This set of commits has some bugs with known fixes, but at this late
stage in the release cycle it seems best to revert and resubmit next
time, along with some new automated test coverage for this whole area.
Commits reverted:
dc88460c: Doc: Review for "Optionally prefetch referenced data in recovery."
1d257577: Optionally prefetch referenced data in recovery.
f003d9f8: Add circular WAL decoding buffer.
323cbe7c: Remove read_page callback from XLogReader.
Remove the new GUC group WAL_RECOVERY recently added by a55a9847, as the
corresponding section of config.sgml is now reverted.
Discussion: https://postgr.es/m/CAOuzzgrn7iKnFRsB4MHp3UisEQAGgZMbk_ViTN4HV4-Ksq8zCg%40mail.gmail.com
Currently, replication slot statistics are updated at prepare, commit, and
rollback. Now, if the transaction is interrupted the stats might not get
updated. Fixed this by updating replication statistics after every
stream/spill.
In passing update the docs to change the description of some of the slot
stats.
Author: Vignesh C, Sawada Masahiko
Reviewed-by: Amit Kapila
Discussion: https://postgr.es/m/20210319185247.ldebgpdaxsowiflw@alap3.anarazel.de
During decoding of an in-progress or prepared transaction, we detect
concurrent abort with an error code ERRCODE_TRANSACTION_ROLLBACK. That is
not sufficient because a callback can decide to throw that error code
at other times as well.
Reported-by: Tom Lane
Author: Amit Kapila
Reviewed-by: Dilip Kumar
Discussion: https://postgr.es/m/CAA4eK1KCjPRS4aZHB48QMM4J8XOC1+TD8jo-4Yu84E+MjwqVhA@mail.gmail.com
Previously, we were using the size of all the changes present in
ReorderBuffer to compute total_bytes after decoding a transaction and that
can lead to counting some of the transactions' changes more than once. Fix
it by using the size of the changes decoded for a transaction to compute
'total_bytes'.
Author: Sawada Masahiko
Reviewed-by: Vignesh C, Amit Kapila
Discussion: https://postgr.es/m/20210319185247.ldebgpdaxsowiflw@alap3.anarazel.de
There were two problems:
a. We were always selecting the next available txn instead of selecting it
when it is larger than the previous transaction.
b. We were selecting the transactions which haven't made any changes to
the database (base snapshot is not set). Later it was hitting an Assert
because we don't decode such transactions and the changes in txn remain as
it is. It is better not to choose such transactions for streaming in the
first place.
Reported-by: Haiying Tang
Author: Dilip Kumar
Reviewed-by: Amit Kapila
Discussion: https://postgr.es/m/OS0PR01MB61133B94E63177040F7ECDA1FB429@OS0PR01MB6113.jpnprd01.prod.outlook.com
Commit 8ff1c94649 allowed TRUNCATE command to truncate foreign tables.
Previously the information about "ONLY" options specified in TRUNCATE
command were passed to the foreign data wrapper. Then postgres_fdw
constructed the TRUNCATE command to issue the remote server and
included "ONLY" options in it based on the passed information.
On the other hand, "ONLY" options specified in SELECT, UPDATE or DELETE
have no effect when accessing or modifying the remote table, i.e.,
are not passed to the foreign data wrapper. So it's inconsistent to
make only TRUNCATE command pass the "ONLY" options to the foreign data
wrapper. Therefore this commit changes the TRUNCATE command so that
it doesn't pass the "ONLY" options to the foreign data wrapper,
for the consistency with other statements. Also this commit changes
postgres_fdw so that it always doesn't include "ONLY" options in
the TRUNCATE command that it constructs.
Author: Fujii Masao
Reviewed-by: Bharath Rupireddy, Kyotaro Horiguchi, Justin Pryzby, Zhihong Yu
Discussion: https://postgr.es/m/551ed8c1-f531-818b-664a-2cecdab99cd8@oss.nttdata.com
Previously, we used to use the array of size max_replication_slots to
store stats for replication slots. But that had two problems in the cases
where a message for dropping a slot gets lost: 1) the stats for the new
slot are not recorded if the array is full and 2) writing beyond the end
of the array if the user reduces the max_replication_slots.
This commit uses HTAB for replication slot statistics, resolving both
problems. Now, pgstat_vacuum_stat() search for all the dead replication
slots in stats hashtable and tell the collector to remove them. To avoid
showing the stats for the already-dropped slots, pg_stat_replication_slots
view searches slot stats by the slot name taken from pg_replication_slots.
Also, we send a message for creating a slot at slot creation, initializing
the stats. This reduces the possibility that the stats are accumulated
into the old slot stats when a message for dropping a slot gets lost.
Reported-by: Andres Freund
Author: Sawada Masahiko, test case by Vignesh C
Reviewed-by: Amit Kapila, Vignesh C, Dilip Kumar
Discussion: https://postgr.es/m/20210319185247.ldebgpdaxsowiflw@alap3.anarazel.de
The Truncate operation acquires an exclusive lock on the target relation
and indexes. It then waits for logical replication of the operation to
finish at commit. Now because we are acquiring the shared lock on the
target index to get index attributes in pgoutput while sending the
changes for the Truncate operation, it leads to a deadlock.
Actually, we don't need to acquire a lock on the target index as we build
the cache entry using a historic snapshot and all the later changes are
absorbed while decoding WAL. So, we wrote a special purpose function for
logical replication to get a bitmap of replica identity attribute numbers
where we get that information without locking the target index.
We decided not to backpatch this as there doesn't seem to be any field
complaint about this issue since it was introduced in commit 5dfd1e5a in
v11.
Reported-by: Haiying Tang
Author: Takamichi Osumi, test case by Li Japin
Reviewed-by: Amit Kapila, Ajin Cherian
Discussion: https://postgr.es/m/OS0PR01MB6113C2499C7DC70EE55ADB82FB759@OS0PR01MB6113.jpnprd01.prod.outlook.com
We send the prepare for the concurrently aborted xacts so that later when
rollback prepared is decoded and sent, the downstream should be able to
rollback such a xact. For 'streaming' case (when we send changes for
in-progress transactions), we were sending prepare twice when concurrent
abort was detected.
Author: Peter Smith
Reviewed-by: Amit Kapila
Discussion: https://postgr.es/m/f82133c6-6055-b400-7922-97dae9f2b50b@enterprisedb.com
Creating a trigger on a relation to which an apply operation is
triggered would cause a relation leak once the change gets committed,
as the executor would miss that the relation needs to be closed
beforehand. This issue got introduced with the refactoring done in
1375422c, where it becomes necessary to track relations within
es_opened_result_relations to make sure that they are closed.
We have discussed using ExecInitResultRelation() coupled with
ExecCloseResultRelations() for the relations in need of tracking by the
apply operations in the subscribers, which would simplify greatly the
opening and closing of indexes, but this requires a larger rework and
reorganization of the worker code, particularly for the tuple routing
part. And that's not really welcome post feature freeze. So, for now,
settle down to the same solution as TRUNCATE which is to fill in
es_opened_result_relations with the relation opened, to make sure that
ExecGetTriggerResultRel() finds them and that they get closed.
The code is lightly refactored so as a relation is not registered three
times for each DML code path, making the whole a bit easier to follow.
Reported-by: Tang Haiying, Shi Yu, Hou Zhijie
Author: Amit Langote, Masahiko Sawada, Hou Zhijie
Reviewed-by: Amit Kapila, Michael Paquier
Discussion: https://postgr.es/m/OS0PR01MB611383FA0FE92EB9DE21946AFB769@OS0PR01MB6113.jpnprd01.prod.outlook.com
This adds the statistics about total transactions count and total
transaction data logically sent to the decoding output plugin from
ReorderBuffer. Users can query the pg_stat_replication_slots view to check
these stats.
Suggested-by: Andres Freund
Author: Vignesh C and Amit Kapila
Reviewed-by: Sawada Masahiko, Amit Kapila
Discussion: https://postgr.es/m/20210319185247.ldebgpdaxsowiflw@alap3.anarazel.de
This will make it consistent with the other usage of slotname in the code.
