/*
* Initialization of shared memory for MultiXact. We use two SLRU areas,
* thus double memory. Also, reserve space for the shared MultiXactState
* struct and the per-backend MultiXactId arrays (two of those, too).
*/
Convert the arithmetic for shared memory size calculation from 'int' to 'Size' (that is, size_t), and install overflow detection checks in it. This allows us to remove the former arbitrary restrictions on NBuffers etc. It won't make any difference in a 32-bit machine, but in a 64-bit machine you could theoretically have terabytes of shared buffers. (How efficiently we could manage 'em remains to be seen.) Similarly, num_temp_buffers, work_mem, and maintenance_work_mem can be set above 2Gb on a 64-bit machine. Original patch from Koichi Suzuki, additional work by moi.
2005-08-21 01:26:37 +02:00
Size
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
MultiXactShmemSize(void)
{
Convert the arithmetic for shared memory size calculation from 'int' to 'Size' (that is, size_t), and install overflow detection checks in it. This allows us to remove the former arbitrary restrictions on NBuffers etc. It won't make any difference in a 32-bit machine, but in a 64-bit machine you could theoretically have terabytes of shared buffers. (How efficiently we could manage 'em remains to be seen.) Similarly, num_temp_buffers, work_mem, and maintenance_work_mem can be set above 2Gb on a 64-bit machine. Original patch from Koichi Suzuki, additional work by moi.
2005-08-21 01:26:37 +02:00
Size size;
Use FLEXIBLE_ARRAY_MEMBER in some more places. Fix a batch of structs that are only visible within individual .c files. Michael Paquier
2015-02-20 23:32:01 +01:00
/* We need 2*MaxOldestSlot + 1 perBackendXactIds[] entries */
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
#define SHARED_MULTIXACT_STATE_SIZE \
Use FLEXIBLE_ARRAY_MEMBER in some more places. Fix a batch of structs that are only visible within individual .c files. Michael Paquier
2015-02-20 23:32:01 +01:00
add_size(offsetof(MultiXactStateData, perBackendXactIds) + sizeof(MultiXactId), \
Revert the addition of GetMaxBackends() and related stuff. This reverts commits 0147fc7, 4567596, aa64f23, and 5ecd018. There is no longer agreement that introducing this function was the right way to address the problem. The consensus now seems to favor trying to make a correct value for MaxBackends available to mdules executing their _PG_init() functions. Nathan Bossart Discussion: http://postgr.es/m/20220323045229.i23skfscdbvrsuxa@jrouhaud
2022-04-12 20:45:23 +02:00
mul_size(sizeof(MultiXactId) * 2, MaxOldestSlot))
Convert the arithmetic for shared memory size calculation from 'int' to 'Size' (that is, size_t), and install overflow detection checks in it. This allows us to remove the former arbitrary restrictions on NBuffers etc. It won't make any difference in a 32-bit machine, but in a 64-bit machine you could theoretically have terabytes of shared buffers. (How efficiently we could manage 'em remains to be seen.) Similarly, num_temp_buffers, work_mem, and maintenance_work_mem can be set above 2Gb on a 64-bit machine. Original patch from Koichi Suzuki, additional work by moi.
2005-08-21 01:26:37 +02:00
size = SHARED_MULTIXACT_STATE_SIZE;
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
size = add_size(size, SimpleLruShmemSize(NUM_MULTIXACTOFFSET_BUFFERS, 0));
size = add_size(size, SimpleLruShmemSize(NUM_MULTIXACTMEMBER_BUFFERS, 0));
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
Convert the arithmetic for shared memory size calculation from 'int' to 'Size' (that is, size_t), and install overflow detection checks in it. This allows us to remove the former arbitrary restrictions on NBuffers etc. It won't make any difference in a 32-bit machine, but in a 64-bit machine you could theoretically have terabytes of shared buffers. (How efficiently we could manage 'em remains to be seen.) Similarly, num_temp_buffers, work_mem, and maintenance_work_mem can be set above 2Gb on a 64-bit machine. Original patch from Koichi Suzuki, additional work by moi.
2005-08-21 01:26:37 +02:00
return size;
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
}
void
MultiXactShmemInit(void)
{
bool found;
debug_elog2(DEBUG2, "Shared Memory Init for MultiXact");
MultiXactOffsetCtl->PagePrecedes = MultiXactOffsetPagePrecedes;
MultiXactMemberCtl->PagePrecedes = MultiXactMemberPagePrecedes;
Get rid of slru.c's hardwired insistence on a fixed number of slots per SLRU area. The number of slots is still a compile-time constant (someday we might want to change that), but at least it's a different constant for each SLRU area. Increase number of subtrans buffers to 32 based on experimentation with a heavily subtrans-bashing test case, and increase number of multixact member buffers to 16, since it's obviously silly for it not to be at least twice the number of multixact offset buffers.
2005-12-07 00:08:34 +01:00
SimpleLruInit(MultiXactOffsetCtl,
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
"MultiXactOffset", NUM_MULTIXACTOFFSET_BUFFERS, 0,
MultiXactOffsetSLRULock, "pg_multixact/offsets",
Defer flushing of SLRU files. Previously, we called fsync() after writing out individual pg_xact, pg_multixact and pg_commit_ts pages due to cache pressure, leading to regular I/O stalls in user backends and recovery. Collapse requests for the same file into a single system call as part of the next checkpoint, as we already did for relation files, using the infrastructure developed by commit 3eb77eba. This can cause a significant improvement to recovery performance, especially when it's otherwise CPU-bound. Hoist ProcessSyncRequests() up into CheckPointGuts() to make it clearer that it applies to all the SLRU mini-buffer-pools as well as the main buffer pool. Rearrange things so that data collected in CheckpointStats includes SLRU activity. Also remove the Shutdown{CLOG,CommitTS,SUBTRANS,MultiXact}() functions, because they were redundant after the shutdown checkpoint that immediately precedes them. (I'm not sure if they were ever needed, but they aren't now.) Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us> (parts) Tested-by: Jakub Wartak <Jakub.Wartak@tomtom.com> Discussion: https://postgr.es/m/CA+hUKGLJ=84YT+NvhkEEDAuUtVHMfQ9i-N7k_o50JmQ6Rpj_OQ@mail.gmail.com
2020-09-25 08:49:43 +02:00
LWTRANCHE_MULTIXACTOFFSET_BUFFER,
SYNC_HANDLER_MULTIXACT_OFFSET);
Prevent excess SimpleLruTruncate() deletion. Every core SLRU wraps around. With the exception of pg_notify, the wrap point can fall in the middle of a page. Account for this in the PagePrecedes callback specification and in SimpleLruTruncate()'s use of said callback. Update each callback implementation to fit the new specification. This changes SerialPagePrecedesLogically() from the style of asyncQueuePagePrecedes() to the style of CLOGPagePrecedes(). (Whereas pg_clog and pg_serial share a key space, pg_serial is nothing like pg_notify.) The bug fixed here has the same symptoms and user followup steps as 592a589a04bd456410b853d86bd05faa9432cbbb. Back-patch to 9.5 (all supported versions). Reviewed by Andrey Borodin and (in earlier versions) by Tom Lane. Discussion: https://postgr.es/m/20190202083822.GC32531@gust.leadboat.com
2021-01-16 21:21:35 +01:00
SlruPagePrecedesUnitTests(MultiXactOffsetCtl, MULTIXACT_OFFSETS_PER_PAGE);
Get rid of slru.c's hardwired insistence on a fixed number of slots per SLRU area. The number of slots is still a compile-time constant (someday we might want to change that), but at least it's a different constant for each SLRU area. Increase number of subtrans buffers to 32 based on experimentation with a heavily subtrans-bashing test case, and increase number of multixact member buffers to 16, since it's obviously silly for it not to be at least twice the number of multixact offset buffers.
2005-12-07 00:08:34 +01:00
SimpleLruInit(MultiXactMemberCtl,
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
"MultiXactMember", NUM_MULTIXACTMEMBER_BUFFERS, 0,
MultiXactMemberSLRULock, "pg_multixact/members",
Defer flushing of SLRU files. Previously, we called fsync() after writing out individual pg_xact, pg_multixact and pg_commit_ts pages due to cache pressure, leading to regular I/O stalls in user backends and recovery. Collapse requests for the same file into a single system call as part of the next checkpoint, as we already did for relation files, using the infrastructure developed by commit 3eb77eba. This can cause a significant improvement to recovery performance, especially when it's otherwise CPU-bound. Hoist ProcessSyncRequests() up into CheckPointGuts() to make it clearer that it applies to all the SLRU mini-buffer-pools as well as the main buffer pool. Rearrange things so that data collected in CheckpointStats includes SLRU activity. Also remove the Shutdown{CLOG,CommitTS,SUBTRANS,MultiXact}() functions, because they were redundant after the shutdown checkpoint that immediately precedes them. (I'm not sure if they were ever needed, but they aren't now.) Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us> (parts) Tested-by: Jakub Wartak <Jakub.Wartak@tomtom.com> Discussion: https://postgr.es/m/CA+hUKGLJ=84YT+NvhkEEDAuUtVHMfQ9i-N7k_o50JmQ6Rpj_OQ@mail.gmail.com
2020-09-25 08:49:43 +02:00
LWTRANCHE_MULTIXACTMEMBER_BUFFER,
SYNC_HANDLER_MULTIXACT_MEMBER);
Prevent excess SimpleLruTruncate() deletion. Every core SLRU wraps around. With the exception of pg_notify, the wrap point can fall in the middle of a page. Account for this in the PagePrecedes callback specification and in SimpleLruTruncate()'s use of said callback. Update each callback implementation to fit the new specification. This changes SerialPagePrecedesLogically() from the style of asyncQueuePagePrecedes() to the style of CLOGPagePrecedes(). (Whereas pg_clog and pg_serial share a key space, pg_serial is nothing like pg_notify.) The bug fixed here has the same symptoms and user followup steps as 592a589a04bd456410b853d86bd05faa9432cbbb. Back-patch to 9.5 (all supported versions). Reviewed by Andrey Borodin and (in earlier versions) by Tom Lane. Discussion: https://postgr.es/m/20190202083822.GC32531@gust.leadboat.com
2021-01-16 21:21:35 +01:00
/* doesn't call SimpleLruTruncate() or meet criteria for unit tests */
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
/* Initialize our shared state struct */
MultiXactState = ShmemInitStruct("Shared MultiXact State",
SHARED_MULTIXACT_STATE_SIZE,
&found);
if (!IsUnderPostmaster)
{
Assert(!found);
/* Make sure we zero out the per-backend state */
MemSet(MultiXactState, 0, SHARED_MULTIXACT_STATE_SIZE);
}
else
Assert(found);
/*
* Set up array pointers. Note that perBackendXactIds[0] is wasted space
Fix an old bug in multixact and two-phase commit. Prepared transactions can be part of multixacts, so allocate a slot for each prepared transaction in the "oldest member" array in multixact.c. On PREPARE TRANSACTION, transfer the oldest member value from the current backends slot to the prepared xact slot. Also save and recover the value from the 2pc state file. The symptom of the bug was that after a transaction prepared, a shared lock still held by the prepared transaction was sometimes ignored by other transactions. Fix back to 8.1, where both 2PC and multixact were introduced.
2009-11-23 10:58:36 +01:00
* since we only use indexes 1..MaxOldestSlot in each array.
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
*/
OldestMemberMXactId = MultiXactState->perBackendXactIds;
Revert the addition of GetMaxBackends() and related stuff. This reverts commits 0147fc7, 4567596, aa64f23, and 5ecd018. There is no longer agreement that introducing this function was the right way to address the problem. The consensus now seems to favor trying to make a correct value for MaxBackends available to mdules executing their _PG_init() functions. Nathan Bossart Discussion: http://postgr.es/m/20220323045229.i23skfscdbvrsuxa@jrouhaud
2022-04-12 20:45:23 +02:00
OldestVisibleMXactId = OldestMemberMXactId + MaxOldestSlot;
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
}
/*
* This func must be called ONCE on system install. It creates the initial
* MultiXact segments. (The MultiXacts directories are assumed to have been
* created by initdb, and MultiXactShmemInit must have been called already.)
*/
void
BootStrapMultiXact(void)
{
int slotno;
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockAcquire(MultiXactOffsetSLRULock, LW_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
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
/* Create and zero the first page of the offsets log */
slotno = ZeroMultiXactOffsetPage(0, false);
/* Make sure it's written out */
Avoid unnecessary public struct declaration in slru.h Instead, declare a public wrapper of the sole function using it for external callers, so that they don't have to always pass a NULL argument. Author: Kevin Grittner
2010-12-30 16:09:17 +01:00
SimpleLruWritePage(MultiXactOffsetCtl, slotno);
Clean up representation of SLRU page state. This is the cleaner fix for the SLRU race condition that I posted a few days ago, but we decided not to use in 8.1 and older branches.
2005-11-05 22:19:47 +01:00
Assert(!MultiXactOffsetCtl->shared->page_dirty[slotno]);
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
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactOffsetSLRULock);
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
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockAcquire(MultiXactMemberSLRULock, LW_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
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
/* Create and zero the first page of the members log */
slotno = ZeroMultiXactMemberPage(0, false);
/* Make sure it's written out */
Avoid unnecessary public struct declaration in slru.h Instead, declare a public wrapper of the sole function using it for external callers, so that they don't have to always pass a NULL argument. Author: Kevin Grittner
2010-12-30 16:09:17 +01:00
SimpleLruWritePage(MultiXactMemberCtl, slotno);
Clean up representation of SLRU page state. This is the cleaner fix for the SLRU race condition that I posted a few days ago, but we decided not to use in 8.1 and older branches.
2005-11-05 22:19:47 +01:00
Assert(!MultiXactMemberCtl->shared->page_dirty[slotno]);
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
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactMemberSLRULock);
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
}
/*
* Initialize (or reinitialize) a page of MultiXactOffset to zeroes.
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
* If writeXlog is true, also emit an XLOG record saying we did this.
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
*
* The page is not actually written, just set up in shared memory.
* The slot number of the new page is returned.
*
* Control lock must be held at entry, and will be held at exit.
*/
static int
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
ZeroMultiXactOffsetPage(int pageno, bool writeXlog)
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
{
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
int slotno;
slotno = SimpleLruZeroPage(MultiXactOffsetCtl, pageno);
if (writeXlog)
WriteMZeroPageXlogRec(pageno, XLOG_MULTIXACT_ZERO_OFF_PAGE);
return slotno;
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
}
/*
* Ditto, for MultiXactMember
*/
static int
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
ZeroMultiXactMemberPage(int pageno, bool writeXlog)
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
{
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
int slotno;
slotno = SimpleLruZeroPage(MultiXactMemberCtl, pageno);
if (writeXlog)
WriteMZeroPageXlogRec(pageno, XLOG_MULTIXACT_ZERO_MEM_PAGE);
return slotno;
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 pg_upgrade failure from servers older than 9.3 When upgrading from servers of versions 9.2 and older, and MultiXactIds have been used in the old server beyond the first page (that is, 2048 multis or more in the default 8kB-page build), pg_upgrade would set the next multixact offset to use beyond what has been allocated in the new cluster. This would cause a failure the first time the new cluster needs to use this value, because the pg_multixact/offsets/ file wouldn't exist or wouldn't be large enough. To fix, ensure that the transient server instances launched by pg_upgrade extend the file as necessary. Per report from Jesse Denardo in CANiVXAj4c88YqipsyFQPboqMudnjcNTdB3pqe8ReXqAFQ=HXyA@mail.gmail.com
2013-08-19 18:33:07 +02:00
/*
* MaybeExtendOffsetSlru
* Extend the offsets SLRU area, if necessary
*
Fix typo in comment. Michael Paquier Discussion: http://postgr.es/m/20180205071404.GB17337@paquier.xyz
2018-03-22 18:36:14 +01:00
* After a binary upgrade from <= 9.2, the pg_multixact/offsets SLRU area might
Fix pg_upgrade failure from servers older than 9.3 When upgrading from servers of versions 9.2 and older, and MultiXactIds have been used in the old server beyond the first page (that is, 2048 multis or more in the default 8kB-page build), pg_upgrade would set the next multixact offset to use beyond what has been allocated in the new cluster. This would cause a failure the first time the new cluster needs to use this value, because the pg_multixact/offsets/ file wouldn't exist or wouldn't be large enough. To fix, ensure that the transient server instances launched by pg_upgrade extend the file as necessary. Per report from Jesse Denardo in CANiVXAj4c88YqipsyFQPboqMudnjcNTdB3pqe8ReXqAFQ=HXyA@mail.gmail.com
2013-08-19 18:33:07 +02:00
* contain files that are shorter than necessary; this would occur if the old
* installation had used multixacts beyond the first page (files cannot be
* copied, because the on-disk representation is different). pg_upgrade would
* update pg_control to set the next offset value to be at that position, so
* that tuples marked as locked by such MultiXacts would be seen as visible
* without having to consult multixact. However, trying to create and use a
* new MultiXactId would result in an error because the page on which the new
* value would reside does not exist. This routine is in charge of creating
* such pages.
*/
static void
MaybeExtendOffsetSlru(void)
{
int pageno;
pageno = MultiXactIdToOffsetPage(MultiXactState->nextMXact);
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
Fix pg_upgrade failure from servers older than 9.3 When upgrading from servers of versions 9.2 and older, and MultiXactIds have been used in the old server beyond the first page (that is, 2048 multis or more in the default 8kB-page build), pg_upgrade would set the next multixact offset to use beyond what has been allocated in the new cluster. This would cause a failure the first time the new cluster needs to use this value, because the pg_multixact/offsets/ file wouldn't exist or wouldn't be large enough. To fix, ensure that the transient server instances launched by pg_upgrade extend the file as necessary. Per report from Jesse Denardo in CANiVXAj4c88YqipsyFQPboqMudnjcNTdB3pqe8ReXqAFQ=HXyA@mail.gmail.com
2013-08-19 18:33:07 +02:00
if (!SimpleLruDoesPhysicalPageExist(MultiXactOffsetCtl, pageno))
{
int slotno;
/*
* Fortunately for us, SimpleLruWritePage is already prepared to deal
* with creating a new segment file even if the page we're writing is
* not the first in it, so this is enough.
*/
slotno = ZeroMultiXactOffsetPage(pageno, false);
SimpleLruWritePage(MultiXactOffsetCtl, slotno);
}
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactOffsetSLRULock);
Fix pg_upgrade failure from servers older than 9.3 When upgrading from servers of versions 9.2 and older, and MultiXactIds have been used in the old server beyond the first page (that is, 2048 multis or more in the default 8kB-page build), pg_upgrade would set the next multixact offset to use beyond what has been allocated in the new cluster. This would cause a failure the first time the new cluster needs to use this value, because the pg_multixact/offsets/ file wouldn't exist or wouldn't be large enough. To fix, ensure that the transient server instances launched by pg_upgrade extend the file as necessary. Per report from Jesse Denardo in CANiVXAj4c88YqipsyFQPboqMudnjcNTdB3pqe8ReXqAFQ=HXyA@mail.gmail.com
2013-08-19 18:33:07 +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
/*
* This must be called ONCE during postmaster or standalone-backend startup.