In the passing, change pgstat_report_replslot signature to use a structure
rather than multiple parameters.
Reported-by: Andres Freund
Author: Vignesh C
Reviewed-by: Sawada Masahiko, Amit Kapila
Discussion: https://postgr.es/m/20210319185247.ldebgpdaxsowiflw@alap3.anarazel.de
This commit introduces new foreign data wrapper API for TRUNCATE.
It extends TRUNCATE command so that it accepts foreign tables as
the targets to truncate and invokes that API. Also it extends postgres_fdw
so that it can issue TRUNCATE command to foreign servers, by adding
new routine for that TRUNCATE API.
The information about options specified in TRUNCATE command, e.g.,
ONLY, CACADE, etc is passed to FDW via API. The list of foreign tables to
truncate is also passed to FDW. FDW truncates the foreign data sources
that the passed foreign tables specify, based on those information.
For example, postgres_fdw constructs TRUNCATE command using them
and issues it to the foreign server.
For performance, TRUNCATE command invokes the FDW routine for
TRUNCATE once per foreign server that foreign tables to truncate belong to.
Author: Kazutaka Onishi, Kohei KaiGai, slightly modified by Fujii Masao
Reviewed-by: Bharath Rupireddy, Michael Paquier, Zhihong Yu, Alvaro Herrera, Stephen Frost, Ashutosh Bapat, Amit Langote, Daniel Gustafsson, Ibrar Ahmed, Fujii Masao
Discussion: https://postgr.es/m/CAOP8fzb_gkReLput7OvOK+8NHgw-RKqNv59vem7=524krQTcWA@mail.gmail.com
Discussion: https://postgr.es/m/CAJuF6cMWDDqU-vn_knZgma+2GMaout68YUgn1uyDnexRhqqM5Q@mail.gmail.com
Teach xlogreader.c to decode its output into a circular buffer, to
support optimizations based on looking ahead.
* XLogReadRecord() works as before, consuming records one by one, and
allowing them to be examined via the traditional XLogRecGetXXX()
macros.
* An alternative new interface XLogNextRecord() is added that returns
pointers to DecodedXLogRecord structs that can be examined directly.
* XLogReadAhead() provides a second cursor that lets you see
further ahead, as long as data is available and there is enough space
in the decoding buffer. This returns DecodedXLogRecord pointers to the
caller, but also adds them to a queue of records that will later be
consumed by XLogNextRecord()/XLogReadRecord().
The buffer's size is controlled with wal_decode_buffer_size. The buffer
could potentially be placed into shared memory, for future projects.
Large records that don't fit in the circular buffer are called
"oversized" and allocated separately with palloc().
Discussion: https://postgr.es/m/CA+hUKGJ4VJN8ttxScUFM8dOKX0BrBiboo5uz1cq=AovOddfHpA@mail.gmail.com
Previously, the XLogReader module would fetch new input data using a
callback function. Redesign the interface so that it tells the caller
to insert more data with a special return value instead. This API suits
later patches for prefetching, encryption and maybe other future
projects that would otherwise require continually extending the callback
interface.
As incidental cleanup work, move global variables readOff, readLen and
readSegNo inside XlogReaderState.
Author: Kyotaro HORIGUCHI <horiguchi.kyotaro@lab.ntt.co.jp>
Author: Heikki Linnakangas <hlinnaka@iki.fi> (parts of earlier version)
Reviewed-by: Antonin Houska <ah@cybertec.at>
Reviewed-by: Alvaro Herrera <alvherre@2ndquadrant.com>
Reviewed-by: Takashi Menjo <takashi.menjo@gmail.com>
Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Thomas Munro <thomas.munro@gmail.com>
Discussion: https://postgr.es/m/20190418.210257.43726183.horiguchi.kyotaro%40lab.ntt.co.jp
Arrange to do some things on-demand, rather than immediately during
executor startup, because there's a fair chance of never having to do
them at all:
* Don't open result relations' indexes until needed.
* Don't initialize partition tuple routing, nor the child-to-root
tuple conversion map, until needed.
This wins in UPDATEs on partitioned tables when only some of the
partitions will actually receive updates; with larger partition
counts the savings is quite noticeable. Also, we can remove some
sketchy heuristics in ExecInitModifyTable about whether to set up
tuple routing.
Also, remove execPartition.c's private hash table tracking which
partitions were already opened by the ModifyTable node. Instead
use the hash added to ModifyTable itself by commit 86dc90056.
To allow lazy computation of the conversion maps, we now set
ri_RootResultRelInfo in all child ResultRelInfos. We formerly set it
only in some, not terribly well-defined, cases. This has user-visible
side effects in that now more error messages refer to the root
relation instead of some partition (and provide error data in the
root's column order, too). It looks to me like this is a strict
improvement in consistency, so I don't have a problem with the
output changes visible in this commit.
Extracted from a larger patch, which seemed to me to be too messy
to push in one commit.
Amit Langote, reviewed at different times by Heikki Linnakangas and
myself
Discussion: https://postgr.es/m/CA+HiwqG7ZruBmmih3wPsBZ4s0H2EhywrnXEduckY5Hr3fWzPWA@mail.gmail.com
Retry the call to heap_prune_page() in rare cases where there is
disagreement between the heap_prune_page() call and the call to
HeapTupleSatisfiesVacuum() that immediately follows. Disagreement is
possible when a concurrently-aborted transaction makes a tuple DEAD
during the tiny window between each step. This was the only case where
a tuple considered DEAD by VACUUM still had storage following pruning.
VACUUM's definition of dead tuples is now uniformly simple and
unambiguous: dead tuples from each page are always LP_DEAD line pointers
that were encountered just after we performed pruning (and just before
we considered freezing remaining items with tuple storage).
Eliminating the tupgone=true special case enables INDEX_CLEANUP=off
style skipping of index vacuuming that takes place based on flexible,
dynamic criteria. The INDEX_CLEANUP=off case had to know about skipping
indexes up-front before now, due to a subtle interaction with the
special case (see commit dd695979) -- this was a special case unto
itself. Now there are no special cases. And so now it won't matter
when or how we decide to skip index vacuuming: it won't affect how
pruning behaves, and it won't be affected by any of the implementation
details of pruning or freezing.
Also remove XLOG_HEAP2_CLEANUP_INFO records. These are no longer
necessary because we now rely entirely on heap pruning taking care of
recovery conflicts. There is no longer any need to generate recovery
conflicts for DEAD tuples that pruning just missed. This also means
that heap vacuuming now uses exactly the same strategy for recovery
conflicts as index vacuuming always has: REDO routines never need to
process a latestRemovedXid from the WAL record, since earlier REDO of
the WAL record from pruning is sufficient in all cases. The generic
XLOG_HEAP2_CLEAN record type is now split into two new record types to
reflect this new division (these are called XLOG_HEAP2_PRUNE and
XLOG_HEAP2_VACUUM).
Also stop acquiring a super-exclusive lock for heap pages when they're
vacuumed during VACUUM's second heap pass. A regular exclusive lock is
enough. This is correct because heap page vacuuming is now strictly a
matter of setting the LP_DEAD line pointers to LP_UNUSED. No other
backend can have a pointer to a tuple located in a pinned buffer that
can be invalidated by a concurrent heap page vacuum operation.
Heap vacuuming can now be thought of as conceptually similar to index
vacuuming and conceptually dissimilar to heap pruning. Heap pruning now
has sole responsibility for anything involving the logical contents of
the database (e.g., managing transaction status information, recovery
conflicts, considering what to do with HOT chains). Index vacuuming and
heap vacuuming are now only concerned with recycling garbage items from
physical data structures that back the logical database.
Bump XLOG_PAGE_MAGIC due to pruning and heap page vacuum WAL record
changes.
Credit for the idea of retrying pruning a page to avoid the tupgone case
goes to Andres Freund.
Author: Peter Geoghegan <pg@bowt.ie>
Reviewed-By: Andres Freund <andres@anarazel.de>
Reviewed-By: Masahiko Sawada <sawada.mshk@gmail.com>
Discussion: https://postgr.es/m/CAH2-WznneCXTzuFmcwx_EyRQgfsfJAAsu+CsqRFmFXCAar=nJw@mail.gmail.com
The output plugin accepts a new parameter (messages) that controls if
logical decoding messages are written into the replication stream. It is
useful for those clients that use pgoutput as an output plugin and needs
to process messages that were written by pg_logical_emit_message().
Although logical streaming replication protocol supports logical
decoding messages now, logical replication does not use this feature yet.
Author: David Pirotte, Euler Taveira
Reviewed-by: Euler Taveira, Andres Freund, Ashutosh Bapat, Amit Kapila
Discussion: https://postgr.es/m/CADK3HHJ-+9SO7KuRLH=9Wa1rAo60Yreq1GFNkH_kd0=CdaWM+A@mail.gmail.com
It is possible that while decoding a prepared transaction, it gets aborted
concurrently via a ROLLBACK PREPARED command. In that case, we were
skipping all the changes and directly sending Rollback Prepared when we
find the same in WAL. However, the downstream has no idea of the GID of
such a transaction. So, ensure to send prepare even when a concurrent
abort is detected.
Author: Ajin Cherian
Reviewed-by: Markus Wanner, Amit Kapila
Discussion: https://postgr.es/m/f82133c6-6055-b400-7922-97dae9f2b50b@enterprisedb.com
Put the remote end's error message into the primary error string,
instead of relegating it to errdetail(). Although this could end up
being awkward if the remote sends us a really long error message,
it seems more in keeping with our message style guidelines, and more
helpful in situations where the errdetail could get dropped.
Peter Smith
Discussion: https://postgr.es/m/CAHut+Ps-Qv2yQceCwobQDP0aJOkfDzRFrOaR6+2Op2K=WHGeWg@mail.gmail.com
Along with gid, this provides a different way to identify the transaction.
The users that use xid in some way to prepare the transactions can use it
to filter prepare transactions. The later commands COMMIT PREPARED or
ROLLBACK PREPARED carries both identifiers, providing an output plugin the
choice of what to use.
Author: Markus Wanner
Reviewed-by: Vignesh C, Amit Kapila
Discussion: https://postgr.es/m/ee280000-7355-c4dc-e47b-2436e7be959c@enterprisedb.com
There is now a per-column COMPRESSION option which can be set to pglz
(the default, and the only option in up until now) or lz4. Or, if you
like, you can set the new default_toast_compression GUC to lz4, and
then that will be the default for new table columns for which no value
is specified. We don't have lz4 support in the PostgreSQL code, so
to use lz4 compression, PostgreSQL must be built --with-lz4.
In general, TOAST compression means compression of individual column
values, not the whole tuple, and those values can either be compressed
inline within the tuple or compressed and then stored externally in
the TOAST table, so those properties also apply to this feature.
Prior to this commit, a TOAST pointer has two unused bits as part of
the va_extsize field, and a compessed datum has two unused bits as
part of the va_rawsize field. These bits are unused because the length
of a varlena is limited to 1GB; we now use them to indicate the
compression type that was used. This means we only have bit space for
2 more built-in compresison types, but we could work around that
problem, if necessary, by introducing a new vartag_external value for
any further types we end up wanting to add. Hopefully, it won't be
too important to offer a wide selection of algorithms here, since
each one we add not only takes more coding but also adds a build
dependency for every packager. Nevertheless, it seems worth doing
at least this much, because LZ4 gets better compression than PGLZ
with less CPU usage.
It's possible for LZ4-compressed datums to leak into composite type
values stored on disk, just as it is for PGLZ. It's also possible for
LZ4-compressed attributes to be copied into a different table via SQL
commands such as CREATE TABLE AS or INSERT .. SELECT. It would be
expensive to force such values to be decompressed, so PostgreSQL has
never done so. For the same reasons, we also don't force recompression
of already-compressed values even if the target table prefers a
different compression method than was used for the source data. These
architectural decisions are perhaps arguable but revisiting them is
well beyond the scope of what seemed possible to do as part of this
project. However, it's relatively cheap to recompress as part of
VACUUM FULL or CLUSTER, so this commit adjusts those commands to do
so, if the configured compression method of the table happens not to
match what was used for some column value stored therein.
Dilip Kumar. The original patches on which this work was based were
written by Ildus Kurbangaliev, and those were patches were based on
even earlier work by Nikita Glukhov, but the design has since changed
very substantially, since allow a potentially large number of
compression methods that could be added and dropped on a running
system proved too problematic given some of the architectural issues
mentioned above; the choice of which specific compression method to
add first is now different; and a lot of the code has been heavily
refactored. More recently, Justin Przyby helped quite a bit with
testing and reviewing and this version also includes some code
contributions from him. Other design input and review from Tomas
Vondra, Álvaro Herrera, Andres Freund, Oleg Bartunov, Alexander
Korotkov, and me.
Discussion: http://postgr.es/m/20170907194236.4cefce96%40wp.localdomain
Discussion: http://postgr.es/m/CAFiTN-uUpX3ck%3DK0mLEk-G_kUQY%3DSNOTeqdaNRR9FMdQrHKebw%40mail.gmail.com
Commit ce0fdbfe97 made tablesync slots permanent and allow Drop
Subscription to drop such slots. However, it is possible that before
tablesync worker could get the acknowledgment of slot creation, drop
subscription stops it and that can lead to a dangling slot on the
publisher. Prevent cancel/die interrupts while creating a slot in the
tablesync worker.
Reported-by: Thomas Munro as per buildfarm
Author: Amit Kapila
Reviewed-by: Vignesh C, Takamichi Osumi
Discussion: https://postgr.es/m/CA+hUKGJG9dWpw1cOQ2nzWU8PHjm=PTraB+KgE5648K9nTfwvxg@mail.gmail.com
Commit 0aa8a01d04 extends the output plugin API to allow decoding of
prepared xacts and allowed the user to enable/disable the two-phase option
via pg_logical_slot_get_changes(). This can lead to a problem such that
the first time when it gets changes via pg_logical_slot_get_changes()
without two_phase option enabled it will not get the prepared even though
prepare is after consistent snapshot. Now next time during getting changes,
if the two_phase option is enabled it can skip prepare because by that
time start decoding point has been moved. So the user will only get commit
prepared.
Allow to enable/disable this option at the create slot time and default
will be false. It will break the existing slots which is fine in a major
release.
Author: Ajin Cherian
Reviewed-by: Amit Kapila and Vignesh C
Discussion: https://postgr.es/m/d0f60d60-133d-bf8d-bd70-47784d8fabf3@enterprisedb.com
In commit a271a1b50e, we allowed decoding at prepare time and the prepare
was decoded again if there is a restart after decoding it. It was done
that way because we can't distinguish between the cases where we have not
decoded the prepare because it was prior to consistent snapshot or we have
decoded it earlier but restarted. To distinguish between these two cases,
we have introduced an initial_consistent_point at the slot level which is
an LSN at which we found a consistent point at the time of slot creation.
This is also the point where we have exported a snapshot for the initial
copy. So, prepare transaction prior to this point are sent along with
commit prepared.
This commit bumps SNAPBUILD_VERSION because of change in SnapBuild. It
will break existing slots which is fine in a major release.
Author: Ajin Cherian, based on idea by Andres Freund
Reviewed-by: Amit Kapila and Vignesh C
Discussion: https://postgr.es/m/d0f60d60-133d-bf8d-bd70-47784d8fabf3@enterprisedb.com
Commit a271a1b50e introduced decoding at prepare time in ReorderBuffer.
This can lead to deadlock for out-of-core logical replication solutions
that uses this feature to build distributed 2PC in case such transactions
lock [user] catalog tables exclusively. They need to inform users to not
have locks on catalog tables (via explicit LOCK command) in such
transactions.