*
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
* StartupXLOG has already established nextMXact/nextOffset by calling
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
* MultiXactSetNextMXact and/or MultiXactAdvanceNextMXact, and the oldestMulti
Truncate pg_multixact/'s contents during crash recovery Commit 9dc842f08 of 8.2 era prevented MultiXact truncation during crash recovery, because there was no guarantee that enough state had been setup, and because it wasn't deemed to be a good idea to remove data during crash recovery anyway. Since then, due to Hot-Standby, streaming replication and PITR, the amount of time a cluster can spend doing crash recovery has increased significantly, to the point that a cluster may even never come out of it. This has made not truncating the content of pg_multixact/ not defensible anymore. To fix, take care to setup enough state for multixact truncation before crash recovery starts (easy since checkpoints contain the required information), and move the current end-of-recovery actions to a new TrimMultiXact() function, analogous to TrimCLOG(). At some later point, this should probably done similarly to the way clog.c is doing it, which is to just WAL log truncations, but we can't do that for the back branches. Back-patch to 9.0. 8.4 also has the problem, but since there's no hot standby there, it's much less pressing. In 9.2 and earlier, this patch is simpler than in newer branches, because multixact access during recovery isn't required. Add appropriate checks to make sure that's not happening. Andres Freund
2013-11-29 15:26:41 +01:00
* info from pg_control and/or MultiXactAdvanceOldest, but we haven't yet
* replayed WAL.
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
*/
void
StartupMultiXact(void)
Truncate pg_multixact/'s contents during crash recovery Commit 9dc842f08 of 8.2 era prevented MultiXact truncation during crash recovery, because there was no guarantee that enough state had been setup, and because it wasn't deemed to be a good idea to remove data during crash recovery anyway. Since then, due to Hot-Standby, streaming replication and PITR, the amount of time a cluster can spend doing crash recovery has increased significantly, to the point that a cluster may even never come out of it. This has made not truncating the content of pg_multixact/ not defensible anymore. To fix, take care to setup enough state for multixact truncation before crash recovery starts (easy since checkpoints contain the required information), and move the current end-of-recovery actions to a new TrimMultiXact() function, analogous to TrimCLOG(). At some later point, this should probably done similarly to the way clog.c is doing it, which is to just WAL log truncations, but we can't do that for the back branches. Back-patch to 9.0. 8.4 also has the problem, but since there's no hot standby there, it's much less pressing. In 9.2 and earlier, this patch is simpler than in newer branches, because multixact access during recovery isn't required. Add appropriate checks to make sure that's not happening. Andres Freund
2013-11-29 15:26:41 +01:00
{
MultiXactId multi = MultiXactState->nextMXact;
MultiXactOffset offset = MultiXactState->nextOffset;
int pageno;
/*
* Initialize offset's idea of the latest page number.
*/
pageno = MultiXactIdToOffsetPage(multi);
MultiXactOffsetCtl->shared->latest_page_number = pageno;
/*
* Initialize member's idea of the latest page number.
*/
pageno = MXOffsetToMemberPage(offset);
MultiXactMemberCtl->shared->latest_page_number = pageno;
}
/*
* This must be called ONCE at the end of startup/recovery.
*/
void
TrimMultiXact(void)
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
{
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
MultiXactId nextMXact;
MultiXactOffset offset;
MultiXactId oldestMXact;
Oid oldestMXactDB;
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
int pageno;
int entryno;
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
int flagsoff;
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
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
LWLockAcquire(MultiXactGenLock, LW_SHARED);
nextMXact = MultiXactState->nextMXact;
offset = MultiXactState->nextOffset;
oldestMXact = MultiXactState->oldestMultiXactId;
oldestMXactDB = MultiXactState->oldestMultiXactDB;
LWLockRelease(MultiXactGenLock);
Fix pg_upgrade failure from servers older than 9.3 When upgrading from servers of versions 9.2 and older, and MultiXactIds have been used in the old server beyond the first page (that is, 2048 multis or more in the default 8kB-page build), pg_upgrade would set the next multixact offset to use beyond what has been allocated in the new cluster. This would cause a failure the first time the new cluster needs to use this value, because the pg_multixact/offsets/ file wouldn't exist or wouldn't be large enough. To fix, ensure that the transient server instances launched by pg_upgrade extend the file as necessary. Per report from Jesse Denardo in CANiVXAj4c88YqipsyFQPboqMudnjcNTdB3pqe8ReXqAFQ=HXyA@mail.gmail.com
2013-08-19 18:33:07 +02:00
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
/* Clean up offsets state */
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockAcquire(MultiXactOffsetSLRULock, LW_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
/*
Have multixact be truncated by checkpoint, not vacuum Instead of truncating pg_multixact at vacuum time, do it only at checkpoint time. The reason for doing it this way is twofold: first, we want it to delete only segments that we're certain will not be required if there's a crash immediately after the removal; and second, we want to do it relatively often so that older files are not left behind if there's an untimely crash. Per my proposal in http://www.postgresql.org/message-id/20140626044519.GJ7340@eldon.alvh.no-ip.org we now execute the truncation in the checkpointer process rather than as part of vacuum. Vacuum is in only charge of maintaining in shared memory the value to which it's possible to truncate the files; that value is stored as part of checkpoints also, and so upon recovery we can reuse the same value to re-execute truncate and reset the oldest-value-still-safe-to-use to one known to remain after truncation. Per bug reported by Jeff Janes in the course of his tests involving bug #8673. While at it, update some comments that hadn't been updated since multixacts were changed. Backpatch to 9.3, where persistency of pg_multixact files was introduced by commit 0ac5ad5134f2.
2014-06-27 20:43:53 +02:00
* (Re-)Initialize our idea of the latest page number for offsets.
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
*/
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
pageno = MultiXactIdToOffsetPage(nextMXact);
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
MultiXactOffsetCtl->shared->latest_page_number = pageno;
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
/*
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
* Zero out the remainder of the current offsets page. See notes in
Fix comments about WAL rule "write xlog before data" versus pg_multixact. Recovery does not achieve its goal of zeroing all pg_multixact entries whose accompanying WAL records never reached disk. Remove that claim and justify its expendability. Detail the need for TrimMultiXact(), which has little in common with the TrimCLOG() rationale. Merge two tightly-related comments. Stop presenting pg_multixact as specific to heap_lock_tuple(); PostgreSQL 9.3 extended its use to heap_update(). Noticed while investigating a report from Andres Freund.
2016-01-01 07:46:46 +01:00
* TrimCLOG() for background. Unlike CLOG, some WAL record covers every
* pg_multixact SLRU mutation. Since, also unlike CLOG, we ignore the WAL
* rule "write xlog before data," nextMXact successors may carry obsolete,
* nonzero offset values. Zero those so case 2 of GetMultiXactIdMembers()
* operates normally.
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
*/
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
entryno = MultiXactIdToOffsetEntry(nextMXact);
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
if (entryno != 0)
{
int slotno;
MultiXactOffset *offptr;
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
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
slotno = SimpleLruReadPage(MultiXactOffsetCtl, pageno, true, nextMXact);
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
offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
offptr += entryno;
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
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
MemSet(offptr, 0, BLCKSZ - (entryno * sizeof(MultiXactOffset)));
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 representation of SLRU page state. This is the cleaner fix for the SLRU race condition that I posted a few days ago, but we decided not to use in 8.1 and older branches.
2005-11-05 22:19:47 +01:00
MultiXactOffsetCtl->shared->page_dirty[slotno] = true;
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
}
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactOffsetSLRULock);
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
/* And the same for members */
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockAcquire(MultiXactMemberSLRULock, LW_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
/*
Have multixact be truncated by checkpoint, not vacuum Instead of truncating pg_multixact at vacuum time, do it only at checkpoint time. The reason for doing it this way is twofold: first, we want it to delete only segments that we're certain will not be required if there's a crash immediately after the removal; and second, we want to do it relatively often so that older files are not left behind if there's an untimely crash. Per my proposal in http://www.postgresql.org/message-id/20140626044519.GJ7340@eldon.alvh.no-ip.org we now execute the truncation in the checkpointer process rather than as part of vacuum. Vacuum is in only charge of maintaining in shared memory the value to which it's possible to truncate the files; that value is stored as part of checkpoints also, and so upon recovery we can reuse the same value to re-execute truncate and reset the oldest-value-still-safe-to-use to one known to remain after truncation. Per bug reported by Jeff Janes in the course of his tests involving bug #8673. While at it, update some comments that hadn't been updated since multixacts were changed. Backpatch to 9.3, where persistency of pg_multixact files was introduced by commit 0ac5ad5134f2.
2014-06-27 20:43:53 +02:00
* (Re-)Initialize our idea of the latest page number for members.
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
*/
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
pageno = MXOffsetToMemberPage(offset);
MultiXactMemberCtl->shared->latest_page_number = pageno;
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
/*
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
* Zero out the remainder of the current members page. See notes in
Fix timing of Startup CLOG and MultiXact during Hot Standby Patch by me, bug report by Chris Redekop, analysis by Florian Pflug
2011-11-02 09:07:44 +01:00
* TrimCLOG() for motivation.
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
flagsoff = MXOffsetToFlagsOffset(offset);
if (flagsoff != 0)
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
{
int slotno;
TransactionId *xidptr;
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
int memberoff;
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
memberoff = MXOffsetToMemberOffset(offset);
Support an optional asynchronous commit mode, in which we don't flush WAL before reporting a transaction committed. Data consistency is still guaranteed (unlike setting fsync = off), but a crash may lose the effects of the last few transactions. Patch by Simon, some editorialization by Tom.
2007-08-02 00:45:09 +02:00
slotno = SimpleLruReadPage(MultiXactMemberCtl, pageno, true, offset);
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
xidptr = (TransactionId *)
(MultiXactMemberCtl->shared->page_buffer[slotno] + memberoff);
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
MemSet(xidptr, 0, BLCKSZ - memberoff);
/*
* Note: we don't need to zero out the flag bits in the remaining
* members of the current group, because they are always reset before
* writing.
*/
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 representation of SLRU page state. This is the cleaner fix for the SLRU race condition that I posted a few days ago, but we decided not to use in 8.1 and older branches.
2005-11-05 22:19:47 +01:00
MultiXactMemberCtl->shared->page_dirty[slotno] = true;
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
}
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
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactMemberSLRULock);
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
/* signal that we're officially up */
LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
MultiXactState->finishedStartup = true;
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
LWLockRelease(MultiXactGenLock);
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
/* Now compute how far away the next members wraparound is. */
Make logging about multixact wraparound protection less chatty. The original messaging design, introduced in commit 068cfadf9, seems too chatty now that some time has elapsed since the bug fix; most installations will be in good shape and don't really need a reminder about this on every postmaster start. Hence, arrange to suppress the "wraparound protections are now enabled" message during startup (specifically, during the TrimMultiXact() call). The message will still appear if protection becomes effective at some later point. Discussion: https://postgr.es/m/17211.1489189214@sss.pgh.pa.us
2017-03-14 17:47:46 +01:00
SetMultiXactIdLimit(oldestMXact, oldestMXactDB, 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
}
/*
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 MultiXact data to save in a checkpoint 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
*/
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
void
MultiXactGetCheckptMulti(bool is_shutdown,
MultiXactId *nextMulti,
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
MultiXactOffset *nextMultiOffset,
MultiXactId *oldestMulti,
Oid *oldestMultiDB)
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
{
LWLockAcquire(MultiXactGenLock, LW_SHARED);
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
*nextMulti = MultiXactState->nextMXact;
*nextMultiOffset = MultiXactState->nextOffset;
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
*oldestMulti = MultiXactState->oldestMultiXactId;
*oldestMultiDB = MultiXactState->oldestMultiXactDB;
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
LWLockRelease(MultiXactGenLock);
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
debug_elog6(DEBUG2,
"MultiXact: checkpoint is nextMulti %u, nextOffset %u, oldestMulti %u in DB %u",
*nextMulti, *nextMultiOffset, *oldestMulti, *oldestMultiDB);
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
}
/*
* Perform a checkpoint --- either during shutdown, or on-the-fly
*/
void
CheckPointMultiXact(void)
{
Add a few more DTrace probes to the backend. Robert Lor
2008-08-01 15:16:09 +02:00
TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_START(true);
Defer flushing of SLRU files. Previously, we called fsync() after writing out individual pg_xact, pg_multixact and pg_commit_ts pages due to cache pressure, leading to regular I/O stalls in user backends and recovery. Collapse requests for the same file into a single system call as part of the next checkpoint, as we already did for relation files, using the infrastructure developed by commit 3eb77eba. This can cause a significant improvement to recovery performance, especially when it's otherwise CPU-bound. Hoist ProcessSyncRequests() up into CheckPointGuts() to make it clearer that it applies to all the SLRU mini-buffer-pools as well as the main buffer pool. Rearrange things so that data collected in CheckpointStats includes SLRU activity. Also remove the Shutdown{CLOG,CommitTS,SUBTRANS,MultiXact}() functions, because they were redundant after the shutdown checkpoint that immediately precedes them. (I'm not sure if they were ever needed, but they aren't now.) Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us> (parts) Tested-by: Jakub Wartak <Jakub.Wartak@tomtom.com> Discussion: https://postgr.es/m/CA+hUKGLJ=84YT+NvhkEEDAuUtVHMfQ9i-N7k_o50JmQ6Rpj_OQ@mail.gmail.com
2020-09-25 08:49:43 +02:00
/*
* Write dirty MultiXact pages to disk. This may result in sync requests
* queued for later handling by ProcessSyncRequests(), as part of the
* checkpoint.
*/
SimpleLruWriteAll(MultiXactOffsetCtl, true);
SimpleLruWriteAll(MultiXactMemberCtl, 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
Add a few more DTrace probes to the backend. Robert Lor
2008-08-01 15:16:09 +02:00
TRACE_POSTGRESQL_MULTIXACT_CHECKPOINT_DONE(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
}
/*
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
* Set the next-to-be-assigned MultiXactId and offset
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
*
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
* This is used when we can determine the correct next ID/offset exactly
Add locking around WAL-replay modification of shared-memory variables. Originally, most of this code assumed that no Postgres backends could be running concurrently with it, and so no locking could be needed. That assumption fails in Hot Standby. While it's still true that Hot Standby backends should never change values like nextXid, they can examine them, and consistency is important in some cases such as when computing a snapshot. Therefore, prudence requires that WAL replay code obtain the relevant locks when modifying such variables, even though it can examine them without taking a lock. We were following that coding rule in some places but not all. This commit applies the coding rule uniformly to all updates of ShmemVariableCache and MultiXactState fields; a search of the replay routines did not find any other cases that seemed to be at risk. In addition, this commit fixes a longstanding thinko in replay of NEXTOID and checkpoint records: we tried to advance nextOid only if it was behind the value in the WAL record, but the comparison would draw the wrong conclusion if OID wraparound had occurred since the previous value. Better to just unconditionally assign the new value, since OID assignment shouldn't be happening during replay anyway. The additional locking seems to be more in the nature of future-proofing than fixing any live bug, so I am not going to back-patch it. The NEXTOID fix will be back-patched separately.
2012-02-06 18:34:10 +01:00
* from a checkpoint record. Although this is only called during bootstrap
* and XLog replay, we take the lock in case any hot-standby backends are
* examining the values.
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
*/
void
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
MultiXactSetNextMXact(MultiXactId nextMulti,
MultiXactOffset nextMultiOffset)
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
{
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
debug_elog4(DEBUG2, "MultiXact: setting next multi to %u offset %u",
nextMulti, nextMultiOffset);
Add locking around WAL-replay modification of shared-memory variables. Originally, most of this code assumed that no Postgres backends could be running concurrently with it, and so no locking could be needed. That assumption fails in Hot Standby. While it's still true that Hot Standby backends should never change values like nextXid, they can examine them, and consistency is important in some cases such as when computing a snapshot. Therefore, prudence requires that WAL replay code obtain the relevant locks when modifying such variables, even though it can examine them without taking a lock. We were following that coding rule in some places but not all. This commit applies the coding rule uniformly to all updates of ShmemVariableCache and MultiXactState fields; a search of the replay routines did not find any other cases that seemed to be at risk. In addition, this commit fixes a longstanding thinko in replay of NEXTOID and checkpoint records: we tried to advance nextOid only if it was behind the value in the WAL record, but the comparison would draw the wrong conclusion if OID wraparound had occurred since the previous value. Better to just unconditionally assign the new value, since OID assignment shouldn't be happening during replay anyway. The additional locking seems to be more in the nature of future-proofing than fixing any live bug, so I am not going to back-patch it. The NEXTOID fix will be back-patched separately.
2012-02-06 18:34:10 +01:00
LWLockAcquire(MultiXactGenLock, LW_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
MultiXactState->nextMXact = nextMulti;
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
MultiXactState->nextOffset = nextMultiOffset;
Add locking around WAL-replay modification of shared-memory variables. Originally, most of this code assumed that no Postgres backends could be running concurrently with it, and so no locking could be needed. That assumption fails in Hot Standby. While it's still true that Hot Standby backends should never change values like nextXid, they can examine them, and consistency is important in some cases such as when computing a snapshot. Therefore, prudence requires that WAL replay code obtain the relevant locks when modifying such variables, even though it can examine them without taking a lock. We were following that coding rule in some places but not all. This commit applies the coding rule uniformly to all updates of ShmemVariableCache and MultiXactState fields; a search of the replay routines did not find any other cases that seemed to be at risk. In addition, this commit fixes a longstanding thinko in replay of NEXTOID and checkpoint records: we tried to advance nextOid only if it was behind the value in the WAL record, but the comparison would draw the wrong conclusion if OID wraparound had occurred since the previous value. Better to just unconditionally assign the new value, since OID assignment shouldn't be happening during replay anyway. The additional locking seems to be more in the nature of future-proofing than fixing any live bug, so I am not going to back-patch it. The NEXTOID fix will be back-patched separately.
2012-02-06 18:34:10 +01:00
LWLockRelease(MultiXactGenLock);
Fix pg_upgrade's multixact handling (again) We need to create the pg_multixact/offsets file deleted by pg_upgrade much earlier than we originally were: it was in TrimMultiXact(), which runs after we exit recovery, but it actually needs to run earlier than the first call to SetMultiXactIdLimit (before recovery), because that routine already wants to read the first offset segment. Per pg_upgrade trouble report from Jeff Janes. While at it, silence a compiler warning about a pointless assert that an unsigned variable was being tested non-negative. This was a signed constant in Thomas Munro's patch which I changed to unsigned before commit. Pointed out by Andres Freund.
2015-04-30 18:55:06 +02:00
/*
* During a binary upgrade, make sure that the offsets SLRU is large
* enough to contain the next value that would be created.
*
* We need to do this pretty early during the first startup in binary
* upgrade mode: before StartupMultiXact() in fact, because this routine
* is called even before that by StartupXLOG(). And we can't do it
* earlier than at this point, because during that first call of this
* routine we determine the MultiXactState->nextMXact value that
* MaybeExtendOffsetSlru needs.