Reported-by: Andres Freund
Discussion: https://postgr.es/m/20210222222847.tpnb6eg3yiykzpky@alap3.anarazel.de
We don't have anything to decode in a transaction if ReorderBufferTXN
doesn't exist by the time we decode the commit prepared. So don't create a
new ReorderBufferTXN here. This is an oversight in commit a271a1b5.
Reported-by: Markus Wanner
Discussion: https://postgr.es/m/dbec82e2-dbd7-95a2-c6b6-e488cbbdf853@bluegap.ch
In 955a684e04 we fixed a bug in initial snapshot creation. In the
course of which several members of struct SnapBuild were obsoleted. As
SnapBuild is serialized to disk we couldn't change the memory layout.
Unfortunately I subsequently forgot about removing the backward compat
gunk, but luckily Heikki just reminded me.
This commit bumps SNAPBUILD_VERSION, therefore breaking existing
slots (which is fine in a major release).
Author: Andres Freund
Reminded-By: Heikki Linnakangas <hlinnaka@iki.fi>
Discussion: https://postgr.es/m/c94be044-818f-15e3-1ad3-7a7ae2dfed0a@iki.fi
For the initial table data synchronization in logical replication, we use
a single transaction to copy the entire table and then synchronize the
position in the stream with the main apply worker.
There are multiple downsides of this approach: (a) We have to perform the
entire copy operation again if there is any error (network breakdown,
error in the database operation, etc.) while we synchronize the WAL
position between tablesync worker and apply worker; this will be onerous
especially for large copies, (b) Using a single transaction in the
synchronization-phase (where we can receive WAL from multiple
transactions) will have the risk of exceeding the CID limit, (c) The slot
will hold the WAL till the entire sync is complete because we never commit
till the end.
This patch solves all the above downsides by allowing multiple
transactions during the tablesync phase. The initial copy is done in a
single transaction and after that, we commit each transaction as we
receive. To allow recovery after any error or crash, we use a permanent
slot and origin to track the progress. The slot and origin will be removed
once we finish the synchronization of the table. We also remove slot and
origin of tablesync workers if the user performs DROP SUBSCRIPTION .. or
ALTER SUBSCRIPTION .. REFERESH and some of the table syncs are still not
finished.
The commands ALTER SUBSCRIPTION ... REFRESH PUBLICATION and
ALTER SUBSCRIPTION ... SET PUBLICATION ... with refresh option as true
cannot be executed inside a transaction block because they can now drop
the slots for which we have no provision to rollback.
This will also open up the path for logical replication of 2PC
transactions on the subscriber side. Previously, we can't do that because
of the requirement of maintaining a single transaction in tablesync
workers.
Bump catalog version due to change of state in the catalog
(pg_subscription_rel).
Author: Peter Smith, Amit Kapila, and Takamichi Osumi
Reviewed-by: Ajin Cherian, Petr Jelinek, Hou Zhijie and Amit Kapila
Discussion: https://postgr.es/m/CAA4eK1KHJxaZS-fod-0fey=0tq3=Gkn4ho=8N4-5HWiCfu0H1A@mail.gmail.com
Currently, we get the origin id from the name and then drop the origin by
taking ExclusiveLock on ReplicationOriginRelationId. So, two concurrent
sessions can get the id from the name at the same time and then when they
try to drop the origin, one of the sessions will get the either
"tuple concurrently deleted" or "cache lookup failed for replication
origin ..".
To prevent this race condition we do the entire operation under lock. This
obviates the need for replorigin_drop() API and we have removed it so if
any extension authors are using it they need to instead use
replorigin_drop_by_name. See it's usage in pg_replication_origin_drop().
Author: Peter Smith
Reviewed-by: Amit Kapila, Euler Taveira, Petr Jelinek, and Alvaro
Herrera
Discussion: https://www.postgresql.org/message-id/CAHut%2BPuW8DWV5fskkMWWMqzt-x7RPcNQOtJQBp6SdwyRghCk7A%40mail.gmail.com
Up until now, we've held this lock when performing a checkpoint or
restartpoint, but commit 076a055acf back
in 2004 and commit 7e48b77b1c from 2009,
taken together, have removed all need for this. In the present code,
there's only ever one process entitled to attempt a checkpoint: either
the checkpointer, during normal operation, or the postmaster, during
single-user operation. So, we don't need the lock.
One possible concern in making this change is that it means that
a substantial amount of code where HOLD_INTERRUPTS() was previously
in effect due to the preceding LWLockAcquire() will now be
running without that. This could mean that ProcessInterrupts()
gets called in places from which it didn't before. However, this
seems unlikely to do very much, because the checkpointer doesn't
have any signal mapped to die(), so it's not clear how,
for example, ProcDiePending = true could happen in the first
place. Similarly with ClientConnectionLost and recovery conflicts.
Also, if there are any such problems, we might want to fix them
rather than reverting this, since running lots of code with
interrupt handling suspended is generally bad.
Patch by me, per an inquiry by Amul Sul. Review by Tom Lane
and Michael Paquier.
Discussion: http://postgr.es/m/CAAJ_b97XnBBfYeSREDJorFsyoD1sHgqnNuCi=02mNQBUMnA=FA@mail.gmail.com
Add an executor aminsert() hint mechanism that informs index AMs that
the incoming index tuple (the tuple that accompanies the hint) is not
being inserted by execution of an SQL statement that logically modifies
any of the index's key columns.
The hint is received by indexes when an UPDATE takes place that does not
apply an optimization like heapam's HOT (though only for indexes where
all key columns are logically unchanged). Any index tuple that receives
the hint on insert is expected to be a duplicate of at least one
existing older version that is needed for the same logical row. Related
versions will typically be stored on the same index page, at least
within index AMs that apply the hint.
Recognizing the difference between MVCC version churn duplicates and
true logical row duplicates at the index AM level can help with cleanup
of garbage index tuples. Cleanup can intelligently target tuples that
are likely to be garbage, without wasting too many cycles on less
promising tuples/pages (index pages with little or no version churn).
This is infrastructure for an upcoming commit that will teach nbtree to
perform bottom-up index deletion. No index AM actually applies the hint
just yet.
Author: Peter Geoghegan <pg@bowt.ie>
Reviewed-By: Victor Yegorov <vyegorov@gmail.com>
Discussion: https://postgr.es/m/CAH2-Wz=CEKFa74EScx_hFVshCOn6AA5T-ajFASTdzipdkLTNQQ@mail.gmail.com
This patch allows PREPARE-time decoding of two-phase transactions (if the
output plugin supports this capability), in which case the transactions
are replayed at PREPARE and then committed later when COMMIT PREPARED
arrives.
Now that we decode the changes before the commit, the concurrent aborts
may cause failures when the output plugin consults catalogs (both system
and user-defined).
We detect such failures with a special sqlerrcode
ERRCODE_TRANSACTION_ROLLBACK introduced by commit 7259736a6e and stop
decoding the remaining changes. Then we rollback the changes when rollback
prepared is encountered.
Author: Ajin Cherian and Amit Kapila based on previous work by Nikhil Sontakke and Stas Kelvich
Reviewed-by: Amit Kapila, Peter Smith, Sawada Masahiko, Arseny Sher, and Dilip Kumar
Tested-by: Takamichi Osumi
Discussion:
https://postgr.es/m/02DA5F5E-CECE-4D9C-8B4B-418077E2C010@postgrespro.ruhttps://postgr.es/m/CAMGcDxeqEpWj3fTXwqhSwBdXd2RS9jzwWscO-XbeCfso6ts3+Q@mail.gmail.com
This adds six methods to the output plugin API, adding support for
streaming changes of two-phase transactions at prepare time.
* begin_prepare
* filter_prepare
* prepare
* commit_prepared
* rollback_prepared
* stream_prepare
Most of this is a simple extension of the existing methods, with the
semantic difference that the transaction is not yet committed and maybe
aborted later.
Until now two-phase transactions were translated into regular transactions
on the subscriber, and the GID was not forwarded to it. None of the
two-phase commands were communicated to the subscriber.
This patch provides the infrastructure for logical decoding plugins to be
informed of two-phase commands Like PREPARE TRANSACTION, COMMIT PREPARED
and ROLLBACK PREPARED commands with the corresponding GID.
This also extends the 'test_decoding' plugin, implementing these new
methods.
This commit simply adds these new APIs and the upcoming patch to "allow
the decoding at prepare time in ReorderBuffer" will use these APIs.
Author: Ajin Cherian and Amit Kapila based on previous work by Nikhil Sontakke and Stas Kelvich
Reviewed-by: Amit Kapila, Peter Smith, Sawada Masahiko, and Dilip Kumar
Discussion:
https://postgr.es/m/02DA5F5E-CECE-4D9C-8B4B-418077E2C010@postgrespro.ruhttps://postgr.es/m/CAMGcDxeqEpWj3fTXwqhSwBdXd2RS9jzwWscO-XbeCfso6ts3+Q@mail.gmail.com
The comments atop of this function describes behaviour in case of a
transactional WAL message only, but it accepts both transactional and
non-transactional WAL messages. Update the comments to describe
behaviour in case of non-transactional WAL message as well.
Ashutosh Bapat, rephrased by Amit Kapila
Discussion: https://postgr.es/m/CAGEoWWTTzNzHOi8bj0wfAo1siGi-YEh6wqH1oaz4DrkTJ6HbTQ@mail.gmail.com
Invent a new flag bit HASH_STRINGS to specify C-string hashing, which
was formerly the default; and add assertions insisting that exactly
one of the bits HASH_STRINGS, HASH_BLOBS, and HASH_FUNCTION be set.
This is in hopes of preventing recurrences of the type of oversight
fixed in commit a1b8aa1e4 (i.e., mistakenly omitting HASH_BLOBS).
Also, when HASH_STRINGS is specified, insist that the keysize be
more than 8 bytes. This is a heuristic, but it should catch
accidental use of HASH_STRINGS for integer or pointer keys.
(Nearly all existing use-cases set the keysize to NAMEDATALEN or
more, so there's little reason to think this restriction should
be problematic.)
Tweak hash_create() to insist that the HASH_ELEM flag be set, and
remove the defaults it had for keysize and entrysize. Since those
defaults were undocumented and basically useless, no callers
omitted HASH_ELEM anyway.
Also, remove memset's zeroing the HASHCTL parameter struct from
those callers that had one. This has never been really necessary,
and while it wasn't a bad coding convention it was confusing that
some callers did it and some did not. We might as well save a few
cycles by standardizing on "not".
Also improve the documentation for hash_create().
In passing, improve reinit.c's usage of a hash table by storing
the key as a binary Oid rather than a string; and, since that's
a temporary hash table, allocate it in CurrentMemoryContext for
neatness.
Discussion: https://postgr.es/m/590625.1607878171@sss.pgh.pa.us
User-visible log messages should go through ereport(), so they are
subject to translation. Many remaining elog(LOG) calls are really
debugging calls.
Reviewed-by: Alvaro Herrera <alvherre@alvh.no-ip.org>
Reviewed-by: Michael Paquier <michael@paquier.xyz>
Reviewed-by: Noah Misch <noah@leadboat.com>
Discussion: https://www.postgresql.org/message-id/flat/92d6f545-5102-65d8-3c87-489f71ea0a37%40enterprisedb.com
We start recording changes in ReorderBufferTXN even before we reach
SNAPBUILD_CONSISTENT state so that if the commit is encountered after
reaching that we should be able to send the changes of the entire transaction.
Now, while recording changes if the reorder buffer memory has exceeded
logical_decoding_work_mem then we can start streaming if it is allowed and
we haven't yet streamed that data. However, we must not allow streaming to
start unless the snapshot has reached SNAPBUILD_CONSISTENT state.
In passing, improve the comments atop ReorderBufferResetTXN to mention the
case when we need to continue streaming after getting an error.
Author: Amit Kapila
Reviewed-by: Dilip Kumar
Discussion: https://postgr.es/m/CAA4eK1KoOH0byboyYY40NBcC7Fe812trwTa+WY3jQF7WQWZbQg@mail.gmail.com
Tablesync worker runs under a single transaction but in streaming mode, we
were committing the transaction on stream_stop, stream_abort, and
stream_commit. We need to avoid committing the transaction in a streaming
mode in tablesync worker.
In passing move the call to process_syncing_tables in
apply_handle_stream_commit after clean up of stream files. This will
allow clean up of files to happen before the exit of tablesync worker
which would otherwise be handled by one of the proc exit routines.
Author: Dilip Kumar
Reviewed-by: Amit Kapila and Peter Smith
Tested-by: Peter Smith
Discussion: https://postgr.es/m/CAHut+Pt4PyKQCwqzQ=EFF=bpKKJD7XKt_S23F6L20ayQNxg77A@mail.gmail.com
Reverts 27838981be (some comments are kept). Per discussion, it does
not seem safe to relax the lock level used for this; in order for it to
be safe, there would have to be memory barriers between the point we set
the flag and the point we set the trasaction Xid, which perhaps would
not be so bad; but there would also have to be barriers at the readers'
side, which from a performance perspective might be bad.
Now maybe this analysis is wrong and it *is* safe for some reason, but
proof of that is not trivial.
Discussion: https://postgr.es/m/20201118190928.vnztes7c2sldu43a@alap3.anarazel.de
Copy.c has grown really large. Split it into more manageable parts:
- copy.c now contains only a few functions that are common to COPY FROM
and COPY TO.
- copyto.c contains code for COPY TO.
- copyfrom.c contains code for initializing COPY FROM, and inserting the
tuples to the correct table.
- copyfromparse.c contains code for reading from the client/file/program,
and parsing the input text/CSV/binary format into tuples.
All of these parts are fairly complicated, and fairly independent of each
other. There is a patch being discussed to implement parallel COPY FROM,
which will add a lot of new code to the COPY FROM path, and another patch
which would allow INSERTs to use the same multi-insert machinery as COPY
FROM, both of which will require refactoring that code. With those two
patches, there's going to be a lot of code churn in copy.c anyway, so now
seems like a good time to do this refactoring.
The CopyStateData struct is also split. All the formatting options, like
FORMAT, QUOTE, ESCAPE, are put in a new CopyFormatOption struct, which
is used by both COPY FROM and TO. Other state data are kept in separate
CopyFromStateData and CopyToStateData structs.
Reviewed-by: Soumyadeep Chakraborty, Erik Rijkers, Vignesh C, Andres Freund
Discussion: https://www.postgresql.org/message-id/8e15b560-f387-7acc-ac90-763986617bfb%40iki.fi
We don't actually need a lock to set PGPROC->statusFlags itself; what we
do need is a shared lock on either XidGenLock or ProcArrayLock in order to
ensure MyProc->pgxactoff keeps still while we modify the mirror array in
ProcGlobal->statusFlags. Some places were using an exclusive lock for
that, which is excessive. Relax those to use shared lock only.
procarray.c has a couple of places with somewhat brittle assumptions
about PGPROC changes: ProcArrayEndTransaction uses only shared lock, so
it's permissible to change MyProc only. On the other hand,
ProcArrayEndTransactionInternal also changes other procs, so it must
hold exclusive lock. Add asserts to ensure those assumptions continue
to hold.
Author: Álvaro Herrera <alvherre@alvh.no-ip.org>
Reviewed-by: Michael Paquier <michael@paquier.xyz>
Discussion: https://postgr.es/m/20201117155501.GA13805@alvherre.pgsql
In streaming mode, the transaction can be decoded in multiple streams and
those streams can be interleaved with streams of other transactions. So,
we can't remember the transaction's write status in the logical decoding
context because that might get changed due to some other transactions and
lead to wrong answers for 'skip-empty-xacts' option. We decided to keep
each transaction's write status in the ReorderBufferTxn to avoid
interleaved streams changing the status of some unrelated transactions.