*/
if (IsBinaryUpgrade)
MaybeExtendOffsetSlru();
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
/*
* Determine the last safe MultiXactId to allocate given the currently oldest
* datminmxid (ie, the oldest MultiXactId that might exist in any database
* of our cluster), and the OID of the (or a) database with that value.
Make logging about multixact wraparound protection less chatty. The original messaging design, introduced in commit 068cfadf9, seems too chatty now that some time has elapsed since the bug fix; most installations will be in good shape and don't really need a reminder about this on every postmaster start. Hence, arrange to suppress the "wraparound protections are now enabled" message during startup (specifically, during the TrimMultiXact() call). The message will still appear if protection becomes effective at some later point. Discussion: https://postgr.es/m/17211.1489189214@sss.pgh.pa.us
2017-03-14 17:47:46 +01:00
*
* is_startup is true when we are just starting the cluster, false when we
* are updating state in a running cluster. This only affects log messages.
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
*/
void
Make logging about multixact wraparound protection less chatty. The original messaging design, introduced in commit 068cfadf9, seems too chatty now that some time has elapsed since the bug fix; most installations will be in good shape and don't really need a reminder about this on every postmaster start. Hence, arrange to suppress the "wraparound protections are now enabled" message during startup (specifically, during the TrimMultiXact() call). The message will still appear if protection becomes effective at some later point. Discussion: https://postgr.es/m/17211.1489189214@sss.pgh.pa.us
2017-03-14 17:47:46 +01:00
SetMultiXactIdLimit(MultiXactId oldest_datminmxid, Oid oldest_datoid,
bool is_startup)
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
{
MultiXactId multiVacLimit;
MultiXactId multiWarnLimit;
MultiXactId multiStopLimit;
MultiXactId multiWrapLimit;
MultiXactId curMulti;
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
bool needs_offset_vacuum;
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(MultiXactIdIsValid(oldest_datminmxid));
/*
Advance the stop point for multixact offset creation only at checkpoint. Commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c advanced the stop point at vacuum time, but this has subsequently been shown to be unsafe as a result of analysis by myself and Thomas Munro and testing by Thomas Munro. The crux of the problem is that the SLRU deletion logic may get confused about what to remove if, at exactly the right time during the checkpoint process, the head of the SLRU crosses what used to be the tail. This patch, by me, fixes the problem by advancing the stop point only following a checkpoint. This has the additional advantage of making the removal logic work during recovery more like the way it works during normal running, which is probably good. At least one of the calls to DetermineSafeOldestOffset which this patch removes was already dead, because MultiXactAdvanceOldest is called only during recovery and DetermineSafeOldestOffset was set up to do nothing during recovery. That, however, is inconsistent with the principle that recovery and normal running should work similarly, and was confusing to boot. Along the way, fix some comments that previous patches in this area neglected to update. It's not clear to me whether there's any concrete basis for the decision to use only half of the multixact ID space, but it's neither necessary nor sufficient to prevent multixact member wraparound, so the comments should not say otherwise.
2015-05-11 04:21:20 +02:00
* We pretend that a wrap will happen halfway through the multixact ID
* space, but that's not really true, because multixacts wrap differently
* from transaction IDs. Note that, separately from any concern about
* multixact IDs wrapping, we must ensure that multixact members do not
Update stale comments, and fix comment typos.
2019-06-08 19:12:26 +02:00
* wrap. Limits for that are set in SetOffsetVacuumLimit, not here.
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
*/
multiWrapLimit = oldest_datminmxid + (MaxMultiXactId >> 1);
if (multiWrapLimit < FirstMultiXactId)
multiWrapLimit += FirstMultiXactId;
/*
Change XID and mxact limits to warn at 40M and stop at 3M. We have edge-case bugs when assigning values in the last few dozen pages before the wrap limit. We may introduce similar bugs in the future. At default BLCKSZ, this makes such bugs unreachable outside of single-user mode. Also, when VACUUM began to consume mxacts, multiStopLimit did not change to compensate. pg_upgrade may fail on a cluster that was already printing "must be vacuumed" warnings. Follow the warning's instructions to clear the warning, then run pg_upgrade again. One can still, peacefully consume 98% of XIDs or mxacts, so DBAs need not change routine VACUUM settings. Discussion: https://postgr.es/m/20200621083513.GA3074645@rfd.leadboat.com
2020-08-02 00:31:01 +02:00
* We'll refuse to continue assigning MultiXactIds once we get within 3M
* multi of data loss. See SetTransactionIdLimit.
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 XID and mxact limits to warn at 40M and stop at 3M. We have edge-case bugs when assigning values in the last few dozen pages before the wrap limit. We may introduce similar bugs in the future. At default BLCKSZ, this makes such bugs unreachable outside of single-user mode. Also, when VACUUM began to consume mxacts, multiStopLimit did not change to compensate. pg_upgrade may fail on a cluster that was already printing "must be vacuumed" warnings. Follow the warning's instructions to clear the warning, then run pg_upgrade again. One can still, peacefully consume 98% of XIDs or mxacts, so DBAs need not change routine VACUUM settings. Discussion: https://postgr.es/m/20200621083513.GA3074645@rfd.leadboat.com
2020-08-02 00:31:01 +02:00
multiStopLimit = multiWrapLimit - 3000000;
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 (multiStopLimit < FirstMultiXactId)
multiStopLimit -= FirstMultiXactId;
/*
Change XID and mxact limits to warn at 40M and stop at 3M. We have edge-case bugs when assigning values in the last few dozen pages before the wrap limit. We may introduce similar bugs in the future. At default BLCKSZ, this makes such bugs unreachable outside of single-user mode. Also, when VACUUM began to consume mxacts, multiStopLimit did not change to compensate. pg_upgrade may fail on a cluster that was already printing "must be vacuumed" warnings. Follow the warning's instructions to clear the warning, then run pg_upgrade again. One can still, peacefully consume 98% of XIDs or mxacts, so DBAs need not change routine VACUUM settings. Discussion: https://postgr.es/m/20200621083513.GA3074645@rfd.leadboat.com
2020-08-02 00:31:01 +02:00
* We'll start complaining loudly when we get within 40M multis of data
* loss. This is kind of arbitrary, but if you let your gas gauge get
* down to 2% of full, would you be looking for the next gas station? We
* need to be fairly liberal about this number because there are lots of
* scenarios where most transactions are done by automatic clients that
* won't pay attention to warnings. (No, we're not gonna make this
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
* configurable. If you know enough to configure it, you know enough to
* not get in this kind of trouble in the first place.)
*/
Change XID and mxact limits to warn at 40M and stop at 3M. We have edge-case bugs when assigning values in the last few dozen pages before the wrap limit. We may introduce similar bugs in the future. At default BLCKSZ, this makes such bugs unreachable outside of single-user mode. Also, when VACUUM began to consume mxacts, multiStopLimit did not change to compensate. pg_upgrade may fail on a cluster that was already printing "must be vacuumed" warnings. Follow the warning's instructions to clear the warning, then run pg_upgrade again. One can still, peacefully consume 98% of XIDs or mxacts, so DBAs need not change routine VACUUM settings. Discussion: https://postgr.es/m/20200621083513.GA3074645@rfd.leadboat.com
2020-08-02 00:31:01 +02:00
multiWarnLimit = multiWrapLimit - 40000000;
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 (multiWarnLimit < FirstMultiXactId)
multiWarnLimit -= FirstMultiXactId;
/*
* We'll start trying to force autovacuums when oldest_datminmxid gets to
Separate multixact freezing parameters from xid's Previously we were piggybacking on transaction ID parameters to freeze multixacts; but since there isn't necessarily any relationship between rates of Xid and multixact consumption, this turns out not to be a good idea. Therefore, we now have multixact-specific freezing parameters: vacuum_multixact_freeze_min_age: when to remove multis as we come across them in vacuum (default to 5 million, i.e. early in comparison to Xid's default of 50 million) vacuum_multixact_freeze_table_age: when to force whole-table scans instead of scanning only the pages marked as not all visible in visibility map (default to 150 million, same as for Xids). Whichever of both which reaches the 150 million mark earlier will cause a whole-table scan. autovacuum_multixact_freeze_max_age: when for cause emergency, uninterruptible whole-table scans (default to 400 million, double as that for Xids). This means there shouldn't be more frequent emergency vacuuming than previously, unless multixacts are being used very rapidly. Backpatch to 9.3 where multixacts were made to persist enough to require freezing. To avoid an ABI break in 9.3, VacuumStmt has a couple of fields in an unnatural place, and StdRdOptions is split in two so that the newly added fields can go at the end. Patch by me, reviewed by Robert Haas, with additional input from Andres Freund and Tom Lane.
2014-02-13 23:30:30 +01:00
* be more than autovacuum_multixact_freeze_max_age mxids old.
Replace hardcoded 200000000 with autovacuum_freeze_max_age Parts of the code used autovacuum_freeze_max_age to determine whether anti-multixact-wraparound vacuums are necessary, while others used a hardcoded 200000000 value. This leads to problems when autovacuum_freeze_max_age is set to a non-default value. Use the latter everywhere. Backpatch to 9.3, where vacuuming of multixacts was introduced. Andres Freund
2013-11-28 20:45:29 +01:00
*
Separate multixact freezing parameters from xid's Previously we were piggybacking on transaction ID parameters to freeze multixacts; but since there isn't necessarily any relationship between rates of Xid and multixact consumption, this turns out not to be a good idea. Therefore, we now have multixact-specific freezing parameters: vacuum_multixact_freeze_min_age: when to remove multis as we come across them in vacuum (default to 5 million, i.e. early in comparison to Xid's default of 50 million) vacuum_multixact_freeze_table_age: when to force whole-table scans instead of scanning only the pages marked as not all visible in visibility map (default to 150 million, same as for Xids). Whichever of both which reaches the 150 million mark earlier will cause a whole-table scan. autovacuum_multixact_freeze_max_age: when for cause emergency, uninterruptible whole-table scans (default to 400 million, double as that for Xids). This means there shouldn't be more frequent emergency vacuuming than previously, unless multixacts are being used very rapidly. Backpatch to 9.3 where multixacts were made to persist enough to require freezing. To avoid an ABI break in 9.3, VacuumStmt has a couple of fields in an unnatural place, and StdRdOptions is split in two so that the newly added fields can go at the end. Patch by me, reviewed by Robert Haas, with additional input from Andres Freund and Tom Lane.
2014-02-13 23:30:30 +01:00
* Note: autovacuum_multixact_freeze_max_age is a PGC_POSTMASTER parameter
* so that we don't have to worry about dealing with on-the-fly changes in
* its value. See SetTransactionIdLimit.
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
*/
Separate multixact freezing parameters from xid's Previously we were piggybacking on transaction ID parameters to freeze multixacts; but since there isn't necessarily any relationship between rates of Xid and multixact consumption, this turns out not to be a good idea. Therefore, we now have multixact-specific freezing parameters: vacuum_multixact_freeze_min_age: when to remove multis as we come across them in vacuum (default to 5 million, i.e. early in comparison to Xid's default of 50 million) vacuum_multixact_freeze_table_age: when to force whole-table scans instead of scanning only the pages marked as not all visible in visibility map (default to 150 million, same as for Xids). Whichever of both which reaches the 150 million mark earlier will cause a whole-table scan. autovacuum_multixact_freeze_max_age: when for cause emergency, uninterruptible whole-table scans (default to 400 million, double as that for Xids). This means there shouldn't be more frequent emergency vacuuming than previously, unless multixacts are being used very rapidly. Backpatch to 9.3 where multixacts were made to persist enough to require freezing. To avoid an ABI break in 9.3, VacuumStmt has a couple of fields in an unnatural place, and StdRdOptions is split in two so that the newly added fields can go at the end. Patch by me, reviewed by Robert Haas, with additional input from Andres Freund and Tom Lane.
2014-02-13 23:30:30 +01:00
multiVacLimit = oldest_datminmxid + autovacuum_multixact_freeze_max_age;
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 (multiVacLimit < FirstMultiXactId)
multiVacLimit += FirstMultiXactId;
/* Grab lock for just long enough to set the new limit values */
LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
MultiXactState->oldestMultiXactId = oldest_datminmxid;
MultiXactState->oldestMultiXactDB = oldest_datoid;
MultiXactState->multiVacLimit = multiVacLimit;
MultiXactState->multiWarnLimit = multiWarnLimit;
MultiXactState->multiStopLimit = multiStopLimit;
MultiXactState->multiWrapLimit = multiWrapLimit;
curMulti = MultiXactState->nextMXact;
LWLockRelease(MultiXactGenLock);
/* Log the info */
ereport(DEBUG1,
Use errmsg_internal for debug messages An inconsistent set of debug-level messages was not using errmsg_internal(), thus uselessly exposing the messages to translation work. Fix those.
2021-02-17 11:24:46 +01:00
(errmsg_internal("MultiXactId wrap limit is %u, limited by database with OID %u",
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
multiWrapLimit, oldest_datoid)));
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
/*
* Computing the actual limits is only possible once the data directory is
* in a consistent state. There's no need to compute the limits while
* still replaying WAL - no decisions about new multis are made even
* though multixact creations might be replayed. So we'll only do further
* checks after TrimMultiXact() has been called.
*/
if (!MultiXactState->finishedStartup)
return;
Assert(!InRecovery);
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
/* Set limits for offset vacuum. */
Make logging about multixact wraparound protection less chatty. The original messaging design, introduced in commit 068cfadf9, seems too chatty now that some time has elapsed since the bug fix; most installations will be in good shape and don't really need a reminder about this on every postmaster start. Hence, arrange to suppress the "wraparound protections are now enabled" message during startup (specifically, during the TrimMultiXact() call). The message will still appear if protection becomes effective at some later point. Discussion: https://postgr.es/m/17211.1489189214@sss.pgh.pa.us
2017-03-14 17:47:46 +01:00
needs_offset_vacuum = SetOffsetVacuumLimit(is_startup);
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +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 past the autovacuum force point, immediately signal an autovac
* request. The reason for this is that autovac only processes one
* database per invocation. Once it's finished cleaning up the oldest
* database, it'll call here, and we'll signal the postmaster to start
* another iteration immediately if there are still any old databases.
*/
Even when autovacuum=off, force it for members as we do in other cases. Thomas Munro, with some adjustments by me.
2015-05-11 16:51:14 +02:00
if ((MultiXactIdPrecedes(multiVacLimit, curMulti) ||
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
needs_offset_vacuum) && IsUnderPostmaster)
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
SendPostmasterSignal(PMSIGNAL_START_AUTOVAC_LAUNCHER);
/* Give an immediate warning if past the wrap warn point */
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
if (MultiXactIdPrecedes(multiWarnLimit, curMulti))
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
{
char *oldest_datname;
/*
* We can be called when not inside a transaction, for example during
* StartupXLOG(). In such a case we cannot do database access, so we
* must just report the oldest DB's OID.
*
* Note: it's also possible that get_database_name fails and returns
* NULL, for example because the database just got dropped. We'll
* still warn, even though the warning might now be unnecessary.
*/
if (IsTransactionState())
oldest_datname = get_database_name(oldest_datoid);
else
oldest_datname = NULL;
if (oldest_datname)
ereport(WARNING,
Message punctuation and pluralization fixes
2013-08-09 13:59:53 +02:00
(errmsg_plural("database \"%s\" must be vacuumed before %u more MultiXactId is used",
"database \"%s\" must be vacuumed before %u more MultiXactIds are used",
multiWrapLimit - curMulti,
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
oldest_datname,
multiWrapLimit - curMulti),
errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
Extend near-wraparound hints to include replication slots Author: Feike Steenbergen Reviewed-by: Michael Paquier
2017-12-29 15:01:25 +01:00
"You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
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
ereport(WARNING,
Message punctuation and pluralization fixes
2013-08-09 13:59:53 +02:00
(errmsg_plural("database with OID %u must be vacuumed before %u more MultiXactId is used",
"database with OID %u must be vacuumed before %u more MultiXactIds are used",
multiWrapLimit - curMulti,
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
oldest_datoid,
multiWrapLimit - curMulti),
errhint("To avoid a database shutdown, execute a database-wide VACUUM in that database.\n"
Extend near-wraparound hints to include replication slots Author: Feike Steenbergen Reviewed-by: Michael Paquier
2017-12-29 15:01:25 +01:00
"You might also need to commit or roll back old prepared transactions, or drop stale replication slots.")));
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
/*
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
* Ensure the next-to-be-assigned MultiXactId is at least minMulti,
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
* and similarly nextOffset is at least minMultiOffset.
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
*
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
* This is used when we can determine minimum safe values from an XLog
* record (either an on-line checkpoint or an mxact creation log entry).
Add locking around WAL-replay modification of shared-memory variables. Originally, most of this code assumed that no Postgres backends could be running concurrently with it, and so no locking could be needed. That assumption fails in Hot Standby. While it's still true that Hot Standby backends should never change values like nextXid, they can examine them, and consistency is important in some cases such as when computing a snapshot. Therefore, prudence requires that WAL replay code obtain the relevant locks when modifying such variables, even though it can examine them without taking a lock. We were following that coding rule in some places but not all. This commit applies the coding rule uniformly to all updates of ShmemVariableCache and MultiXactState fields; a search of the replay routines did not find any other cases that seemed to be at risk. In addition, this commit fixes a longstanding thinko in replay of NEXTOID and checkpoint records: we tried to advance nextOid only if it was behind the value in the WAL record, but the comparison would draw the wrong conclusion if OID wraparound had occurred since the previous value. Better to just unconditionally assign the new value, since OID assignment shouldn't be happening during replay anyway. The additional locking seems to be more in the nature of future-proofing than fixing any live bug, so I am not going to back-patch it. The NEXTOID fix will be back-patched separately.
2012-02-06 18:34:10 +01:00
* Although this is only called during XLog replay, we take the lock in case
* any hot-standby backends are examining the values.
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
*/
void
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
MultiXactAdvanceNextMXact(MultiXactId minMulti,
MultiXactOffset minMultiOffset)
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
{
Add locking around WAL-replay modification of shared-memory variables. Originally, most of this code assumed that no Postgres backends could be running concurrently with it, and so no locking could be needed. That assumption fails in Hot Standby. While it's still true that Hot Standby backends should never change values like nextXid, they can examine them, and consistency is important in some cases such as when computing a snapshot. Therefore, prudence requires that WAL replay code obtain the relevant locks when modifying such variables, even though it can examine them without taking a lock. We were following that coding rule in some places but not all. This commit applies the coding rule uniformly to all updates of ShmemVariableCache and MultiXactState fields; a search of the replay routines did not find any other cases that seemed to be at risk. In addition, this commit fixes a longstanding thinko in replay of NEXTOID and checkpoint records: we tried to advance nextOid only if it was behind the value in the WAL record, but the comparison would draw the wrong conclusion if OID wraparound had occurred since the previous value. Better to just unconditionally assign the new value, since OID assignment shouldn't be happening during replay anyway. The additional locking seems to be more in the nature of future-proofing than fixing any live bug, so I am not going to back-patch it. The NEXTOID fix will be back-patched separately.