Diagnosed-by: Amit Kapila
Author: Dilip Kumar
Reviewed-by: Amit Kapila
Discussion: https://postgr.es/m/CAA4eK1LR7=XNM_TLmpZMFuV8ZQpoxkem--NZJYf8YXmesbvwLA@mail.gmail.com
Move the system catalog index declarations from catalog/indexing.h to
the respective parent tables' catalog/pg_*.h files. The original
reason for having it split was that the old genbki system produced the
output in the order of the catalog files it read, so all the indexing
stuff needed to come separately. But this is no longer the case, and
keeping it together makes more sense.
Reviewed-by: John Naylor <john.naylor@enterprisedb.com>
Discussion: https://www.postgresql.org/message-id/flat/c7cc82d6-f976-75d6-2e3e-b03d2cab26bb@2ndquadrant.com
Logical replication protocol uses a single byte character to identify a
message type in logical replication protocol. The code uses string
literals for the same. Use Enum so that
1. All the string literals used can be found at a single place. This
makes it easy to add more types without the risk of conflicts.
2. It's easy to locate the code handling a given message type.
3. When used with switch statements, it is easy to identify the missing
cases using -Wswitch.
Author: Ashutosh Bapat
Reviewed-by: Kyotaro Horiguchi, Andres Freund, Peter Smith and Amit Kapila
Discussion: https://postgr.es/m/CAExHW5uPzQ7L0oAd_ENyvaiYMOPgkrAoJpE+ZY5-obdcVT6NPg@mail.gmail.com
This adds the statistics about transactions streamed to the decoding
output plugin from ReorderBuffer. Users can query the
pg_stat_replication_slots view to check these stats and call
pg_stat_reset_replication_slot to reset the stats of a particular slot.
Users can pass NULL in pg_stat_reset_replication_slot to reset stats of
all the slots.
Commit 9868167500 has added the basic infrastructure to capture the stats
of slot and this commit extends the statistics collector to track
additional information about slots.
Bump the catversion as we have added new columns in the catalog entry.
Author: Ajin Cherian and Amit Kapila
Reviewed-by: Sawada Masahiko and Dilip Kumar
Discussion: https://postgr.es/m/CAA4eK1+chpEomLzgSoky-D31qev19AmECNiEAietPQUGEFhtVA@mail.gmail.com
It's unsafe to do this at parse time because addition of generated
columns to a table would not invalidate stored rules containing
UPDATEs on the table ... but there might now be dependent generated
columns that were not there when the rule was made. This also fixes
an oversight that rewriteTargetView failed to update extraUpdatedCols
when transforming an UPDATE on an updatable view. (Since the new
calculation is downstream of that, rewriteTargetView doesn't actually
need to do anything; but before, there was a demonstrable bug there.)
In v13 and HEAD, this leads to easily-visible bugs because (since
commit c6679e4fc) we won't recalculate generated columns that aren't
listed in extraUpdatedCols. In v12 this bitmap is mostly just used
for trigger-firing decisions, so you'd only notice a problem if a
trigger cared whether a generated column had been updated.
I'd complained about this back in May, but then forgot about it
until bug #16671 from Michael Paul Killian revived the issue.
Back-patch to v12 where this field was introduced. If existing
stored rules contain any extraUpdatedCols values, they'll be
ignored because the rewriter will overwrite them, so the bug will
be fixed even for existing rules. (But note that if someone were
to update to 13.1 or 12.5, store some rules with UPDATEs on tables
having generated columns, and then downgrade to a prior minor version,
they might observe issues similar to what this patch fixes. That
seems unlikely enough to not be worth going to a lot of effort to fix.)
Discussion: https://postgr.es/m/10206.1588964727@sss.pgh.pa.us
Discussion: https://postgr.es/m/16671-2fa55851859fb166@postgresql.org
Most importantly, remove optimization in LogicalRepSyncTableStart that
skips the normal walrcv_startstreaming/endstreaming dance. The
optimization is not critically important for production uses anyway,
since it only fires in cases with no activity, and saves an
uninteresting amount of work even then. Critically, it obscures bugs by
hiding the interesting code path from test cases.
Also: in GetSubscriptionRelState, remove pointless relation open; access
pg_subscription_rel->srsubstate with GETSTRUCT as is typical rather than
SysCacheGetAttr; remove unused 'missing_ok' argument.
In wait_for_relation_state_change, use explicit catalog snapshot
invalidation rather than obscurely (and expensively) through
GetLatestSnapshot.
In various places: sprinkle comments more liberally and rewrite a number
of them. Other cosmetic code improvements.
No backpatch, since no bug is being fixed here.
Author: Álvaro Herrera <alvherre@alvh.no-ip.org>
Reviewed-by: Petr Jelínek <petr.jelinek@2ndquadrant.com>
Discussion: https://postgr.es/m/20201010190637.GA5774@alvherre.pgsql
A number of places were using appendStringInfo() when they could have been
using appendStringInfoString() instead. While there's no functionality
change there, it's just more efficient to use appendStringInfoString()
when no formatting is required. Likewise for some
appendStringInfoString() calls which were just appending a single char.
We can just use appendStringInfoChar() for that.
Additionally, many places were using appendPQExpBuffer() when they could
have used appendPQExpBufferStr(). Change those too.
Patch by Zhijie Hou, but further searching by me found significantly more
places that deserved the same treatment.
Author: Zhijie Hou, David Rowley
Discussion: https://postgr.es/m/cb172cf4361e4c7ba7167429070979d4@G08CNEXMBPEKD05.g08.fujitsu.local
during logical decoding.
Prior to commit c55040ccd0 we have no way of knowing the invalidations
before commit. So, while decoding we use to execute all the invalidations
at each command end as we had no way of knowing which invalidations
happened before that command. Due to this, transactions involving large
amounts of DDLs use to take more time and also lead to high CPU usage. But
now we know specific invalidations at each command end so we execute only
required invalidations.
It has been observed that decoding of a transaction containing truncation
of a table with 1000 partitions would be finished in 1s whereas before
this patch it used to take 4-5 minutes.
Author: Dilip Kumar
Reviewed-by: Amit Kapila and Keisuke Kuroda
Discussion: https://postgr.es/m/CANDwggKYveEtXjXjqHA6RL3AKSHMsQyfRY6bK+NqhAWJyw8psQ@mail.gmail.com
I removed the duplicate command tags for START_REPLICATION inadvertently
in commit 07082b08cc, but the replication protocol requires them. The
fact that the replication protocol was broken was not noticed because
all our test cases use an optimized code path that exits early, failing
to verify that the behavior is correct for non-optimized cases. Put
them back.
Also document this protocol quirk.
Add a test case that shows the failure. It might still succeed even
without the patch when run on a fast enough server, but it suffices to
show the bug in enough cases that it would be noticed in buildfarm.
Author: Álvaro Herrera <alvherre@alvh.no-ip.org>
Reported-by: Henry Hinze <henry.hinze@gmail.com>
Reviewed-by: Petr Jelínek <petr.jelinek@2ndquadrant.com>
Discussion: https://postgr.es/m/16643-eaadeb2a1a58d28c@postgresql.org
Maintaining 'es_result_relation_info' correctly at all times has become
cumbersome, especially with partitioning where each partition gets its
own result relation info. Having to set and reset it across arbitrary
operations has caused bugs in the past.
This changes all the places that used 'es_result_relation_info', to
receive the currently active ResultRelInfo via function parameters
instead.
Author: Amit Langote
Discussion: https://www.postgresql.org/message-id/CA%2BHiwqGEmiib8FLiHMhKB%2BCH5dRgHSLc5N5wnvc4kym%2BZYpQEQ%40mail.gmail.com
Instead of allocating all the ResultRelInfos upfront in one big array,
allocate them in ExecInitModifyTable(). es_result_relations is now an
array of ResultRelInfo pointers, rather than an array of structs, and it
is indexed by the RT index.
This simplifies things: we get rid of the separate concept of a "result
rel index", and don't need to set it in setrefs.c anymore. This also
allows follow-up optimizations (not included in this commit yet) to skip
initializing ResultRelInfos for target relations that were not needed at
runtime, and removal of the es_result_relation_info pointer.