2012-02-06 18:34:10 +01:00
LWLockAcquire(MultiXactGenLock, LW_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
if (MultiXactIdPrecedes(MultiXactState->nextMXact, minMulti))
{
debug_elog3(DEBUG2, "MultiXact: setting next multi to %u", minMulti);
MultiXactState->nextMXact = minMulti;
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
}
if (MultiXactOffsetPrecedes(MultiXactState->nextOffset, minMultiOffset))
{
debug_elog3(DEBUG2, "MultiXact: setting next offset to %u",
minMultiOffset);
MultiXactState->nextOffset = minMultiOffset;
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
}
Add locking around WAL-replay modification of shared-memory variables. Originally, most of this code assumed that no Postgres backends could be running concurrently with it, and so no locking could be needed. That assumption fails in Hot Standby. While it's still true that Hot Standby backends should never change values like nextXid, they can examine them, and consistency is important in some cases such as when computing a snapshot. Therefore, prudence requires that WAL replay code obtain the relevant locks when modifying such variables, even though it can examine them without taking a lock. We were following that coding rule in some places but not all. This commit applies the coding rule uniformly to all updates of ShmemVariableCache and MultiXactState fields; a search of the replay routines did not find any other cases that seemed to be at risk. In addition, this commit fixes a longstanding thinko in replay of NEXTOID and checkpoint records: we tried to advance nextOid only if it was behind the value in the WAL record, but the comparison would draw the wrong conclusion if OID wraparound had occurred since the previous value. Better to just unconditionally assign the new value, since OID assignment shouldn't be happening during replay anyway. The additional locking seems to be more in the nature of future-proofing than fixing any live bug, so I am not going to back-patch it. The NEXTOID fix will be back-patched separately.
2012-02-06 18:34:10 +01:00
LWLockRelease(MultiXactGenLock);
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
/*
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* Update our oldestMultiXactId value, but only if it's more recent than what
* we had.
*
* This may only be called during WAL replay.
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
*/
void
MultiXactAdvanceOldest(MultiXactId oldestMulti, Oid oldestMultiDB)
{
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
Assert(InRecovery);
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 (MultiXactIdPrecedes(MultiXactState->oldestMultiXactId, oldestMulti))
Make logging about multixact wraparound protection less chatty. The original messaging design, introduced in commit 068cfadf9, seems too chatty now that some time has elapsed since the bug fix; most installations will be in good shape and don't really need a reminder about this on every postmaster start. Hence, arrange to suppress the "wraparound protections are now enabled" message during startup (specifically, during the TrimMultiXact() call). The message will still appear if protection becomes effective at some later point. Discussion: https://postgr.es/m/17211.1489189214@sss.pgh.pa.us
2017-03-14 17:47:46 +01:00
SetMultiXactIdLimit(oldestMulti, oldestMultiDB, false);
Have multixact be truncated by checkpoint, not vacuum Instead of truncating pg_multixact at vacuum time, do it only at checkpoint time. The reason for doing it this way is twofold: first, we want it to delete only segments that we're certain will not be required if there's a crash immediately after the removal; and second, we want to do it relatively often so that older files are not left behind if there's an untimely crash. Per my proposal in http://www.postgresql.org/message-id/20140626044519.GJ7340@eldon.alvh.no-ip.org we now execute the truncation in the checkpointer process rather than as part of vacuum. Vacuum is in only charge of maintaining in shared memory the value to which it's possible to truncate the files; that value is stored as part of checkpoints also, and so upon recovery we can reuse the same value to re-execute truncate and reset the oldest-value-still-safe-to-use to one known to remain after truncation. Per bug reported by Jeff Janes in the course of his tests involving bug #8673. While at it, update some comments that hadn't been updated since multixacts were changed. Backpatch to 9.3, where persistency of pg_multixact files was introduced by commit 0ac5ad5134f2.
2014-06-27 20:43:53 +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
/*
* Make sure that MultiXactOffset has room for a newly-allocated MultiXactId.
*
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
* NB: this is called while holding MultiXactGenLock. We want it to be very
* fast most of the time; even when it's not so fast, no actual I/O need
* happen unless we're forced to write out a dirty log or xlog page to make
* room in shared memory.
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
*/
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
static void
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
ExtendMultiXactOffset(MultiXactId multi)
{
int pageno;
/*
* No work except at first MultiXactId of a page. But beware: just after
* wraparound, the first MultiXactId of page zero is FirstMultiXactId.
*/
if (MultiXactIdToOffsetEntry(multi) != 0 &&
multi != FirstMultiXactId)
return;
pageno = MultiXactIdToOffsetPage(multi);
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
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
/* Zero the page and make an XLOG entry about it */
ZeroMultiXactOffsetPage(pageno, true);
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactOffsetSLRULock);
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 sure that MultiXactMember has room for the members of a newly-
* allocated MultiXactId.
*
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
* Like the above routine, this is called while holding MultiXactGenLock;
* same comments apply.
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
*/
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
static void
ExtendMultiXactMember(MultiXactOffset offset, int nmembers)
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
{
/*
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
* It's possible that the members span more than one page of the members
* file, so we loop to ensure we consider each page. The coding is not
* optimal if the members span several pages, but that seems unusual
* enough to not worry much about.
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
*/
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
while (nmembers > 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
int flagsoff;
int flagsbit;
Wrap multixact/members correctly during extension, take 2 In a50d97625497b7 I already changed this, but got it wrong for the case where the number of members is larger than the number of entries that fit in the last page of the last segment. As reported by Serge Negodyuck in a followup to bug #8673.
2014-06-09 21:17:23 +02:00
uint32 difference;
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
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
/*
* Only zero when at first entry of a page.
*/
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
flagsoff = MXOffsetToFlagsOffset(offset);
flagsbit = MXOffsetToFlagsBitShift(offset);
if (flagsoff == 0 && flagsbit == 0)
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
{
int pageno;
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
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
pageno = MXOffsetToMemberPage(offset);
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockAcquire(MultiXactMemberSLRULock, LW_EXCLUSIVE);
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
/* Zero the page and make an XLOG entry about it */
ZeroMultiXactMemberPage(pageno, true);
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactMemberSLRULock);
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
}
Wrap multixact/members correctly during extension In the 9.2 code for extending multixact/members, the logic was very simple because the number of entries in a members page was a proper divisor of 2^32, and thus at 2^32 wraparound the logic for page switch was identical than at any other page boundary. In commit 0ac5ad5134f I failed to realize this and introduced code that was not able to go over the 2^32 boundary. Fix that by ensuring that when we reach the last page of the last segment we correctly zero the initial page of the initial segment, using correct uint32-wraparound-safe arithmetic. Noticed while investigating bug #8673 reported by Serge Negodyuck, as diagnosed by Andres Freund.
2014-01-02 22:17:07 +01:00
/*
Wrap multixact/members correctly during extension, take 2 In a50d97625497b7 I already changed this, but got it wrong for the case where the number of members is larger than the number of entries that fit in the last page of the last segment. As reported by Serge Negodyuck in a followup to bug #8673.
2014-06-09 21:17:23 +02:00
* Compute the number of items till end of current page. Careful: if
* addition of unsigned ints wraps around, we're at the last page of
* the last segment; since that page holds a different number of items
* than other pages, we need to do it differently.
Wrap multixact/members correctly during extension In the 9.2 code for extending multixact/members, the logic was very simple because the number of entries in a members page was a proper divisor of 2^32, and thus at 2^32 wraparound the logic for page switch was identical than at any other page boundary. In commit 0ac5ad5134f I failed to realize this and introduced code that was not able to go over the 2^32 boundary. Fix that by ensuring that when we reach the last page of the last segment we correctly zero the initial page of the initial segment, using correct uint32-wraparound-safe arithmetic. Noticed while investigating bug #8673 reported by Serge Negodyuck, as diagnosed by Andres Freund.
2014-01-02 22:17:07 +01:00
*/
Wrap multixact/members correctly during extension, take 2 In a50d97625497b7 I already changed this, but got it wrong for the case where the number of members is larger than the number of entries that fit in the last page of the last segment. As reported by Serge Negodyuck in a followup to bug #8673.
2014-06-09 21:17:23 +02:00
if (offset + MAX_MEMBERS_IN_LAST_MEMBERS_PAGE < offset)
Wrap multixact/members correctly during extension In the 9.2 code for extending multixact/members, the logic was very simple because the number of entries in a members page was a proper divisor of 2^32, and thus at 2^32 wraparound the logic for page switch was identical than at any other page boundary. In commit 0ac5ad5134f I failed to realize this and introduced code that was not able to go over the 2^32 boundary. Fix that by ensuring that when we reach the last page of the last segment we correctly zero the initial page of the initial segment, using correct uint32-wraparound-safe arithmetic. Noticed while investigating bug #8673 reported by Serge Negodyuck, as diagnosed by Andres Freund.
2014-01-02 22:17:07 +01:00
{
Wrap multixact/members correctly during extension, take 2 In a50d97625497b7 I already changed this, but got it wrong for the case where the number of members is larger than the number of entries that fit in the last page of the last segment. As reported by Serge Negodyuck in a followup to bug #8673.
2014-06-09 21:17:23 +02:00
/*
* This is the last page of the last segment; we can compute the
* number of items left to allocate in it without modulo
* arithmetic.
*/
difference = MaxMultiXactOffset - offset + 1;
Wrap multixact/members correctly during extension In the 9.2 code for extending multixact/members, the logic was very simple because the number of entries in a members page was a proper divisor of 2^32, and thus at 2^32 wraparound the logic for page switch was identical than at any other page boundary. In commit 0ac5ad5134f I failed to realize this and introduced code that was not able to go over the 2^32 boundary. Fix that by ensuring that when we reach the last page of the last segment we correctly zero the initial page of the initial segment, using correct uint32-wraparound-safe arithmetic. Noticed while investigating bug #8673 reported by Serge Negodyuck, as diagnosed by Andres Freund.
2014-01-02 22:17:07 +01:00
}
else
difference = MULTIXACT_MEMBERS_PER_PAGE - offset % MULTIXACT_MEMBERS_PER_PAGE;
Wrap multixact/members correctly during extension, take 2 In a50d97625497b7 I already changed this, but got it wrong for the case where the number of members is larger than the number of entries that fit in the last page of the last segment. As reported by Serge Negodyuck in a followup to bug #8673.
2014-06-09 21:17:23 +02:00
/*
* Advance to next page, taking care to properly handle the wraparound
* case. OK if nmembers goes negative.
*/
nmembers -= difference;
offset += difference;
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
}
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
* GetOldestMultiXactId
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
* Return the oldest MultiXactId that's still possibly still seen as live by
* any running transaction. Older ones might still exist on disk, but they no
* longer have any running member transaction.
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
*
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
* It's not safe to truncate MultiXact SLRU segments on the value returned by
Update obsolete multixact.c comments. Commit 4f627f89 switched SLRU truncation for multixacts back to being a task performed during VACUUM, but missed some comments that continued to reference truncation happening as part of checkpointing. Update those comments now. Also update comments that became obsolete when commit c3ffa731 changed the way that vacuum_multixact_freeze_min_age is applied by VACUUM as it computes its MultiXactCutoff cutoff (which is used by VACUUM to decide what to freeze). Explain the same issues by referencing how OldestMxact is the latest valid value that relminmxid can ever be advanced to at the end of a VACUUM (following the work in commit 0b018fab).
2023-01-04 01:54:35 +01:00
* this function; however, it can be set as the new relminmxid for any table
* that VACUUM knows has no remaining MXIDs < the same value. It is only safe
* to truncate SLRUs when no table can possibly still have a referencing MXID.
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
MultiXactId
GetOldestMultiXactId(void)
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
MultiXactId oldestMXact;
MultiXactId nextMXact;
int i;
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
* This is the oldest valid value among all the OldestMemberMXactId[] and
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
* OldestVisibleMXactId[] entries, or nextMXact if none are valid.
*/
LWLockAcquire(MultiXactGenLock, LW_SHARED);
/*
* We have to beware of the possibility that nextMXact is in the
* wrapped-around state. We don't fix the counter itself here, but we
* must be sure to use a valid value in our calculation.
*/
nextMXact = MultiXactState->nextMXact;
if (nextMXact < FirstMultiXactId)
nextMXact = FirstMultiXactId;
oldestMXact = nextMXact;
Revert the addition of GetMaxBackends() and related stuff. This reverts commits 0147fc7, 4567596, aa64f23, and 5ecd018. There is no longer agreement that introducing this function was the right way to address the problem. The consensus now seems to favor trying to make a correct value for MaxBackends available to mdules executing their _PG_init() functions. Nathan Bossart Discussion: http://postgr.es/m/20220323045229.i23skfscdbvrsuxa@jrouhaud
2022-04-12 20:45:23 +02:00
for (i = 1; i <= MaxOldestSlot; i++)
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
{
MultiXactId thisoldest;
thisoldest = OldestMemberMXactId[i];
if (MultiXactIdIsValid(thisoldest) &&
MultiXactIdPrecedes(thisoldest, oldestMXact))
oldestMXact = thisoldest;
thisoldest = OldestVisibleMXactId[i];
if (MultiXactIdIsValid(thisoldest) &&
MultiXactIdPrecedes(thisoldest, oldestMXact))
oldestMXact = thisoldest;
}
LWLockRelease(MultiXactGenLock);
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 oldestMXact;
}
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
/*
* Determine how aggressively we need to vacuum in order to prevent member
* wraparound.
*
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* To do so determine what's the oldest member offset and install the limit
* info in MultiXactState, where it can be used to prevent overrun of old data
* in the members SLRU area.
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
*
* The return value is true if emergency autovacuum is required and false
* otherwise.
*/
static bool
Make logging about multixact wraparound protection less chatty. The original messaging design, introduced in commit 068cfadf9, seems too chatty now that some time has elapsed since the bug fix; most installations will be in good shape and don't really need a reminder about this on every postmaster start. Hence, arrange to suppress the "wraparound protections are now enabled" message during startup (specifically, during the TrimMultiXact() call). The message will still appear if protection becomes effective at some later point. Discussion: https://postgr.es/m/17211.1489189214@sss.pgh.pa.us
2017-03-14 17:47:46 +01:00
SetOffsetVacuumLimit(bool is_startup)
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
{
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
MultiXactId oldestMultiXactId;
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
MultiXactId nextMXact;
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
MultiXactOffset oldestOffset = 0; /* placate compiler */
MultiXactOffset prevOldestOffset;
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
MultiXactOffset nextOffset;
bool oldestOffsetKnown = false;
bool prevOldestOffsetKnown;
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
MultiXactOffset offsetStopLimit = 0;
Fix bug in SetOffsetVacuumLimit() triggered by find_multixact_start() failure. Previously, if find_multixact_start() failed, SetOffsetVacuumLimit() would install 0 into MultiXactState->offsetStopLimit if it previously succeeded. Luckily, there are no known cases where find_multixact_start() will return an error in 9.5 and above. But if it were to happen, for example due to filesystem permission issues, it'd be somewhat bad: GetNewMultiXactId() could continue allocating mxids even if close to a wraparound, or it could erroneously stop allocating mxids, even if no wraparound is looming. The wrong value would be corrected the next time SetOffsetVacuumLimit() is called, or by a restart. Reported-By: Noah Misch, although this is not his preferred fix Discussion: 20151210140450.GA22278@alap3.anarazel.de Backpatch: 9.5, where the bug was introduced as part of 4f627f
2015-12-14 11:34:16 +01:00
MultiXactOffset prevOffsetStopLimit;
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
/*
* NB: Have to prevent concurrent truncation, we might otherwise try to
Fix misc typos, mostly in comments. A collection of typos I happened to spot while reading code, as well as grepping for common mistakes. Backpatch to all supported versions, as applicable, to avoid conflicts when backporting other commits in the future.
2018-07-09 15:10:44 +02:00
* lookup an oldestMulti that's concurrently getting truncated away.
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
*/
LWLockAcquire(MultiXactTruncationLock, LW_SHARED);
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
/* Read relevant fields from shared memory. */
LWLockAcquire(MultiXactGenLock, LW_SHARED);
oldestMultiXactId = MultiXactState->oldestMultiXactId;
nextMXact = MultiXactState->nextMXact;
nextOffset = MultiXactState->nextOffset;
prevOldestOffsetKnown = MultiXactState->oldestOffsetKnown;
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
prevOldestOffset = MultiXactState->oldestOffset;
Fix bug in SetOffsetVacuumLimit() triggered by find_multixact_start() failure. Previously, if find_multixact_start() failed, SetOffsetVacuumLimit() would install 0 into MultiXactState->offsetStopLimit if it previously succeeded. Luckily, there are no known cases where find_multixact_start() will return an error in 9.5 and above. But if it were to happen, for example due to filesystem permission issues, it'd be somewhat bad: GetNewMultiXactId() could continue allocating mxids even if close to a wraparound, or it could erroneously stop allocating mxids, even if no wraparound is looming. The wrong value would be corrected the next time SetOffsetVacuumLimit() is called, or by a restart. Reported-By: Noah Misch, although this is not his preferred fix Discussion: 20151210140450.GA22278@alap3.anarazel.de Backpatch: 9.5, where the bug was introduced as part of 4f627f
2015-12-14 11:34:16 +01:00
prevOffsetStopLimit = MultiXactState->offsetStopLimit;
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
Assert(MultiXactState->finishedStartup);
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
LWLockRelease(MultiXactGenLock);
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
/*
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* Determine the offset of the oldest multixact. Normally, we can read
* the offset from the multixact itself, but there's an important special
* case: if there are no multixacts in existence at all, oldestMXact
* obviously can't point to one. It will instead point to the multixact
* ID that will be assigned the next time one is needed.
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
*/
if (oldestMultiXactId == nextMXact)
{
/*
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* When the next multixact gets created, it will be stored at the next
* offset.
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
*/
oldestOffset = nextOffset;
oldestOffsetKnown = true;
}
else
{
/*
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* Figure out where the oldest existing multixact's offsets are
* stored. Due to bugs in early release of PostgreSQL 9.3.X and 9.4.X,
* the supposedly-earliest multixact might not really exist. We are
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
* careful not to fail in that case.
*/
oldestOffsetKnown =
find_multixact_start(oldestMultiXactId, &oldestOffset);
if (oldestOffsetKnown)
ereport(DEBUG1,
Use errmsg_internal for debug messages An inconsistent set of debug-level messages was not using errmsg_internal(), thus uselessly exposing the messages to translation work. Fix those.
2021-02-17 11:24:46 +01:00
(errmsg_internal("oldest MultiXactId member is at offset %u",
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
oldestOffset)));
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
else
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
ereport(LOG,
(errmsg("MultiXact member wraparound protections are disabled because oldest checkpointed MultiXact %u does not exist on disk",
oldestMultiXactId)));
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
}
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
LWLockRelease(MultiXactTruncationLock);
Fix corner case in autovacuum-forcing logic for multixact wraparound. Since find_multixact_start() relies on SimpleLruDoesPhysicalPageExist(), and that function looks only at the on-disk state, it's possible for it to fail to find a page that exists in the in-memory SLRU that has not been written yet. If that happens, SetOffsetVacuumLimit() will erroneously decide to force emergency autovacuuming immediately. We should probably fix find_multixact_start() to consider the data cached in memory as well as on the on-disk state, but that's no excuse for SetOffsetVacuumLimit() to be stupid about the case where it can no longer read the value after having previously succeeded in doing so. Report by Andres Freund.
2015-06-19 17:28:30 +02:00
/*
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* If we can, compute limits (and install them MultiXactState) to prevent
* overrun of old data in the members SLRU area. We can only do so if the
* oldest offset is known though.
Fix corner case in autovacuum-forcing logic for multixact wraparound. Since find_multixact_start() relies on SimpleLruDoesPhysicalPageExist(), and that function looks only at the on-disk state, it's possible for it to fail to find a page that exists in the in-memory SLRU that has not been written yet. If that happens, SetOffsetVacuumLimit() will erroneously decide to force emergency autovacuuming immediately. We should probably fix find_multixact_start() to consider the data cached in memory as well as on the on-disk state, but that's no excuse for SetOffsetVacuumLimit() to be stupid about the case where it can no longer read the value after having previously succeeded in doing so. Report by Andres Freund.
2015-06-19 17:28:30 +02:00
*/
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
if (oldestOffsetKnown)
{
/* move back to start of the corresponding segment */
offsetStopLimit = oldestOffset - (oldestOffset %
(MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT));
/* always leave one segment before the wraparound point */
offsetStopLimit -= (MULTIXACT_MEMBERS_PER_PAGE * SLRU_PAGES_PER_SEGMENT);
Make logging about multixact wraparound protection less chatty. The original messaging design, introduced in commit 068cfadf9, seems too chatty now that some time has elapsed since the bug fix; most installations will be in good shape and don't really need a reminder about this on every postmaster start. Hence, arrange to suppress the "wraparound protections are now enabled" message during startup (specifically, during the TrimMultiXact() call). The message will still appear if protection becomes effective at some later point. Discussion: https://postgr.es/m/17211.1489189214@sss.pgh.pa.us
2017-03-14 17:47:46 +01:00
if (!prevOldestOffsetKnown && !is_startup)
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
ereport(LOG,
(errmsg("MultiXact member wraparound protections are now enabled")));
Make logging about multixact wraparound protection less chatty. The original messaging design, introduced in commit 068cfadf9, seems too chatty now that some time has elapsed since the bug fix; most installations will be in good shape and don't really need a reminder about this on every postmaster start. Hence, arrange to suppress the "wraparound protections are now enabled" message during startup (specifically, during the TrimMultiXact() call). The message will still appear if protection becomes effective at some later point. Discussion: https://postgr.es/m/17211.1489189214@sss.pgh.pa.us
2017-03-14 17:47:46 +01:00
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
ereport(DEBUG1,
Use errmsg_internal for debug messages An inconsistent set of debug-level messages was not using errmsg_internal(), thus uselessly exposing the messages to translation work. Fix those.
2021-02-17 11:24:46 +01:00
(errmsg_internal("MultiXact member stop limit is now %u based on MultiXact %u",
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
offsetStopLimit, oldestMultiXactId)));
}
else if (prevOldestOffsetKnown)
Fix corner case in autovacuum-forcing logic for multixact wraparound. Since find_multixact_start() relies on SimpleLruDoesPhysicalPageExist(), and that function looks only at the on-disk state, it's possible for it to fail to find a page that exists in the in-memory SLRU that has not been written yet. If that happens, SetOffsetVacuumLimit() will erroneously decide to force emergency autovacuuming immediately. We should probably fix find_multixact_start() to consider the data cached in memory as well as on the on-disk state, but that's no excuse for SetOffsetVacuumLimit() to be stupid about the case where it can no longer read the value after having previously succeeded in doing so. Report by Andres Freund.
2015-06-19 17:28:30 +02:00
{
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
/*
* If we failed to get the oldest offset this time, but we have a
Fix bug in SetOffsetVacuumLimit() triggered by find_multixact_start() failure. Previously, if find_multixact_start() failed, SetOffsetVacuumLimit() would install 0 into MultiXactState->offsetStopLimit if it previously succeeded. Luckily, there are no known cases where find_multixact_start() will return an error in 9.5 and above. But if it were to happen, for example due to filesystem permission issues, it'd be somewhat bad: GetNewMultiXactId() could continue allocating mxids even if close to a wraparound, or it could erroneously stop allocating mxids, even if no wraparound is looming. The wrong value would be corrected the next time SetOffsetVacuumLimit() is called, or by a restart. Reported-By: Noah Misch, although this is not his preferred fix Discussion: 20151210140450.GA22278@alap3.anarazel.de Backpatch: 9.5, where the bug was introduced as part of 4f627f
2015-12-14 11:34:16 +01:00
* value from a previous pass through this function, use the old
* values rather than automatically forcing an emergency autovacuum
* cycle again.
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
*/
Fix corner case in autovacuum-forcing logic for multixact wraparound. Since find_multixact_start() relies on SimpleLruDoesPhysicalPageExist(), and that function looks only at the on-disk state, it's possible for it to fail to find a page that exists in the in-memory SLRU that has not been written yet. If that happens, SetOffsetVacuumLimit() will erroneously decide to force emergency autovacuuming immediately. We should probably fix find_multixact_start() to consider the data cached in memory as well as on the on-disk state, but that's no excuse for SetOffsetVacuumLimit() to be stupid about the case where it can no longer read the value after having previously succeeded in doing so. Report by Andres Freund.
2015-06-19 17:28:30 +02:00
oldestOffset = prevOldestOffset;
oldestOffsetKnown = true;
Fix bug in SetOffsetVacuumLimit() triggered by find_multixact_start() failure. Previously, if find_multixact_start() failed, SetOffsetVacuumLimit() would install 0 into MultiXactState->offsetStopLimit if it previously succeeded. Luckily, there are no known cases where find_multixact_start() will return an error in 9.5 and above. But if it were to happen, for example due to filesystem permission issues, it'd be somewhat bad: GetNewMultiXactId() could continue allocating mxids even if close to a wraparound, or it could erroneously stop allocating mxids, even if no wraparound is looming. The wrong value would be corrected the next time SetOffsetVacuumLimit() is called, or by a restart. Reported-By: Noah Misch, although this is not his preferred fix Discussion: 20151210140450.GA22278@alap3.anarazel.de Backpatch: 9.5, where the bug was introduced as part of 4f627f
2015-12-14 11:34:16 +01:00
offsetStopLimit = prevOffsetStopLimit;
Fix corner case in autovacuum-forcing logic for multixact wraparound. Since find_multixact_start() relies on SimpleLruDoesPhysicalPageExist(), and that function looks only at the on-disk state, it's possible for it to fail to find a page that exists in the in-memory SLRU that has not been written yet. If that happens, SetOffsetVacuumLimit() will erroneously decide to force emergency autovacuuming immediately. We should probably fix find_multixact_start() to consider the data cached in memory as well as on the on-disk state, but that's no excuse for SetOffsetVacuumLimit() to be stupid about the case where it can no longer read the value after having previously succeeded in doing so. Report by Andres Freund.
2015-06-19 17:28:30 +02:00
}
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
/* Install the computed values */
LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
MultiXactState->oldestOffset = oldestOffset;
MultiXactState->oldestOffsetKnown = oldestOffsetKnown;
MultiXactState->offsetStopLimit = offsetStopLimit;
LWLockRelease(MultiXactGenLock);
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
/*
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* Do we need an emergency autovacuum? If we're not sure, assume yes.
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
*/
return !oldestOffsetKnown ||
(nextOffset - oldestOffset > MULTIXACT_MEMBER_SAFE_THRESHOLD);
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
}
/*
* Return whether adding "distance" to "start" would move past "boundary".
*
* We use this to determine whether the addition is "wrapping around" the
* boundary point, hence the name. The reason we don't want to use the regular
* 2^31-modulo arithmetic here is that we want to be able to use the whole of
Fix some typos Author: Justin Pryzby Discussion: https://postgr.es/m/20200408165653.GF2228@telsasoft.com
2020-04-27 07:59:36 +02:00
* the 2^32-1 space here, allowing for more multixacts than would fit
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* otherwise.
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
*/
static bool
MultiXactOffsetWouldWrap(MultiXactOffset boundary, MultiXactOffset start,
uint32 distance)
{
MultiXactOffset finish;
/*
* Note that offset number 0 is not used (see GetMultiXactIdMembers), so
* if the addition wraps around the UINT_MAX boundary, skip that value.
*/
finish = start + distance;
if (finish < start)
finish++;
/*-----------------------------------------------------------------------
* When the boundary is numerically greater than the starting point, any
* value numerically between the two is not wrapped:
*
* <----S----B---->
* [---) = F wrapped past B (and UINT_MAX)
* [---) = F not wrapped
* [----] = F wrapped past B
*
* When the boundary is numerically less than the starting point (i.e. the
* UINT_MAX wraparound occurs somewhere in between) then all values in
* between are wrapped:
*
* <----B----S---->
* [---) = F not wrapped past B (but wrapped past UINT_MAX)
* [---) = F wrapped past B (and UINT_MAX)
* [----] = F not wrapped
*-----------------------------------------------------------------------
*/
if (start < boundary)
return finish >= boundary || finish < start;
else
return finish >= boundary && finish < start;
}
/*
Code review for multixact bugfix Reword messages, rename a confusingly named function. Per Robert Haas.
2015-04-28 19:52:29 +02:00
* Find the starting offset of the given MultiXactId.
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
*
* Returns false if the file containing the multi does not exist on disk.
* Otherwise, returns true and sets *result to the starting member offset.
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
*
* This function does not prevent concurrent truncation, so if that's
* required, the caller has to protect against that.
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
*/
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
static bool
find_multixact_start(MultiXactId multi, MultiXactOffset *result)
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
{
MultiXactOffset offset;
int pageno;
int entryno;
int slotno;
MultiXactOffset *offptr;
Remove legacy multixact truncation support. In 9.5 and master there is no need to support legacy truncation. This is just committed separately to make it easier to backpatch the WAL logged multixact truncation to 9.3 and 9.4 if we later decide to do so. I bumped master's magic from 0xD086 to 0xD088 and 9.5's from 0xD085 to 0xD087 to avoid 9.5 reusing a value that has been in use on master while keeping the numbers increasing between major versions. Discussion: 20150621192409.GA4797@alap3.anarazel.de Backpatch: 9.5
2015-09-26 19:04:25 +02:00
Assert(MultiXactState->finishedStartup);
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
pageno = MultiXactIdToOffsetPage(multi);
entryno = MultiXactIdToOffsetEntry(multi);
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
/*
Defer flushing of SLRU files. Previously, we called fsync() after writing out individual pg_xact, pg_multixact and pg_commit_ts pages due to cache pressure, leading to regular I/O stalls in user backends and recovery. Collapse requests for the same file into a single system call as part of the next checkpoint, as we already did for relation files, using the infrastructure developed by commit 3eb77eba. This can cause a significant improvement to recovery performance, especially when it's otherwise CPU-bound. Hoist ProcessSyncRequests() up into CheckPointGuts() to make it clearer that it applies to all the SLRU mini-buffer-pools as well as the main buffer pool. Rearrange things so that data collected in CheckpointStats includes SLRU activity. Also remove the Shutdown{CLOG,CommitTS,SUBTRANS,MultiXact}() functions, because they were redundant after the shutdown checkpoint that immediately precedes them. (I'm not sure if they were ever needed, but they aren't now.) Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us> (parts) Tested-by: Jakub Wartak <Jakub.Wartak@tomtom.com> Discussion: https://postgr.es/m/CA+hUKGLJ=84YT+NvhkEEDAuUtVHMfQ9i-N7k_o50JmQ6Rpj_OQ@mail.gmail.com
2020-09-25 08:49:43 +02:00
* Write out dirty data, so PhysicalPageExists can work correctly.
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
*/
Defer flushing of SLRU files. Previously, we called fsync() after writing out individual pg_xact, pg_multixact and pg_commit_ts pages due to cache pressure, leading to regular I/O stalls in user backends and recovery. Collapse requests for the same file into a single system call as part of the next checkpoint, as we already did for relation files, using the infrastructure developed by commit 3eb77eba. This can cause a significant improvement to recovery performance, especially when it's otherwise CPU-bound. Hoist ProcessSyncRequests() up into CheckPointGuts() to make it clearer that it applies to all the SLRU mini-buffer-pools as well as the main buffer pool. Rearrange things so that data collected in CheckpointStats includes SLRU activity. Also remove the Shutdown{CLOG,CommitTS,SUBTRANS,MultiXact}() functions, because they were redundant after the shutdown checkpoint that immediately precedes them. (I'm not sure if they were ever needed, but they aren't now.) Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us> (parts) Tested-by: Jakub Wartak <Jakub.Wartak@tomtom.com> Discussion: https://postgr.es/m/CA+hUKGLJ=84YT+NvhkEEDAuUtVHMfQ9i-N7k_o50JmQ6Rpj_OQ@mail.gmail.com
2020-09-25 08:49:43 +02:00
SimpleLruWriteAll(MultiXactOffsetCtl, true);
SimpleLruWriteAll(MultiXactMemberCtl, true);
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
if (!SimpleLruDoesPhysicalPageExist(MultiXactOffsetCtl, pageno))
return false;
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
/* lock is acquired by SimpleLruReadPage_ReadOnly */
slotno = SimpleLruReadPage_ReadOnly(MultiXactOffsetCtl, pageno, multi);
offptr = (MultiXactOffset *) MultiXactOffsetCtl->shared->page_buffer[slotno];
offptr += entryno;
offset = *offptr;
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactOffsetSLRULock);
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
*result = offset;
return true;
Protect against multixact members wraparound Multixact member files are subject to early wraparound overflow and removal: if the average multixact size is above a certain threshold (see note below) the protections against offset overflow are not enough: during multixact truncation at checkpoint time, some pg_multixact/members files would be removed because the server considers them to be old and not needed anymore. This leads to loss of files that are critical to interpret existing tuples's Xmax values. To protect against this, since we don't have enough info in pg_control and we can't modify it in old branches, we maintain shared memory state about the oldest value that we need to keep; we use this during new multixact creation to abort if an old still-needed file would get overwritten. This value is kept up to date by checkpoints, which makes it not completely accurate but should be good enough. We start emitting warnings sometime earlier, so that the eventual multixact-shutdown doesn't take DBAs completely by surprise (more precisely: once 20 members SLRU segments are remaining before shutdown.) On troublesome average multixact size: The threshold size depends on the multixact freeze parameters. The oldest age is related to the greater of multixact_freeze_table_age and multixact_freeze_min_age: anything older than that should be removed promptly by autovacuum. If autovacuum is keeping up with multixact freezing, the troublesome multixact average size is (2^32-1) / Max(freeze table age, freeze min age) or around 28 members per multixact. Having an average multixact size larger than that will eventually cause new multixact data to overwrite the data area for older multixacts. (If autovacuum is not able to keep up, or there are errors in vacuuming, the actual maximum is multixact_freeeze_max_age instead, at which point multixact generation is stopped completely. The default value for this limit is 400 million, which means that the multixact size that would cause trouble is about 10 members). Initial bug report by Timothy Garnett, bug #12990 Backpatch to 9.3, where the problem was introduced. Authors: Álvaro Herrera, Thomas Munro Reviews: Thomas Munro, Amit Kapila, Robert Haas, Kevin Grittner
2015-04-28 16:32:53 +02:00
}
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
/*
* Determine how many multixacts, and how many multixact members, currently
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
* exist. Return false if unable to determine.
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
*/
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
static bool
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
ReadMultiXactCounts(uint32 *multixacts, MultiXactOffset *members)
{
MultiXactOffset nextOffset;
MultiXactOffset oldestOffset;
MultiXactId oldestMultiXactId;
MultiXactId nextMultiXactId;
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
bool oldestOffsetKnown;
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
LWLockAcquire(MultiXactGenLock, LW_SHARED);
nextOffset = MultiXactState->nextOffset;
oldestMultiXactId = MultiXactState->oldestMultiXactId;
nextMultiXactId = MultiXactState->nextMXact;
Even when autovacuum=off, force it for members as we do in other cases. Thomas Munro, with some adjustments by me.
2015-05-11 16:51:14 +02:00
oldestOffset = MultiXactState->oldestOffset;
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
oldestOffsetKnown = MultiXactState->oldestOffsetKnown;
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
LWLockRelease(MultiXactGenLock);
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
if (!oldestOffsetKnown)
return false;
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
*members = nextOffset - oldestOffset;
*multixacts = nextMultiXactId - oldestMultiXactId;
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
return true;
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
}
/*
* Multixact members can be removed once the multixacts that refer to them
Fix inconsistencies in the code This addresses a couple of issues in the code: - Typos and inconsistencies in comments and function declarations. - Removal of unreferenced function declarations. - Removal of unnecessary compile flags. - A cleanup error in regressplans.sh. Author: Alexander Lakhin Discussion: https://postgr.es/m/0c991fdf-2670-1997-c027-772a420c4604@gmail.com
2019-07-08 06:15:09 +02:00
* are older than every datminmxid. autovacuum_multixact_freeze_max_age and
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
* vacuum_multixact_freeze_table_age work together to make sure we never have
* too many multixacts; we hope that, at least under normal circumstances,
* this will also be sufficient to keep us from using too many offsets.
* However, if the average multixact has many members, we might exhaust the
* members space while still using few enough members that these limits fail
Update obsolete multixact.c comments. Commit 4f627f89 switched SLRU truncation for multixacts back to being a task performed during VACUUM, but missed some comments that continued to reference truncation happening as part of checkpointing. Update those comments now. Also update comments that became obsolete when commit c3ffa731 changed the way that vacuum_multixact_freeze_min_age is applied by VACUUM as it computes its MultiXactCutoff cutoff (which is used by VACUUM to decide what to freeze). Explain the same issues by referencing how OldestMxact is the latest valid value that relminmxid can ever be advanced to at the end of a VACUUM (following the work in commit 0b018fab).
2023-01-04 01:54:35 +01:00
* to trigger relminmxid advancement by VACUUM. At that point, we'd have no
* choice but to start failing multixact-creating operations with an error.
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
*
* To prevent that, if more than a threshold portion of the members space is
* used, we effectively reduce autovacuum_multixact_freeze_max_age and
* to a value just less than the number of multixacts in use. We hope that
* this will quickly trigger autovacuuming on the table or tables with the
* oldest relminmxid, thus allowing datminmxid values to advance and removing
Fix whitespace
2015-05-08 20:45:53 +02:00
* some members.
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
*
* As the fraction of the member space currently in use grows, we become
* more aggressive in clamping this value. That not only causes autovacuum
* to ramp up, but also makes any manual vacuums the user issues more
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
* aggressive. This happens because vacuum_get_cutoffs() will clamp the
* freeze table and the minimum freeze age cutoffs based on the effective
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
* autovacuum_multixact_freeze_max_age this function returns. In the worst
* case, we'll claim the freeze_max_age to zero, and every vacuum of any
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
* table will freeze every multixact.