The EState arrays of regular result rels and root result rels are merged
into one array. Similarly, the resultRelations and rootResultRelations
lists in PlannedStmt are merged into one. It's not actually clear to me
why they were kept separate in the first place, but now that the
es_result_relations array is indexed by RT index, it certainly seems
pointless.
The PlannedStmt->resultRelations list is now only needed for
ExecRelationIsTargetRelation(). One visible effect of this change is that
ExecRelationIsTargetRelation() will now return 'true' also for the
partition root, if a partitioned table is updated. That seems like a good
thing, although the function isn't used in core code, and I don't see any
reason for an FDW to call it on a partition root.
Author: Amit Langote
Discussion: https://www.postgresql.org/message-id/CA%2BHiwqGEmiib8FLiHMhKB%2BCH5dRgHSLc5N5wnvc4kym%2BZYpQEQ%40mail.gmail.com
This adds the statistics about transactions spilled to disk from
ReorderBuffer. Users can query the pg_stat_replication_slots view to check
these stats and call pg_stat_reset_replication_slot to reset the stats of
a particular slot. Users can pass NULL in pg_stat_reset_replication_slot
to reset stats of all the slots.
This commit extends the statistics collector to track this information
about slots.
Author: Sawada Masahiko and Amit Kapila
Reviewed-by: Amit Kapila and Dilip Kumar
Discussion: https://postgr.es/m/CA+fd4k5_pPAYRTDrO2PbtTOe0eHQpBvuqmCr8ic39uTNmR49Eg@mail.gmail.com
In logical replication when a subscriber is missing some columns, it
currently emits an error message that says "some" columns are missing, but
it doesn't specify the missing column names. Change that to display
missing column names which makes an error to be more informative to the
user.
We have decided not to backpatch this commit as this is a minor usability
improvement and no user has reported this.
Reported-by: Bharath Rupireddy
Author: Bharath Rupireddy
Reviewed-by: Kyotaro Horiguchi and Amit Kapila
Discussion: https://postgr.es/m/CALj2ACVkW-EXH_4pmBK8tNeHRz5ksUC4WddGactuCjPiBch-cg@mail.gmail.com
Commit 464824323e changed the logical replication protocol to allow the
streaming of in-progress transactions and used the new version of protocol
irrespective of the server version. Use the appropriate version of the
protocol based on the server version.
Reported-by: Ashutosh Sharma
Author: Dilip Kumar
Reviewed-by: Ashutosh Sharma and Amit Kapila
Discussion: https://postgr.es/m/CAE9k0P=9OpXcNrcU5Gsvd5MZ8GFpiN833vNHzX6Uc=8+h1ft1Q@mail.gmail.com
Since there is only one place that actually needs the partition check
expression, namely ExecPartitionCheck, it's better to fetch it from
the relcache there. In this way we will never fetch it at all if
the query never has use for it, and we still fetch it just once when
we do need it.
The reason for taking an interest in this is that if the relcache
doesn't already have the check expression cached, fetching it
requires obtaining AccessShareLock on the partition root. That
means that operations that look like they should only touch the
partition itself will also take a lock on the root. In particular
we observed that TRUNCATE on a partition may take a lock on the
partition's root, contributing to a deadlock situation in parallel
pg_restore.
As written, this patch does have a small cost, which is that we
are microscopically reducing efficiency for the case where a partition
has an empty check expression. ExecPartitionCheck will be called,
and will go through the motions of setting up and checking an empty
qual, where before it would not have been called at all. We could
avoid that by adding a separate boolean flag to track whether there
is a partition expression to test. However, this case only arises
for a default partition with no siblings, which surely is not an
interesting case in practice. Hence adding complexity for it
does not seem like a good trade-off.
Amit Langote, per a suggestion by me
Discussion: https://postgr.es/m/VI1PR03MB31670CA1BD9625C3A8C5DD05EB230@VI1PR03MB3167.eurprd03.prod.outlook.com
The code recorded cache invalidation events by zeroing the "localreloid"
field of affected cache entries. However, it's possible for an inval
event to occur even while we have the entry open and locked. So an
ill-timed inval could result in "cache lookup failed for relation 0"
errors, if the worker's code tried to use the cleared field. We can
fix that by creating a separate bool field to record whether the entry
needs to be revalidated. (In the back branches, cram the bool into
what had been padding space, to avoid an ABI break in the somewhat
unlikely event that any extension is looking at this struct.)
Also, rearrange the logic in logicalrep_rel_open so that it
does the right thing in cases where table_open would fail.
We should retry the lookup by name in that case, but we didn't.
The real-world impact of this is probably small. In the first place,
the error conditions are very low probability, and in the second place,
the worker would just exit and get restarted. We only noticed because
in a CLOBBER_CACHE_ALWAYS build, the failure can occur repeatedly,
preventing the worker from making progress. Nonetheless, it's clearly
a bug, and it impedes a useful type of testing; so back-patch to v10
where this code was introduced.
Discussion: https://postgr.es/m/1032727.1600096803@sss.pgh.pa.us
Commit 3f60f690f only partially fixed the broken-status-tracking
issue in LogicalRepApplyLoop: we need ping_sent to have the same
lifetime as last_recv_timestamp. The effects are much less serious
than what that commit fixed, though. AFAICS this would just lead to
extra ping requests being sent, once per second until the sender
responds. Still, it's a bug, so backpatch to v10 as before.
Discussion: https://postgr.es/m/959627.1599248476@sss.pgh.pa.us
To add support for streaming of in-progress transactions into the
built-in logical replication, we need to do three things:
* Extend the logical replication protocol, so identify in-progress
transactions, and allow adding additional bits of information (e.g.
XID of subtransactions).
* Modify the output plugin (pgoutput) to implement the new stream
API callbacks, by leveraging the extended replication protocol.
* Modify the replication apply worker, to properly handle streamed
in-progress transaction by spilling the data to disk and then
replaying them on commit.
We however must explicitly disable streaming replication during
replication slot creation, even if the plugin supports it. We
don't need to replicate the changes accumulated during this phase,
and moreover we don't have a replication connection open so we
don't have where to send the data anyway.
Author: Tomas Vondra, Dilip Kumar and Amit Kapila
Reviewed-by: Amit Kapila, Kuntal Ghosh and Ajin Cherian
Tested-by: Neha Sharma, Mahendra Singh Thalor and Ajin Cherian
Discussion: https://postgr.es/m/688b0b7f-2f6c-d827-c27b-216a8e3ea700@2ndquadrant.com
Previous commits made it faster/more scalable to compute snapshots. But not
building a snapshot is still faster. Now that GetSnapshotData() does not
maintain RecentGlobal* anymore, that is actually not too hard:
This commit introduces xactCompletionCount, which tracks the number of
top-level transactions with xids (i.e. which may have modified the database)
that completed in some form since the start of the server.
We can avoid rebuilding the snapshot's contents whenever the current
xactCompletionCount is the same as it was when the snapshot was
originally built. Currently this check happens while holding
ProcArrayLock. While it's likely possible to perform the check without
acquiring ProcArrayLock, it seems better to do that separately /
later, some careful analysis is required. Even with the lock this is a
significant win on its own.
On a smaller two socket machine this gains another ~1.03x, on a larger
machine the effect is roughly double (earlier patch version tested
though). If we were able to safely avoid the lock there'd be another
significant gain on top of that.
Author: Andres Freund <andres@anarazel.de>
Reviewed-By: Robert Haas <robertmhaas@gmail.com>
Reviewed-By: Thomas Munro <thomas.munro@gmail.com>
Reviewed-By: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
Similar to the previous commit this increases the chance that data
frequently needed by GetSnapshotData() stays in l2 cache. As we now
take care to not unnecessarily write to ProcGlobal->vacuumFlags, there
should be very few modifications to the ProcGlobal->vacuumFlags array.
Author: Andres Freund <andres@anarazel.de>
Reviewed-By: Robert Haas <robertmhaas@gmail.com>
Reviewed-By: Thomas Munro <thomas.munro@gmail.com>
Reviewed-By: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
Now that xmin isn't needed for GetSnapshotData() anymore, it leads to
unnecessary cacheline ping-pong to have it in PGXACT, as it is updated
considerably more frequently than the other PGXACT members.