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
*/
int
MultiXactMemberFreezeThreshold(void)
{
MultiXactOffset members;
uint32 multixacts;
uint32 victim_multixacts;
double fraction;
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
/* If we can't determine member space utilization, assume the worst. */
if (!ReadMultiXactCounts(&multixacts, &members))
return 0;
Teach autovacuum about multixact member wraparound. The logic introduced in commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and repaired in commits 669c7d20e6374850593cb430d332e11a3992bbcf and 7be47c56af3d3013955c91c2877c08f2a0e3e6a2 helps to ensure that we don't overwrite old multixact member information while it is still needed, but a user who creates many large multixacts can still exhaust the member space (and thus start getting errors) while autovacuum stands idly by. To fix this, progressively ramp down the effective value (but not the actual contents) of autovacuum_multixact_freeze_max_age as member space utilization increases. This makes autovacuum more aggressive and also reduces the threshold for a manual VACUUM to perform a full-table scan. This patch leaves unsolved the problem of ensuring that emergency autovacuums are triggered even when autovacuum=off. We'll need to fix that via a separate patch. Thomas Munro and Robert Haas
2015-05-08 18:09:14 +02:00
/* If member space utilization is low, no special action is required. */
if (members <= MULTIXACT_MEMBER_SAFE_THRESHOLD)
return autovacuum_multixact_freeze_max_age;
/*
* Compute a target for relminmxid advancement. The number of multixacts
* we try to eliminate from the system is based on how far we are past
* MULTIXACT_MEMBER_SAFE_THRESHOLD.
*/
fraction = (double) (members - MULTIXACT_MEMBER_SAFE_THRESHOLD) /
(MULTIXACT_MEMBER_DANGER_THRESHOLD - MULTIXACT_MEMBER_SAFE_THRESHOLD);
victim_multixacts = multixacts * fraction;
/* fraction could be > 1.0, but lowest possible freeze age is zero */
if (victim_multixacts > multixacts)
return 0;
return multixacts - victim_multixacts;
}
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
typedef struct mxtruncinfo
{
int earliestExistingPage;
} mxtruncinfo;
/*
* SlruScanDirectory callback
* This callback determines the earliest existing page number.
*/
static bool
SlruScanDirCbFindEarliest(SlruCtl ctl, char *filename, int segpage, void *data)
{
mxtruncinfo *trunc = (mxtruncinfo *) data;
if (trunc->earliestExistingPage == -1 ||
ctl->PagePrecedes(segpage, trunc->earliestExistingPage))
{
trunc->earliestExistingPage = segpage;
}
return false; /* keep going */
}
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
/*
* Delete members segments [oldest, newOldest)
*
* The members SLRU can, in contrast to the offsets one, be filled to almost
* the full range at once. This means SimpleLruTruncate() can't trivially be
* used - instead the to-be-deleted range is computed using the offsets
* SLRU. C.f. TruncateMultiXact().
*/
static void
PerformMembersTruncation(MultiXactOffset oldestOffset, MultiXactOffset newOldestOffset)
{
const int maxsegment = MXOffsetToMemberSegment(MaxMultiXactOffset);
int startsegment = MXOffsetToMemberSegment(oldestOffset);
int endsegment = MXOffsetToMemberSegment(newOldestOffset);
int segment = startsegment;
/*
* Delete all the segments but the last one. The last segment can still
* contain, possibly partially, valid data.
*/
while (segment != endsegment)
{
elog(DEBUG2, "truncating multixact members segment %x", segment);
SlruDeleteSegment(MultiXactMemberCtl, segment);
/* move to next segment, handling wraparound correctly */
if (segment == maxsegment)
segment = 0;
else
segment += 1;
}
}
/*
* Delete offsets segments [oldest, newOldest)
*/
static void
PerformOffsetsTruncation(MultiXactId oldestMulti, MultiXactId newOldestMulti)
{
/*
* We step back one multixact to avoid passing a cutoff page that hasn't
* been created yet in the rare case that oldestMulti would be the first
* item on a page and oldestMulti == nextMulti. In that case, if we
* didn't subtract one, we'd trigger SimpleLruTruncate's wraparound
* detection.
*/
SimpleLruTruncate(MultiXactOffsetCtl,
MultiXactIdToOffsetPage(PreviousMultiXactId(newOldestMulti)));
}
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 all MultiXactOffset and MultiXactMember segments before the oldest
* ones still of interest.
*
Remove legacy multixact truncation support. In 9.5 and master there is no need to support legacy truncation. This is just committed separately to make it easier to backpatch the WAL logged multixact truncation to 9.3 and 9.4 if we later decide to do so. I bumped master's magic from 0xD086 to 0xD088 and 9.5's from 0xD085 to 0xD087 to avoid 9.5 reusing a value that has been in use on master while keeping the numbers increasing between major versions. Discussion: 20150621192409.GA4797@alap3.anarazel.de Backpatch: 9.5
2015-09-26 19:04:25 +02:00
* This is only called on a primary as part of vacuum (via
* vac_truncate_clog()). During recovery truncation is done by replaying
* truncation WAL records logged here.
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
*
* newOldestMulti is the oldest currently required multixact, newOldestMultiDB
* is one of the databases preventing newOldestMulti from increasing.
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
*/
void
Remove legacy multixact truncation support. In 9.5 and master there is no need to support legacy truncation. This is just committed separately to make it easier to backpatch the WAL logged multixact truncation to 9.3 and 9.4 if we later decide to do so. I bumped master's magic from 0xD086 to 0xD088 and 9.5's from 0xD085 to 0xD087 to avoid 9.5 reusing a value that has been in use on master while keeping the numbers increasing between major versions. Discussion: 20150621192409.GA4797@alap3.anarazel.de Backpatch: 9.5
2015-09-26 19:04:25 +02:00
TruncateMultiXact(MultiXactId newOldestMulti, Oid newOldestMultiDB)
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
{
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
MultiXactId oldestMulti;
MultiXactId nextMulti;
MultiXactOffset newOldestOffset;
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
MultiXactOffset oldestOffset;
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
MultiXactOffset nextOffset;
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
mxtruncinfo trunc;
MultiXactId earliest;
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
Remove legacy multixact truncation support. In 9.5 and master there is no need to support legacy truncation. This is just committed separately to make it easier to backpatch the WAL logged multixact truncation to 9.3 and 9.4 if we later decide to do so. I bumped master's magic from 0xD086 to 0xD088 and 9.5's from 0xD085 to 0xD087 to avoid 9.5 reusing a value that has been in use on master while keeping the numbers increasing between major versions. Discussion: 20150621192409.GA4797@alap3.anarazel.de Backpatch: 9.5
2015-09-26 19:04:25 +02:00
Assert(!RecoveryInProgress());
Assert(MultiXactState->finishedStartup);
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
/*
* We can only allow one truncation to happen at once. Otherwise parts of
* members might vanish while we're doing lookups or similar. There's no
* need to have an interlock with creating new multis or such, since those
* are constrained by the limits (which only grow, never shrink).
*/
LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
Have multixact be truncated by checkpoint, not vacuum Instead of truncating pg_multixact at vacuum time, do it only at checkpoint time. The reason for doing it this way is twofold: first, we want it to delete only segments that we're certain will not be required if there's a crash immediately after the removal; and second, we want to do it relatively often so that older files are not left behind if there's an untimely crash. Per my proposal in http://www.postgresql.org/message-id/20140626044519.GJ7340@eldon.alvh.no-ip.org we now execute the truncation in the checkpointer process rather than as part of vacuum. Vacuum is in only charge of maintaining in shared memory the value to which it's possible to truncate the files; that value is stored as part of checkpoints also, and so upon recovery we can reuse the same value to re-execute truncate and reset the oldest-value-still-safe-to-use to one known to remain after truncation. Per bug reported by Jeff Janes in the course of his tests involving bug #8673. While at it, update some comments that hadn't been updated since multixacts were changed. Backpatch to 9.3, where persistency of pg_multixact files was introduced by commit 0ac5ad5134f2.
2014-06-27 20:43:53 +02:00
LWLockAcquire(MultiXactGenLock, LW_SHARED);
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
nextMulti = MultiXactState->nextMXact;
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
nextOffset = MultiXactState->nextOffset;
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
oldestMulti = MultiXactState->oldestMultiXactId;
Have multixact be truncated by checkpoint, not vacuum Instead of truncating pg_multixact at vacuum time, do it only at checkpoint time. The reason for doing it this way is twofold: first, we want it to delete only segments that we're certain will not be required if there's a crash immediately after the removal; and second, we want to do it relatively often so that older files are not left behind if there's an untimely crash. Per my proposal in http://www.postgresql.org/message-id/20140626044519.GJ7340@eldon.alvh.no-ip.org we now execute the truncation in the checkpointer process rather than as part of vacuum. Vacuum is in only charge of maintaining in shared memory the value to which it's possible to truncate the files; that value is stored as part of checkpoints also, and so upon recovery we can reuse the same value to re-execute truncate and reset the oldest-value-still-safe-to-use to one known to remain after truncation. Per bug reported by Jeff Janes in the course of his tests involving bug #8673. While at it, update some comments that hadn't been updated since multixacts were changed. Backpatch to 9.3, where persistency of pg_multixact files was introduced by commit 0ac5ad5134f2.
2014-06-27 20:43:53 +02:00
LWLockRelease(MultiXactGenLock);
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
Assert(MultiXactIdIsValid(oldestMulti));
/*
* Make sure to only attempt truncation if there's values to truncate
* away. In normal processing values shouldn't go backwards, but there's
* some corner cases (due to bugs) where that's possible.
*/
if (MultiXactIdPrecedesOrEquals(newOldestMulti, oldestMulti))
{
LWLockRelease(MultiXactTruncationLock);
return;
}
Have multixact be truncated by checkpoint, not vacuum Instead of truncating pg_multixact at vacuum time, do it only at checkpoint time. The reason for doing it this way is twofold: first, we want it to delete only segments that we're certain will not be required if there's a crash immediately after the removal; and second, we want to do it relatively often so that older files are not left behind if there's an untimely crash. Per my proposal in http://www.postgresql.org/message-id/20140626044519.GJ7340@eldon.alvh.no-ip.org we now execute the truncation in the checkpointer process rather than as part of vacuum. Vacuum is in only charge of maintaining in shared memory the value to which it's possible to truncate the files; that value is stored as part of checkpoints also, and so upon recovery we can reuse the same value to re-execute truncate and reset the oldest-value-still-safe-to-use to one known to remain after truncation. Per bug reported by Jeff Janes in the course of his tests involving bug #8673. While at it, update some comments that hadn't been updated since multixacts were changed. Backpatch to 9.3, where persistency of pg_multixact files was introduced by commit 0ac5ad5134f2.
2014-06-27 20:43:53 +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
* Note we can't just plow ahead with the truncation; it's possible that
* there are no segments to truncate, which is a problem because we are
* going to attempt to read the offsets page to determine where to
* truncate the members SLRU. So we first scan the directory to determine
* the earliest offsets page number that we can read without error.
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
*
Prevent excess SimpleLruTruncate() deletion. Every core SLRU wraps around. With the exception of pg_notify, the wrap point can fall in the middle of a page. Account for this in the PagePrecedes callback specification and in SimpleLruTruncate()'s use of said callback. Update each callback implementation to fit the new specification. This changes SerialPagePrecedesLogically() from the style of asyncQueuePagePrecedes() to the style of CLOGPagePrecedes(). (Whereas pg_clog and pg_serial share a key space, pg_serial is nothing like pg_notify.) The bug fixed here has the same symptoms and user followup steps as 592a589a04bd456410b853d86bd05faa9432cbbb. Back-patch to 9.5 (all supported versions). Reviewed by Andrey Borodin and (in earlier versions) by Tom Lane. Discussion: https://postgr.es/m/20190202083822.GC32531@gust.leadboat.com
2021-01-16 21:21:35 +01:00
* When nextMXact is less than one segment away from multiWrapLimit,
* SlruScanDirCbFindEarliest can find some early segment other than the
* actual earliest. (MultiXactOffsetPagePrecedes(EARLIEST, LATEST)
* returns false, because not all pairs of entries have the same answer.)
* That can also arise when an earlier truncation attempt failed unlink()
* or returned early from this function. The only consequence is
* returning early, which wastes space that we could have liberated.
*
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* NB: It's also possible that the page that oldestMulti is on has already
* been truncated away, and we crashed before updating oldestMulti.
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
trunc.earliestExistingPage = -1;
SlruScanDirectory(MultiXactOffsetCtl, SlruScanDirCbFindEarliest, &trunc);
earliest = trunc.earliestExistingPage * MULTIXACT_OFFSETS_PER_PAGE;
Have multixact be truncated by checkpoint, not vacuum Instead of truncating pg_multixact at vacuum time, do it only at checkpoint time. The reason for doing it this way is twofold: first, we want it to delete only segments that we're certain will not be required if there's a crash immediately after the removal; and second, we want to do it relatively often so that older files are not left behind if there's an untimely crash. Per my proposal in http://www.postgresql.org/message-id/20140626044519.GJ7340@eldon.alvh.no-ip.org we now execute the truncation in the checkpointer process rather than as part of vacuum. Vacuum is in only charge of maintaining in shared memory the value to which it's possible to truncate the files; that value is stored as part of checkpoints also, and so upon recovery we can reuse the same value to re-execute truncate and reset the oldest-value-still-safe-to-use to one known to remain after truncation. Per bug reported by Jeff Janes in the course of his tests involving bug #8673. While at it, update some comments that hadn't been updated since multixacts were changed. Backpatch to 9.3, where persistency of pg_multixact files was introduced by commit 0ac5ad5134f2.
2014-06-27 20:43:53 +02:00
if (earliest < FirstMultiXactId)
earliest = FirstMultiXactId;
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
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
/* If there's nothing to remove, we can bail out early. */
if (MultiXactIdPrecedes(oldestMulti, earliest))
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
{
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
LWLockRelease(MultiXactTruncationLock);
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
return;
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +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
* First, compute the safe truncation point for MultiXactMember. This is
Have multixact be truncated by checkpoint, not vacuum Instead of truncating pg_multixact at vacuum time, do it only at checkpoint time. The reason for doing it this way is twofold: first, we want it to delete only segments that we're certain will not be required if there's a crash immediately after the removal; and second, we want to do it relatively often so that older files are not left behind if there's an untimely crash. Per my proposal in http://www.postgresql.org/message-id/20140626044519.GJ7340@eldon.alvh.no-ip.org we now execute the truncation in the checkpointer process rather than as part of vacuum. Vacuum is in only charge of maintaining in shared memory the value to which it's possible to truncate the files; that value is stored as part of checkpoints also, and so upon recovery we can reuse the same value to re-execute truncate and reset the oldest-value-still-safe-to-use to one known to remain after truncation. Per bug reported by Jeff Janes in the course of his tests involving bug #8673. While at it, update some comments that hadn't been updated since multixacts were changed. Backpatch to 9.3, where persistency of pg_multixact files was introduced by commit 0ac5ad5134f2.
2014-06-27 20:43:53 +02:00
* the starting offset of the oldest multixact.
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
*
* Hopefully, find_multixact_start will always work here, because we've
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* already checked that it doesn't precede the earliest MultiXact on disk.
* But if it fails, don't truncate anything, and log a message.
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
*/
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
if (oldestMulti == nextMulti)
{
/* there are NO MultiXacts */
oldestOffset = nextOffset;
}
else if (!find_multixact_start(oldestMulti, &oldestOffset))
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
{
ereport(LOG,
(errmsg("oldest MultiXact %u not found, earliest MultiXact %u, skipping truncation",
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
oldestMulti, earliest)));
LWLockRelease(MultiXactTruncationLock);
Cope with possible failure of the oldest MultiXact to exist. Recent commits, mainly b69bf30b9bfacafc733a9ba77c9587cf54d06c0c and 53bb309d2d5a9432d2602c93ed18e58bd2924e15, introduced mechanisms to protect against wraparound of the MultiXact member space: the number of multixacts that can exist at one time is limited to 2^32, but the total number of members in those multixacts is also limited to 2^32, and older code did not take care to enforce the second limit, potentially allowing old data to be overwritten while it was still needed. Unfortunately, these new mechanisms failed to account for the fact that the code paths in which they run might be executed during recovery or while the cluster was in an inconsistent state. Also, they failed to account for the fact that users who used pg_upgrade to upgrade a PostgreSQL version between 9.3.0 and 9.3.4 might have might oldestMultiXid = 1 in the control file despite the true value being larger. To fix these problems, first, avoid unnecessarily examining the mmembers of MultiXacts when the cluster is not known to be consistent. TruncateMultiXact has done this for a long time, and this patch does not fix that. But the new calls used to prevent member wraparound are not needed until we reach normal running, so avoid calling them earlier. (SetMultiXactIdLimit is actually called before InRecovery is set, so we can't rely on that; we invent our own multixact-specific flag instead.) Second, make failure to look up the members of a MultiXact a non-fatal error. Instead, if we're unable to determine the member offset at which wraparound would occur, postpone arming the member wraparound defenses until we are able to do so. If we're unable to determine the member offset that should force autovacuum, force it continuously until we are able to do so. If we're unable to deterine the member offset at which we should truncate the members SLRU, log a message and skip truncation. An important consequence of these changes is that anyone who does have a bogus oldestMultiXid = 1 value in pg_control will experience immediate emergency autovacuuming when upgrading to a release that contains this fix. The release notes should highlight this fact. If a user has no pg_multixact/offsets/0000 file, but has oldestMultiXid = 1 in the control file, they may wish to vacuum any tables with relminmxid = 1 prior to upgrading in order to avoid an immediate emergency autovacuum after the upgrade. This must be done with a PostgreSQL version 9.3.5 or newer and with vacuum_multixact_freeze_min_age and vacuum_multixact_freeze_table_age set to 0. This patch also adds an additional log message at each database server startup, indicating either that protections against member wraparound have been engaged, or that they have not. In the latter case, once autovacuum has advanced oldestMultiXid to a sane value, the message indicating that the guards have been engaged will appear at the next checkpoint. A few additional messages have also been added at the DEBUG1 level so that the correct operation of this code can be properly audited. Along the way, this patch fixes another, related bug in TruncateMultiXact that has existed since PostgreSQL 9.3.0: when no MultiXacts exist at all, the truncation code looks up NextMultiXactId, which doesn't exist yet. This can lead to TruncateMultiXact removing every file in pg_multixact/offsets instead of keeping one around, as it should. This in turn will cause the database server to refuse to start afterwards. Patch by me. Review by Álvaro Herrera, Andres Freund, Noah Misch, and Thomas Munro.
2015-06-05 14:34:52 +02:00
return;
}
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
Handle wraparound during truncation in multixact/members In pg_multixact/members, relying on modulo-2^32 arithmetic for wraparound handling doesn't work all that well. Because we don't explicitely track wraparound of the allocation counter for members, it is possible that the "live" area exceeds 2^31 entries; trying to remove SLRU segments that are "old" according to the original logic might lead to removal of segments still in use. To fix, have the truncation routine use a tailored SlruScanDirectory callback that keeps track of the live area in actual use; that way, when the live range exceeds 2^31 entries, the oldest segments still live will not get removed untimely. This new SlruScanDir callback needs to take care not to remove segments that are "in the future": if new SLRU segments appear while the truncation is ongoing, make sure we don't remove them. This requires examination of shared memory state to recheck for false positives, but testing suggests that this doesn't cause a problem. The original coding didn't suffer from this pitfall because segments created when truncation is running are never considered to be removable. Per Andres Freund's investigation of bug #8673 reported by Serge Negodyuck.
2014-01-02 22:16:54 +01:00
/*
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* Secondly compute up to where to truncate. Lookup the corresponding
* member offset for newOldestMulti for that.
Handle wraparound during truncation in multixact/members In pg_multixact/members, relying on modulo-2^32 arithmetic for wraparound handling doesn't work all that well. Because we don't explicitely track wraparound of the allocation counter for members, it is possible that the "live" area exceeds 2^31 entries; trying to remove SLRU segments that are "old" according to the original logic might lead to removal of segments still in use. To fix, have the truncation routine use a tailored SlruScanDirectory callback that keeps track of the live area in actual use; that way, when the live range exceeds 2^31 entries, the oldest segments still live will not get removed untimely. This new SlruScanDir callback needs to take care not to remove segments that are "in the future": if new SLRU segments appear while the truncation is ongoing, make sure we don't remove them. This requires examination of shared memory state to recheck for false positives, but testing suggests that this doesn't cause a problem. The original coding didn't suffer from this pitfall because segments created when truncation is running are never considered to be removable. Per Andres Freund's investigation of bug #8673 reported by Serge Negodyuck.
2014-01-02 22:16:54 +01:00
*/
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
if (newOldestMulti == nextMulti)
{
/* there are NO MultiXacts */
newOldestOffset = nextOffset;
}
else if (!find_multixact_start(newOldestMulti, &newOldestOffset))
{
ereport(LOG,
(errmsg("cannot truncate up to MultiXact %u because it does not exist on disk, skipping truncation",
newOldestMulti)));
LWLockRelease(MultiXactTruncationLock);
return;
}
Handle wraparound during truncation in multixact/members In pg_multixact/members, relying on modulo-2^32 arithmetic for wraparound handling doesn't work all that well. Because we don't explicitely track wraparound of the allocation counter for members, it is possible that the "live" area exceeds 2^31 entries; trying to remove SLRU segments that are "old" according to the original logic might lead to removal of segments still in use. To fix, have the truncation routine use a tailored SlruScanDirectory callback that keeps track of the live area in actual use; that way, when the live range exceeds 2^31 entries, the oldest segments still live will not get removed untimely. This new SlruScanDir callback needs to take care not to remove segments that are "in the future": if new SLRU segments appear while the truncation is ongoing, make sure we don't remove them. This requires examination of shared memory state to recheck for false positives, but testing suggests that this doesn't cause a problem. The original coding didn't suffer from this pitfall because segments created when truncation is running are never considered to be removable. Per Andres Freund's investigation of bug #8673 reported by Serge Negodyuck.
2014-01-02 22:16:54 +01:00
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
elog(DEBUG1, "performing multixact truncation: "
"offsets [%u, %u), offsets segments [%x, %x), "
"members [%u, %u), members segments [%x, %x)",
oldestMulti, newOldestMulti,
MultiXactIdToOffsetSegment(oldestMulti),
MultiXactIdToOffsetSegment(newOldestMulti),
oldestOffset, newOldestOffset,
MXOffsetToMemberSegment(oldestOffset),
MXOffsetToMemberSegment(newOldestOffset));
Have multixact be truncated by checkpoint, not vacuum Instead of truncating pg_multixact at vacuum time, do it only at checkpoint time. The reason for doing it this way is twofold: first, we want it to delete only segments that we're certain will not be required if there's a crash immediately after the removal; and second, we want to do it relatively often so that older files are not left behind if there's an untimely crash. Per my proposal in http://www.postgresql.org/message-id/20140626044519.GJ7340@eldon.alvh.no-ip.org we now execute the truncation in the checkpointer process rather than as part of vacuum. Vacuum is in only charge of maintaining in shared memory the value to which it's possible to truncate the files; that value is stored as part of checkpoints also, and so upon recovery we can reuse the same value to re-execute truncate and reset the oldest-value-still-safe-to-use to one known to remain after truncation. Per bug reported by Jeff Janes in the course of his tests involving bug #8673. While at it, update some comments that hadn't been updated since multixacts were changed. Backpatch to 9.3, where persistency of pg_multixact files was introduced by commit 0ac5ad5134f2.
2014-06-27 20:43:53 +02:00
Fix off-by-one error in calculating subtrans/multixact truncation point. If there were no subtransactions (or multixacts) active, we would calculate the oldestxid == next xid. That's correct, but if next XID happens to be on the next pg_subtrans (pg_multixact) page, the page does not exist yet, and SimpleLruTruncate will produce an "apparent wraparound" warning. The warning is harmless in this case, but looks very alarming to users. Backpatch to all supported versions. Patch and analysis by Thomas Munro.
2015-07-23 00:29:59 +02:00
/*
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* Do truncation, and the WAL logging of the truncation, in a critical
* section. That way offsets/members cannot get out of sync anymore, i.e.
* once consistent the newOldestMulti will always exist in members, even
* if we crashed in the wrong moment.
Fix off-by-one error in calculating subtrans/multixact truncation point. If there were no subtransactions (or multixacts) active, we would calculate the oldestxid == next xid. That's correct, but if next XID happens to be on the next pg_subtrans (pg_multixact) page, the page does not exist yet, and SimpleLruTruncate will produce an "apparent wraparound" warning. The warning is harmless in this case, but looks very alarming to users. Backpatch to all supported versions. Patch and analysis by Thomas Munro.
2015-07-23 00:29:59 +02:00
*/
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
START_CRIT_SECTION();
Fix whitespace
2015-05-17 02:43:32 +02:00
Advance the stop point for multixact offset creation only at checkpoint. Commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c advanced the stop point at vacuum time, but this has subsequently been shown to be unsafe as a result of analysis by myself and Thomas Munro and testing by Thomas Munro. The crux of the problem is that the SLRU deletion logic may get confused about what to remove if, at exactly the right time during the checkpoint process, the head of the SLRU crosses what used to be the tail. This patch, by me, fixes the problem by advancing the stop point only following a checkpoint. This has the additional advantage of making the removal logic work during recovery more like the way it works during normal running, which is probably good. At least one of the calls to DetermineSafeOldestOffset which this patch removes was already dead, because MultiXactAdvanceOldest is called only during recovery and DetermineSafeOldestOffset was set up to do nothing during recovery. That, however, is inconsistent with the principle that recovery and normal running should work similarly, and was confusing to boot. Along the way, fix some comments that previous patches in this area neglected to update. It's not clear to me whether there's any concrete basis for the decision to use only half of the multixact ID space, but it's neither necessary nor sufficient to prevent multixact member wraparound, so the comments should not say otherwise.
2015-05-11 04:21:20 +02:00
/*
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
* Prevent checkpoints from being scheduled concurrently. This is critical
* because otherwise a truncation record might not be replayed after a
* crash/basebackup, even though the state of the data directory would
Remove legacy multixact truncation support. In 9.5 and master there is no need to support legacy truncation. This is just committed separately to make it easier to backpatch the WAL logged multixact truncation to 9.3 and 9.4 if we later decide to do so. I bumped master's magic from 0xD086 to 0xD088 and 9.5's from 0xD085 to 0xD087 to avoid 9.5 reusing a value that has been in use on master while keeping the numbers increasing between major versions. Discussion: 20150621192409.GA4797@alap3.anarazel.de Backpatch: 9.5
2015-09-26 19:04:25 +02:00
* require it.
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
*/
Rename delayChkpt to delayChkptFlags. Before commit 412ad7a55639516f284cd0ef9757d6ae5c7abd43, delayChkpt was a Boolean. Now it's an integer. Extensions using it need to be appropriately updated, so let's rename the field to make sure that a hard compilation failure occurs. Replacing delayChkpt with delayChkptFlags made a few comments extend past 80 characters, so I reflowed them and changed some wording very slightly. The back-branches will need a different change to restore compatibility with existing minor releases; this is just for master. Per suggestion from Tom Lane. Discussion: http://postgr.es/m/a7880f4d-1d74-582a-ada7-dad168d046d1@enterprisedb.com
2022-04-08 17:44:17 +02:00
Assert((MyProc->delayChkptFlags & DELAY_CHKPT_START) == 0);
MyProc->delayChkptFlags |= DELAY_CHKPT_START;
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
/* WAL log truncation */
Remove legacy multixact truncation support. In 9.5 and master there is no need to support legacy truncation. This is just committed separately to make it easier to backpatch the WAL logged multixact truncation to 9.3 and 9.4 if we later decide to do so. I bumped master's magic from 0xD086 to 0xD088 and 9.5's from 0xD085 to 0xD087 to avoid 9.5 reusing a value that has been in use on master while keeping the numbers increasing between major versions. Discussion: 20150621192409.GA4797@alap3.anarazel.de Backpatch: 9.5
2015-09-26 19:04:25 +02:00
WriteMTruncateXlogRec(newOldestMultiDB,
oldestMulti, newOldestMulti,
oldestOffset, newOldestOffset);
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
/*
* Update in-memory limits before performing the truncation, while inside
* the critical section: Have to do it before truncation, to prevent
* concurrent lookups of those values. Has to be inside the critical
* section as otherwise a future call to this function would error out,
* while looking up the oldest member in offsets, if our caller crashes
* before updating the limits.
Advance the stop point for multixact offset creation only at checkpoint. Commit b69bf30b9bfacafc733a9ba77c9587cf54d06c0c advanced the stop point at vacuum time, but this has subsequently been shown to be unsafe as a result of analysis by myself and Thomas Munro and testing by Thomas Munro. The crux of the problem is that the SLRU deletion logic may get confused about what to remove if, at exactly the right time during the checkpoint process, the head of the SLRU crosses what used to be the tail. This patch, by me, fixes the problem by advancing the stop point only following a checkpoint. This has the additional advantage of making the removal logic work during recovery more like the way it works during normal running, which is probably good. At least one of the calls to DetermineSafeOldestOffset which this patch removes was already dead, because MultiXactAdvanceOldest is called only during recovery and DetermineSafeOldestOffset was set up to do nothing during recovery. That, however, is inconsistent with the principle that recovery and normal running should work similarly, and was confusing to boot. Along the way, fix some comments that previous patches in this area neglected to update. It's not clear to me whether there's any concrete basis for the decision to use only half of the multixact ID space, but it's neither necessary nor sufficient to prevent multixact member wraparound, so the comments should not say otherwise.
2015-05-11 04:21:20 +02:00
*/
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
LWLockAcquire(MultiXactGenLock, LW_EXCLUSIVE);
MultiXactState->oldestMultiXactId = newOldestMulti;
MultiXactState->oldestMultiXactDB = newOldestMultiDB;
LWLockRelease(MultiXactGenLock);
/* First truncate members */
PerformMembersTruncation(oldestOffset, newOldestOffset);
/* Then offsets */
PerformOffsetsTruncation(oldestMulti, newOldestMulti);
Rename delayChkpt to delayChkptFlags. Before commit 412ad7a55639516f284cd0ef9757d6ae5c7abd43, delayChkpt was a Boolean. Now it's an integer. Extensions using it need to be appropriately updated, so let's rename the field to make sure that a hard compilation failure occurs. Replacing delayChkpt with delayChkptFlags made a few comments extend past 80 characters, so I reflowed them and changed some wording very slightly. The back-branches will need a different change to restore compatibility with existing minor releases; this is just for master. Per suggestion from Tom Lane. Discussion: http://postgr.es/m/a7880f4d-1d74-582a-ada7-dad168d046d1@enterprisedb.com
2022-04-08 17:44:17 +02:00
MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
END_CRIT_SECTION();
LWLockRelease(MultiXactTruncationLock);
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
}
/*
Prevent excess SimpleLruTruncate() deletion. Every core SLRU wraps around. With the exception of pg_notify, the wrap point can fall in the middle of a page. Account for this in the PagePrecedes callback specification and in SimpleLruTruncate()'s use of said callback. Update each callback implementation to fit the new specification. This changes SerialPagePrecedesLogically() from the style of asyncQueuePagePrecedes() to the style of CLOGPagePrecedes(). (Whereas pg_clog and pg_serial share a key space, pg_serial is nothing like pg_notify.) The bug fixed here has the same symptoms and user followup steps as 592a589a04bd456410b853d86bd05faa9432cbbb. Back-patch to 9.5 (all supported versions). Reviewed by Andrey Borodin and (in earlier versions) by Tom Lane. Discussion: https://postgr.es/m/20190202083822.GC32531@gust.leadboat.com
2021-01-16 21:21:35 +01:00
* Decide whether a MultiXactOffset page number is "older" for truncation
* purposes. Analogous to CLOGPagePrecedes().
*
* Offsetting the values is optional, because MultiXactIdPrecedes() has
* translational symmetry.
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 bool
MultiXactOffsetPagePrecedes(int page1, int page2)
{
MultiXactId multi1;
MultiXactId multi2;
multi1 = ((MultiXactId) page1) * MULTIXACT_OFFSETS_PER_PAGE;
Prevent excess SimpleLruTruncate() deletion. Every core SLRU wraps around. With the exception of pg_notify, the wrap point can fall in the middle of a page. Account for this in the PagePrecedes callback specification and in SimpleLruTruncate()'s use of said callback. Update each callback implementation to fit the new specification. This changes SerialPagePrecedesLogically() from the style of asyncQueuePagePrecedes() to the style of CLOGPagePrecedes(). (Whereas pg_clog and pg_serial share a key space, pg_serial is nothing like pg_notify.) The bug fixed here has the same symptoms and user followup steps as 592a589a04bd456410b853d86bd05faa9432cbbb. Back-patch to 9.5 (all supported versions). Reviewed by Andrey Borodin and (in earlier versions) by Tom Lane. Discussion: https://postgr.es/m/20190202083822.GC32531@gust.leadboat.com
2021-01-16 21:21:35 +01:00
multi1 += FirstMultiXactId + 1;
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
multi2 = ((MultiXactId) page2) * MULTIXACT_OFFSETS_PER_PAGE;
Prevent excess SimpleLruTruncate() deletion. Every core SLRU wraps around. With the exception of pg_notify, the wrap point can fall in the middle of a page. Account for this in the PagePrecedes callback specification and in SimpleLruTruncate()'s use of said callback. Update each callback implementation to fit the new specification. This changes SerialPagePrecedesLogically() from the style of asyncQueuePagePrecedes() to the style of CLOGPagePrecedes(). (Whereas pg_clog and pg_serial share a key space, pg_serial is nothing like pg_notify.) The bug fixed here has the same symptoms and user followup steps as 592a589a04bd456410b853d86bd05faa9432cbbb. Back-patch to 9.5 (all supported versions). Reviewed by Andrey Borodin and (in earlier versions) by Tom Lane. Discussion: https://postgr.es/m/20190202083822.GC32531@gust.leadboat.com
2021-01-16 21:21:35 +01:00
multi2 += FirstMultiXactId + 1;
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
Prevent excess SimpleLruTruncate() deletion. Every core SLRU wraps around. With the exception of pg_notify, the wrap point can fall in the middle of a page. Account for this in the PagePrecedes callback specification and in SimpleLruTruncate()'s use of said callback. Update each callback implementation to fit the new specification. This changes SerialPagePrecedesLogically() from the style of asyncQueuePagePrecedes() to the style of CLOGPagePrecedes(). (Whereas pg_clog and pg_serial share a key space, pg_serial is nothing like pg_notify.) The bug fixed here has the same symptoms and user followup steps as 592a589a04bd456410b853d86bd05faa9432cbbb. Back-patch to 9.5 (all supported versions). Reviewed by Andrey Borodin and (in earlier versions) by Tom Lane. Discussion: https://postgr.es/m/20190202083822.GC32531@gust.leadboat.com
2021-01-16 21:21:35 +01:00
return (MultiXactIdPrecedes(multi1, multi2) &&
MultiXactIdPrecedes(multi1,
multi2 + MULTIXACT_OFFSETS_PER_PAGE - 1));
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
}
/*
Prevent excess SimpleLruTruncate() deletion. Every core SLRU wraps around. With the exception of pg_notify, the wrap point can fall in the middle of a page. Account for this in the PagePrecedes callback specification and in SimpleLruTruncate()'s use of said callback. Update each callback implementation to fit the new specification. This changes SerialPagePrecedesLogically() from the style of asyncQueuePagePrecedes() to the style of CLOGPagePrecedes(). (Whereas pg_clog and pg_serial share a key space, pg_serial is nothing like pg_notify.) The bug fixed here has the same symptoms and user followup steps as 592a589a04bd456410b853d86bd05faa9432cbbb. Back-patch to 9.5 (all supported versions). Reviewed by Andrey Borodin and (in earlier versions) by Tom Lane. Discussion: https://postgr.es/m/20190202083822.GC32531@gust.leadboat.com
2021-01-16 21:21:35 +01:00
* Decide whether a MultiXactMember page number is "older" for truncation
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
* purposes. There is no "invalid offset number" so use the numbers verbatim.
*/
static bool
MultiXactMemberPagePrecedes(int page1, int page2)
{
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
MultiXactOffset offset1;
MultiXactOffset offset2;
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
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
offset1 = ((MultiXactOffset) page1) * MULTIXACT_MEMBERS_PER_PAGE;
offset2 = ((MultiXactOffset) page2) * MULTIXACT_MEMBERS_PER_PAGE;
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
Prevent excess SimpleLruTruncate() deletion. Every core SLRU wraps around. With the exception of pg_notify, the wrap point can fall in the middle of a page. Account for this in the PagePrecedes callback specification and in SimpleLruTruncate()'s use of said callback. Update each callback implementation to fit the new specification. This changes SerialPagePrecedesLogically() from the style of asyncQueuePagePrecedes() to the style of CLOGPagePrecedes(). (Whereas pg_clog and pg_serial share a key space, pg_serial is nothing like pg_notify.) The bug fixed here has the same symptoms and user followup steps as 592a589a04bd456410b853d86bd05faa9432cbbb. Back-patch to 9.5 (all supported versions). Reviewed by Andrey Borodin and (in earlier versions) by Tom Lane. Discussion: https://postgr.es/m/20190202083822.GC32531@gust.leadboat.com
2021-01-16 21:21:35 +01:00
return (MultiXactOffsetPrecedes(offset1, offset2) &&
MultiXactOffsetPrecedes(offset1,
offset2 + MULTIXACT_MEMBERS_PER_PAGE - 1));
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
}
/*
* Decide which of two MultiXactIds is earlier.
*
* XXX do we need to do something special for InvalidMultiXactId?
* (Doesn't look like 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
bool
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
MultiXactIdPrecedes(MultiXactId multi1, MultiXactId multi2)
{
int32 diff = (int32) (multi1 - multi2);
return (diff < 0);
}
Fix full-table-vacuum request mechanism for MultiXactIds While autovacuum dutifully launched anti-multixact-wraparound vacuums when the multixact "age" was reached, the vacuum code was not aware that it needed to make them be full table vacuums. As the resulting partial-table vacuums aren't capable of actually increasing relminmxid, autovacuum continued to launch anti-wraparound vacuums that didn't have the intended effect, until age of relfrozenxid caused the vacuum to finally be a full table one via vacuum_freeze_table_age. To fix, introduce logic for multixacts similar to that for plain TransactionIds, using the same GUCs. Backpatch to 9.3, where permanent MultiXactIds were introduced. Andres Freund, some cleanup by Álvaro
2013-11-28 20:52:54 +01:00
/*
* MultiXactIdPrecedesOrEquals -- is multi1 logically <= multi2?