After the changes in dc7420c2c9, this is a very straight-forward change.
For highly concurrent, snapshot acquisition heavy, workloads this change alone
can significantly increase scalability. E.g. plain pgbench on a smaller 2
socket machine gains 1.07x for read-only pgbench, 1.22x for read-only pgbench
when submitting queries in batches of 100, and 2.85x for batches of 100
'SELECT';. The latter numbers are obviously not to be expected in the
real-world, but micro-benchmark the snapshot computation
scalability (previously spending ~80% of the time in GetSnapshotData()).
Author: Andres Freund <andres@anarazel.de>
Reviewed-By: Robert Haas <robertmhaas@gmail.com>
Reviewed-By: Thomas Munro <thomas.munro@gmail.com>
Reviewed-By: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
To make GetSnapshotData() more scalable, it cannot not look at at each proc's
xmin: While snapshot contents do not need to change whenever a read-only
transaction commits or a snapshot is released, a proc's xmin is modified in
those cases. The frequency of xmin modifications leads to, particularly on
higher core count systems, many cache misses inside GetSnapshotData(), despite
the data underlying a snapshot not changing. That is the most
significant source of GetSnapshotData() scaling poorly on larger systems.
Without accessing xmins, GetSnapshotData() cannot calculate accurate horizons /
thresholds as it has so far. But we don't really have to: The horizons don't
actually change that much between GetSnapshotData() calls. Nor are the horizons
actually used every time a snapshot is built.
The trick this commit introduces is to delay computation of accurate horizons
until there use and using horizon boundaries to determine whether accurate
horizons need to be computed.
The use of RecentGlobal[Data]Xmin to decide whether a row version could be
removed has been replaces with new GlobalVisTest* functions. These use two
thresholds to determine whether a row can be pruned:
1) definitely_needed, indicating that rows deleted by XIDs >= definitely_needed
are definitely still visible.
2) maybe_needed, indicating that rows deleted by XIDs < maybe_needed can
definitely be removed
GetSnapshotData() updates definitely_needed to be the xmin of the computed
snapshot.
When testing whether a row can be removed (with GlobalVisTestIsRemovableXid())
and the tested XID falls in between the two (i.e. XID >= maybe_needed && XID <
definitely_needed) the boundaries can be recomputed to be more accurate. As it
is not cheap to compute accurate boundaries, we limit the number of times that
happens in short succession. As the boundaries used by
GlobalVisTestIsRemovableXid() are never reset (with maybe_needed updated by
GetSnapshotData()), it is likely that further test can benefit from an earlier
computation of accurate horizons.
To avoid regressing performance when old_snapshot_threshold is set (as that
requires an accurate horizon to be computed), heap_page_prune_opt() doesn't
unconditionally call TransactionIdLimitedForOldSnapshots() anymore. Both the
computation of the limited horizon, and the triggering of errors (with
SetOldSnapshotThresholdTimestamp()) is now only done when necessary to remove
tuples.
This commit just removes the accesses to PGXACT->xmin from
GetSnapshotData(), but other members of PGXACT residing in the same
cache line are accessed. Therefore this in itself does not result in a
significant improvement. Subsequent commits will take advantage of the
fact that GetSnapshotData() now does not need to access xmins anymore.
Note: This contains a workaround in heap_page_prune_opt() to keep the
snapshot_too_old tests working. While that workaround is ugly, the tests
currently are not meaningful, and it seems best to address them separately.
Author: Andres Freund <andres@anarazel.de>
Reviewed-By: Robert Haas <robertmhaas@gmail.com>
Reviewed-By: Thomas Munro <thomas.munro@gmail.com>
Reviewed-By: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
They are equivalent, except that StrNCpy() zero-fills the entire
destination buffer instead of providing just one trailing zero. For
all but a tiny number of callers, that's just overhead rather than
being desirable.
Remove StrNCpy() as it is now unused.
In some cases, namestrcpy() is the more appropriate function to use.
While we're here, simplify the API of namestrcpy(): Remove the return
value, don't check for NULL input. Nothing was using that anyway.
Also, remove a few unused name-related functions.
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://www.postgresql.org/message-id/flat/44f5e198-36f6-6cdb-7fa9-60e34784daae%402ndquadrant.com
This is like CVE-2018-1058 commit
582edc369c. Today, a malicious user of a
publisher or subscriber database can invoke arbitrary SQL functions
under an identity running replication, often a superuser. This fix may
cause "does not exist" or "no schema has been selected to create in"
errors in a replication process. After upgrading, consider watching
server logs for these errors. Objects accruing schema qualification in
the wake of the earlier commit are unlikely to need further correction.
Back-patch to v10, which introduced logical replication.
Security: CVE-2020-14349
Testing that an unsigned variable is >= 0 is pretty pointless,
as noted by Coverity and numerous buildfarm members.
In passing, add comment about new uses of "volatile" --- Coverity
doesn't much like that either, but it seems probably necessary.
Instead of serializing the transaction to disk after reaching the
logical_decoding_work_mem limit in memory, we consume the changes we have
in memory and invoke stream API methods added by commit 45fdc9738b.
However, sometimes if we have incomplete toast or speculative insert we
spill to the disk because we can't generate the complete tuple and stream.
And, as soon as we get the complete tuple we stream the transaction
including the serialized changes.
We can do this incremental processing thanks to having assignments
(associating subxact with toplevel xacts) in WAL right away, and
thanks to logging the invalidation messages at each command end. These
features are added by commits 0bead9af48 and c55040ccd0 respectively.
Now that we can stream in-progress transactions, the concurrent aborts
may cause failures when the output plugin consults catalogs (both system
and user-defined).
We handle such failures by returning ERRCODE_TRANSACTION_ROLLBACK
sqlerrcode from system table scan APIs to the backend or WALSender
decoding a specific uncommitted transaction. The decoding logic on the
receipt of such a sqlerrcode aborts the decoding of the current
transaction and continue with the decoding of other transactions.
We have ReorderBufferTXN pointer in each ReorderBufferChange by which we
know which xact it belongs to. The output plugin can use this to decide
which changes to discard in case of stream_abort_cb (e.g. when a subxact
gets discarded).
We also provide a new option via SQL APIs to fetch the changes being
streamed.
Author: Dilip Kumar, Tomas Vondra, Amit Kapila, Nikhil Sontakke
Reviewed-by: Amit Kapila, Kuntal Ghosh, Ajin Cherian
Tested-by: Neha Sharma, Mahendra Singh Thalor and Ajin Cherian
Discussion: https://postgr.es/m/688b0b7f-2f6c-d827-c27b-216a8e3ea700@2ndquadrant.com
This adds seven methods to the output plugin API, adding support for
streaming changes of large in-progress transactions.
* stream_start
* stream_stop
* stream_abort
* stream_commit
* stream_change
* stream_message
* stream_truncate
Most of this is a simple extension of the existing methods, with
the semantic difference that the transaction (or subtransaction)
is incomplete and may be aborted later (which is something the
regular API does not really need to deal with).
This also extends the 'test_decoding' plugin, implementing these
new stream methods.
The stream_start/start_stop are used to demarcate a chunk of changes
streamed for a particular toplevel transaction.
This commit simply adds these new APIs and the upcoming patch to "allow
the streaming mode in ReorderBuffer" will use these APIs.
Author: Tomas Vondra, Dilip Kumar, Amit Kapila
Reviewed-by: Amit Kapila
Tested-by: Neha Sharma and Mahendra Singh Thalor
Discussion: https://postgr.es/m/688b0b7f-2f6c-d827-c27b-216a8e3ea700@2ndquadrant.com
Thomas Munro
Amit Kapila
Fujii Masao
Michael Paquier
Peter Eisentraut
Tom Lane
Peter Geoghegan
Robert Haas
Andres Freund
Bruce Momjian
Noah Misch
Alvaro Herrera
Heikki Linnakangas
David Rowley