*
* XXX do we need to do something special for InvalidMultiXactId?
* (Doesn't look like it.)
*/
bool
MultiXactIdPrecedesOrEquals(MultiXactId multi1, MultiXactId multi2)
{
int32 diff = (int32) (multi1 - multi2);
return (diff <= 0);
}
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
/*
* Decide which of two offsets is earlier.
*/
static bool
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
MultiXactOffsetPrecedes(MultiXactOffset offset1, MultiXactOffset offset2)
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
{
int32 diff = (int32) (offset1 - offset2);
return (diff < 0);
}
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
/*
* Write an xlog record reflecting the zeroing of either a MEMBERs or
* OFFSETs page (info shows which)
*/
static void
WriteMZeroPageXlogRec(int pageno, uint8 info)
{
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 *) (&pageno), sizeof(int));
(void) XLogInsert(RM_MULTIXACT_ID, info);
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
}
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
/*
* Write a TRUNCATE xlog record
*
* We must flush the xlog record to disk before returning --- see notes in
* TruncateCLOG().
*/
static void
WriteMTruncateXlogRec(Oid oldestMultiDB,
MultiXactId startTruncOff, MultiXactId endTruncOff,
MultiXactOffset startTruncMemb, MultiXactOffset endTruncMemb)
{
XLogRecPtr recptr;
xl_multixact_truncate xlrec;
xlrec.oldestMultiDB = oldestMultiDB;
xlrec.startTruncOff = startTruncOff;
xlrec.endTruncOff = endTruncOff;
xlrec.startTruncMemb = startTruncMemb;
xlrec.endTruncMemb = endTruncMemb;
XLogBeginInsert();
XLogRegisterData((char *) (&xlrec), SizeOfMultiXactTruncate);
recptr = XLogInsert(RM_MULTIXACT_ID, XLOG_MULTIXACT_TRUNCATE_ID);
XLogFlush(recptr);
}
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
/*
* MULTIXACT resource manager's routines
*/
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
multixact_redo(XLogReaderState *record)
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
{
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;
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
Add a new option to RestoreBkpBlocks() to indicate if a cleanup lock should be used instead of the normal exclusive lock, and make WAL redo functions responsible for calling RestoreBkpBlocks(). They know better what kind of a lock they need. At the moment, this just moves things around with no functional change, but makes the hot standby patch that's under review cleaner.
2009-01-20 19:59:37 +01:00
/* Backup blocks are not used in multixact records */
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(!XLogRecHasAnyBlockRefs(record));
Add a new option to RestoreBkpBlocks() to indicate if a cleanup lock should be used instead of the normal exclusive lock, and make WAL redo functions responsible for calling RestoreBkpBlocks(). They know better what kind of a lock they need. At the moment, this just moves things around with no functional change, but makes the hot standby patch that's under review cleaner.
2009-01-20 19:59:37 +01:00
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
if (info == XLOG_MULTIXACT_ZERO_OFF_PAGE)
{
int pageno;
int slotno;
memcpy(&pageno, XLogRecGetData(record), sizeof(int));
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockAcquire(MultiXactOffsetSLRULock, LW_EXCLUSIVE);
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
slotno = ZeroMultiXactOffsetPage(pageno, false);
Avoid unnecessary public struct declaration in slru.h Instead, declare a public wrapper of the sole function using it for external callers, so that they don't have to always pass a NULL argument. Author: Kevin Grittner
2010-12-30 16:09:17 +01:00
SimpleLruWritePage(MultiXactOffsetCtl, slotno);
Clean up representation of SLRU page state. This is the cleaner fix for the SLRU race condition that I posted a few days ago, but we decided not to use in 8.1 and older branches.
2005-11-05 22:19:47 +01:00
Assert(!MultiXactOffsetCtl->shared->page_dirty[slotno]);
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
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactOffsetSLRULock);
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
}
else if (info == XLOG_MULTIXACT_ZERO_MEM_PAGE)
{
int pageno;
int slotno;
memcpy(&pageno, XLogRecGetData(record), sizeof(int));
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockAcquire(MultiXactMemberSLRULock, LW_EXCLUSIVE);
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
slotno = ZeroMultiXactMemberPage(pageno, false);
Avoid unnecessary public struct declaration in slru.h Instead, declare a public wrapper of the sole function using it for external callers, so that they don't have to always pass a NULL argument. Author: Kevin Grittner
2010-12-30 16:09:17 +01:00
SimpleLruWritePage(MultiXactMemberCtl, slotno);
Clean up representation of SLRU page state. This is the cleaner fix for the SLRU race condition that I posted a few days ago, but we decided not to use in 8.1 and older branches.
2005-11-05 22:19:47 +01:00
Assert(!MultiXactMemberCtl->shared->page_dirty[slotno]);
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
Rename SLRU structures and associated LWLocks. Originally, the names assigned to SLRUs had no purpose other than being shmem lookup keys, so not a lot of thought went into them. As of v13, though, we're exposing them in the pg_stat_slru view and the pg_stat_reset_slru function, so it seems advisable to take a bit more care. Rename them to names based on the associated on-disk storage directories (which fortunately we *did* think about, to some extent; since those are also visible to DBAs, consistency seems like a good thing). Also rename the associated LWLocks, since those names are likewise user-exposed now as wait event names. For the most part I only touched symbols used in the respective modules' SimpleLruInit() calls, not the names of other related objects. This renaming could have been taken further, and maybe someday we will do so. But for now it seems undesirable to change the names of any globally visible functions or structs, so some inconsistency is unavoidable. (But I *did* terminate "oldserxid" with prejudice, as I found that name both unreadable and not descriptive of the SLRU's contents.) Table 27.12 needs re-alphabetization now, but I'll leave that till after the other LWLock renamings I have in mind. Discussion: https://postgr.es/m/28683.1589405363@sss.pgh.pa.us
2020-05-15 20:28:19 +02:00
LWLockRelease(MultiXactMemberSLRULock);
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
}
else if (info == XLOG_MULTIXACT_CREATE_ID)
{
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
xl_multixact_create *xlrec =
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
(xl_multixact_create *) XLogRecGetData(record);
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
TransactionId max_xid;
int i;
/* Store the data back into the SLRU files */
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
RecordNewMultiXact(xlrec->mid, xlrec->moff, xlrec->nmembers,
xlrec->members);
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
/* Make sure nextMXact/nextOffset are beyond what this record has */
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
MultiXactAdvanceNextMXact(xlrec->mid + 1,
xlrec->moff + xlrec->nmembers);
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
/*
Rename VariableCacheData.nextFullXid to nextXid. Including Full in variable names duplicates the type information and leads to overly long names. As FullTransactionId cannot accidentally be casted to TransactionId that does not seem necessary. Author: Andres Freund Discussion: https://postgr.es/m/20200724011143.jccsyvsvymuiqfxu@alap3.anarazel.de
2020-08-11 20:25:23 +02:00
* Make sure nextXid is beyond any XID mentioned in the record. This
Add basic infrastructure for 64 bit transaction IDs. Instead of inferring epoch progress from xids and checkpoints, introduce a 64 bit FullTransactionId type and use it to track xid generation. This fixes an unlikely bug where the epoch is reported incorrectly if the range of active xids wraps around more than once between checkpoints. The only user-visible effect of this commit is to correct the epoch used by txid_current() and txid_status(), also visible with pg_controldata, in those rare circumstances. It also creates some basic infrastructure so that later patches can use 64 bit transaction IDs in more places. The new type is a struct that we pass by value, as a form of strong typedef. This prevents the sort of accidental confusion between TransactionId and FullTransactionId that would be possible if we were to use a plain old uint64. Author: Thomas Munro Reported-by: Amit Kapila Reviewed-by: Andres Freund, Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/CAA4eK1%2BMv%2Bmb0HFfWM9Srtc6MVe160WFurXV68iAFMcagRZ0dQ%40mail.gmail.com
2019-03-27 22:34:43 +01:00
* should be unnecessary, since any XID found here ought to have other
* evidence in the XLOG, but let's be safe.
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
*/
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
max_xid = XLogRecGetXid(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
for (i = 0; i < xlrec->nmembers; i++)
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
{
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 (TransactionIdPrecedes(max_xid, xlrec->members[i].xid))
max_xid = xlrec->members[i].xid;
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
}
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
Add basic infrastructure for 64 bit transaction IDs. Instead of inferring epoch progress from xids and checkpoints, introduce a 64 bit FullTransactionId type and use it to track xid generation. This fixes an unlikely bug where the epoch is reported incorrectly if the range of active xids wraps around more than once between checkpoints. The only user-visible effect of this commit is to correct the epoch used by txid_current() and txid_status(), also visible with pg_controldata, in those rare circumstances. It also creates some basic infrastructure so that later patches can use 64 bit transaction IDs in more places. The new type is a struct that we pass by value, as a form of strong typedef. This prevents the sort of accidental confusion between TransactionId and FullTransactionId that would be possible if we were to use a plain old uint64. Author: Thomas Munro Reported-by: Amit Kapila Reviewed-by: Andres Freund, Tom Lane, Heikki Linnakangas Discussion: https://postgr.es/m/CAA4eK1%2BMv%2Bmb0HFfWM9Srtc6MVe160WFurXV68iAFMcagRZ0dQ%40mail.gmail.com
2019-03-27 22:34:43 +01:00
AdvanceNextFullTransactionIdPastXid(max_xid);
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
}
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
else if (info == XLOG_MULTIXACT_TRUNCATE_ID)
{
xl_multixact_truncate xlrec;
int pageno;
memcpy(&xlrec, XLogRecGetData(record),
SizeOfMultiXactTruncate);
elog(DEBUG1, "replaying multixact truncation: "
"offsets [%u, %u), offsets segments [%x, %x), "
"members [%u, %u), members segments [%x, %x)",
xlrec.startTruncOff, xlrec.endTruncOff,
MultiXactIdToOffsetSegment(xlrec.startTruncOff),
MultiXactIdToOffsetSegment(xlrec.endTruncOff),
xlrec.startTruncMemb, xlrec.endTruncMemb,
MXOffsetToMemberSegment(xlrec.startTruncMemb),
MXOffsetToMemberSegment(xlrec.endTruncMemb));
/* should not be required, but more than cheap enough */
LWLockAcquire(MultiXactTruncationLock, LW_EXCLUSIVE);
/*
* Advance the horizon values, so they're current at the end of
* recovery.
*/
Make logging about multixact wraparound protection less chatty. The original messaging design, introduced in commit 068cfadf9, seems too chatty now that some time has elapsed since the bug fix; most installations will be in good shape and don't really need a reminder about this on every postmaster start. Hence, arrange to suppress the "wraparound protections are now enabled" message during startup (specifically, during the TrimMultiXact() call). The message will still appear if protection becomes effective at some later point. Discussion: https://postgr.es/m/17211.1489189214@sss.pgh.pa.us
2017-03-14 17:47:46 +01:00
SetMultiXactIdLimit(xlrec.endTruncOff, xlrec.oldestMultiDB, false);
Rework the way multixact truncations work. The fact that multixact truncations are not WAL logged has caused a fair share of problems. Amongst others it requires to do computations during recovery while the database is not in a consistent state, delaying truncations till checkpoints, and handling members being truncated, but offset not. We tried to put bandaids on lots of these issues over the last years, but it seems time to change course. Thus this patch introduces WAL logging for multixact truncations. This allows: 1) to perform the truncation directly during VACUUM, instead of delaying it to the checkpoint. 2) to avoid looking at the offsets SLRU for truncation during recovery, we can just use the master's values. 3) simplify a fair amount of logic to keep in memory limits straight, this has gotten much easier During the course of fixing this a bunch of additional bugs had to be fixed: 1) Data was not purged from memory the member's SLRU before deleting segments. This happened to be hard or impossible to hit due to the interlock between checkpoints and truncation. 2) find_multixact_start() relied on SimpleLruDoesPhysicalPageExist - but that doesn't work for offsets that haven't yet been flushed to disk. Add code to flush the SLRUs to fix. Not pretty, but it feels slightly safer to only make decisions based on actual on-disk state. 3) find_multixact_start() could be called concurrently with a truncation and thus fail. Via SetOffsetVacuumLimit() that could lead to a round of emergency vacuuming. The problem remains in pg_get_multixact_members(), but that's quite harmless. For now this is going to only get applied to 9.5+, leaving the issues in the older branches in place. It is quite possible that we need to backpatch at a later point though. For the case this gets backpatched we need to handle that an updated standby may be replaying WAL from a not-yet upgraded primary. We have to recognize that situation and use "old style" truncation (i.e. looking at the SLRUs) during WAL replay. In contrast to before, this now happens in the startup process, when replaying a checkpoint record, instead of the checkpointer. Doing truncation in the restartpoint is incorrect, they can happen much later than the original checkpoint, thereby leading to wraparound. To avoid "multixact_redo: unknown op code 48" errors standbys would have to be upgraded before primaries. A later patch will bump the WAL page magic, and remove the legacy truncation codepaths. Legacy truncation support is just included to make a possible future backpatch easier. Discussion: 20150621192409.GA4797@alap3.anarazel.de Reviewed-By: Robert Haas, Alvaro Herrera, Thomas Munro Backpatch: 9.5 for now
2015-09-26 19:04:25 +02:00
PerformMembersTruncation(xlrec.startTruncMemb, xlrec.endTruncMemb);
/*
* During XLOG replay, latest_page_number isn't necessarily set up
* yet; insert a suitable value to bypass the sanity test in
* SimpleLruTruncate.
*/
pageno = MultiXactIdToOffsetPage(xlrec.endTruncOff);
MultiXactOffsetCtl->shared->latest_page_number = pageno;
PerformOffsetsTruncation(xlrec.startTruncOff, xlrec.endTruncOff);
LWLockRelease(MultiXactTruncationLock);
}
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
else
elog(PANIC, "multixact_redo: unknown op code %u", info);
}
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
Datum
pg_get_multixact_members(PG_FUNCTION_ARGS)
{
typedef struct
{
MultiXactMember *members;
int nmembers;
int iter;
} mxact;
Use PG_GETARG_TRANSACTIONID where appropriate Some places were using PG_GETARG_UINT32 where PG_GETARG_TRANSACTIONID would be more appropriate. (Of course, they are the same internally, so there is no externally visible effect.) To do that, export PG_GETARG_TRANSACTIONID outside of xid.c. We also export PG_RETURN_TRANSACTIONID for symmetry, even though there are currently no external users. Author: Ashutosh Bapat <ashutosh.bapat@2ndquadrant.com> Discussion: https://www.postgresql.org/message-id/flat/d8f6bdd536df403b9b33816e9f7e0b9d@G08CNEXMBPEKD05.g08.fujitsu.local
2020-11-02 16:48:22 +01:00
MultiXactId mxid = PG_GETARG_TRANSACTIONID(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
mxact *multi;
FuncCallContext *funccxt;
if (mxid < FirstMultiXactId)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("invalid MultiXactId: %u", mxid)));
if (SRF_IS_FIRSTCALL())
{
MemoryContext oldcxt;
TupleDesc tupdesc;
funccxt = SRF_FIRSTCALL_INIT();
oldcxt = MemoryContextSwitchTo(funccxt->multi_call_memory_ctx);
multi = palloc(sizeof(mxact));
/* no need to allow for old values here */
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
multi->nmembers = GetMultiXactIdMembers(mxid, &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
multi->iter = 0;
Switch some system functions to use get_call_result_type() This shaves some code by replacing the combinations of CreateTemplateTupleDesc()/TupleDescInitEntry() hardcoding a mapping of the attributes listed in pg_proc.dat by get_call_result_type() to build the TupleDesc needed for the rows generated. get_call_result_type() is more expensive than the former style, but this removes some duplication with the lists of OUT parameters (pg_proc.dat and the attributes hardcoded in these code paths). This is applied to functions that are not considered as critical (aka that could be called repeatedly for monitoring purposes). Author: Bharath Rupireddy Reviewed-by: Robert Haas, Álvaro Herrera, Tom Lane, Michael Paquier Discussion: https://postgr.es/m/CALj2ACV23HW5HP5hFjd89FNS-z5X8r2jNXdMXcpN2BgTtKd87w@mail.gmail.com
2022-12-21 02:11:22 +01:00
if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
elog(ERROR, "return type must be a row type");
funccxt->tuple_desc = tupdesc;
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
funccxt->attinmeta = TupleDescGetAttInMetadata(tupdesc);
funccxt->user_fctx = multi;
MemoryContextSwitchTo(oldcxt);
}
funccxt = SRF_PERCALL_SETUP();
multi = (mxact *) funccxt->user_fctx;
while (multi->iter < multi->nmembers)
{
HeapTuple tuple;
char *values[2];
Add more use of psprintf()
2014-01-07 03:30:26 +01:00
values[0] = psprintf("%u", multi->members[multi->iter].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
values[1] = mxstatus_to_string(multi->members[multi->iter].status);
tuple = BuildTupleFromCStrings(funccxt->attinmeta, values);
multi->iter++;
pfree(values[0]);
SRF_RETURN_NEXT(funccxt, HeapTupleGetDatum(tuple));
}
SRF_RETURN_DONE(funccxt);
}
Defer flushing of SLRU files. Previously, we called fsync() after writing out individual pg_xact, pg_multixact and pg_commit_ts pages due to cache pressure, leading to regular I/O stalls in user backends and recovery. Collapse requests for the same file into a single system call as part of the next checkpoint, as we already did for relation files, using the infrastructure developed by commit 3eb77eba. This can cause a significant improvement to recovery performance, especially when it's otherwise CPU-bound. Hoist ProcessSyncRequests() up into CheckPointGuts() to make it clearer that it applies to all the SLRU mini-buffer-pools as well as the main buffer pool. Rearrange things so that data collected in CheckpointStats includes SLRU activity. Also remove the Shutdown{CLOG,CommitTS,SUBTRANS,MultiXact}() functions, because they were redundant after the shutdown checkpoint that immediately precedes them. (I'm not sure if they were ever needed, but they aren't now.) Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us> (parts) Tested-by: Jakub Wartak <Jakub.Wartak@tomtom.com> Discussion: https://postgr.es/m/CA+hUKGLJ=84YT+NvhkEEDAuUtVHMfQ9i-N7k_o50JmQ6Rpj_OQ@mail.gmail.com
2020-09-25 08:49:43 +02:00
/*
* Entrypoint for sync.c to sync offsets files.
*/
int
multixactoffsetssyncfiletag(const FileTag *ftag, char *path)
{
return SlruSyncFileTag(MultiXactOffsetCtl, ftag, path);
}
/*
* Entrypoint for sync.c to sync members files.
*/
int
multixactmemberssyncfiletag(const FileTag *ftag, char *path)
{
return SlruSyncFileTag(MultiXactMemberCtl, ftag, path);
}