synchronous_standby_names cannot be reloaded safely by backends, and the
checkpointer is in charge of updating a state in shared memory if the
GUC is enabled in WalSndCtl, to let the backends know if they should
wait or not for a given LSN. This provides a strict control on the
timing of the waiting queues if the GUC is enabled or disabled, then
reloaded. The checkpointer is also in charge of waking up the backends
that could be waiting for a LSN when the GUC is disabled.
This logic had a race condition at startup, where it would be possible
for backends to not wait for a LSN even if synchronous_standby_names is
enabled. This would cause visibility issues with transactions that we
should be waiting for but they were not. The problem lasts until the
checkpointer does its initial update of the shared memory state when it
loads synchronous_standby_names.
In order to take care of this problem, the shared memory state in
WalSndCtl is extended to detect if it has been initialized by the
checkpointer, and not only check if synchronous_standby_names is
defined. In WalSndCtlData, sync_standbys_defined is renamed to
sync_standbys_status, a bits8 able to know about two states:
- If the shared memory state has been initialized. This flag is set by
the checkpointer at startup once, and never removed.
- If synchronous_standby_names is known as defined in the shared memory
state. This is the same as the previous sync_standbys_defined in
WalSndCtl.
This method gives a way for backends to decide what they should do until
the shared memory area is initialized, and they now ultimately fall back
to a check on the GUC value in this case, which is the best thing that
can be done.
Fortunately, SyncRepUpdateSyncStandbysDefined() is called immediately by
the checkpointer when this process starts, so the window is very narrow.
It is possible to enlarge the problematic window by making the
checkpointer wait at the beginning of SyncRepUpdateSyncStandbysDefined()
with a hardcoded sleep for example, and doing so has showed that a 2PC
visibility test is indeed failing. On machines slow enough, this bug
would cause spurious failures.
In 17~, we have looked at the possibility of adding an injection point
to have a reproducible test, but as the problematic window happens at
early startup, we would need to invent a way to make an injection point
optionally persistent across restarts when attached, something that
would be fine for this case as it would involve the checkpointer. This
issue is quite old, and can be reproduced on all the stable branches.
Author: Melnikov Maksim <m.melnikov@postgrespro.ru>
Co-authored-by: Michael Paquier <michael@paquier.xyz>
Discussion: https://postgr.es/m/163fcbec-900b-4b07-beaa-d2ead8634bec@postgrespro.ru
Backpatch-through: 13
Data loss can happen when the DDLs like ALTER PUBLICATION ... ADD TABLE ...
or ALTER TYPE ... that don't take a strong lock on table happens
concurrently to DMLs on the tables involved in the DDL. This happens
because logical decoding doesn't distribute invalidations to concurrent
transactions and those transactions use stale cache data to decode the
changes. The problem becomes bigger because we keep using the stale cache
even after those in-progress transactions are finished and skip the
changes required to be sent to the client.
This commit fixes the issue by distributing invalidation messages from
catalog-modifying transactions to all concurrent in-progress transactions.
This allows the necessary rebuild of the catalog cache when decoding new
changes after concurrent DDL.
We observed performance regression primarily during frequent execution of
*publication DDL* statements that modify the published tables. The
regression is minor or nearly nonexistent for DDLs that do not affect the
published tables or occur infrequently, making this a worthwhile cost to
resolve a longstanding data loss issue.
An alternative approach considered was to take a strong lock on each
affected table during publication modification. However, this would only
address issues related to publication DDLs (but not the ALTER TYPE ...)
and require locking every relation in the database for publications
created as FOR ALL TABLES, which is impractical.
The bug exists in all supported branches, but we are backpatching till 14.
The fix for 13 requires somewhat bigger changes than this fix, so the fix
for that branch is still under discussion.
Reported-by: hubert depesz lubaczewski <depesz@depesz.com>
Reported-by: Tomas Vondra <tomas.vondra@enterprisedb.com>
Author: Shlok Kyal <shlok.kyal.oss@gmail.com>
Author: Hayato Kuroda <kuroda.hayato@fujitsu.com>
Reviewed-by: Zhijie Hou <houzj.fnst@fujitsu.com>
Reviewed-by: Masahiko Sawada <sawada.mshk@gmail.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Tested-by: Benoit Lobréau <benoit.lobreau@dalibo.com>
Backpatch-through: 14
Discussion: https://postgr.es/m/de52b282-1166-1180-45a2-8d8917ca74c6@enterprisedb.com
Discussion: https://postgr.es/m/CAD21AoAenVqiMjpN-PvGHL1N9DWnHSq673bfgr6phmBUzx=kLQ@mail.gmail.com
The issue happens when building conflict information during apply of
INSERT or UPDATE operations that violate unique constraints on leaf
partitions.
The problem was introduced in commit 9ff68679b5, which removed the
redundant calls to ExecOpenIndices/ExecCloseIndices. The previous code was
relying on the redundant ExecOpenIndices call in
apply_handle_tuple_routing() to build the index information required for
unique key conflict detection.
The fix is to delay building the index information until a conflict is
detected instead of relying on ExecOpenIndices to do the same. The
additional benefit of this approach is that it avoids building index
information when there is no conflict.
Author: Hou Zhijie <houzj.fnst@fujitsu.com>
Reviewed-by:Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Discussion: https://postgr.es/m/TYAPR01MB57244ADA33DDA57119B9D26494A62@TYAPR01MB5724.jpnprd01.prod.outlook.com
WAL senders do not flush their statistics until they exit, limiting the
monitoring possible for live processes. This is penalizing when WAL
senders are running for a long time, like in streaming or logical
replication setups, because it is not possible to know the amount of IO
they generate while running.
This commit makes WAL senders more aggressive with their statistics
flush, using an internal of 1 second, with the flush timing calculated
based on the existing GetCurrentTimestamp() done before the sleeps done
to wait for some activity. Note that the sleep done for logical and
physical WAL senders happens in two different code paths, so the stats
flushes need to happen in these two places.
One test is added for the physical WAL sender case, and one for the
logical WAL sender case. This can be done in a stable fashion by
relying on the WAL generated by the TAP tests in combination with a
stats reset while a server is running, but only on HEAD as WAL data has
been added to pg_stat_io in a051e71e28.
This issue exists since a9c70b46db and the introduction of pg_stat_io,
so backpatch down to v16.
Author: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
Reviewed-by: vignesh C <vignesh21@gmail.com>
Reviewed-by: Xuneng Zhou <xunengzhou@gmail.com>
Discussion: https://postgr.es/m/Z73IsKBceoVd4t55@ip-10-97-1-34.eu-west-3.compute.internal
Backpatch-through: 16
This gets rid of the bespoken ProcessWalRcvInterrupts() function,
which lets walreceiver terminate at any CHECK_FOR_INTERRUPTS() call.
And it's less code anyway.
We can now use the standard libpqsrv_connect_params() libpq wrapper
from libpq-be-fe-helpers.h, removing more code. We attempted to do
that earlier already in commit 728f86fec6, but that was reverted
because it didn't call ProcessWalRcvInterrupts() and therefore didn't
react to shutdown requests. Now that ProcessWalRcvInterrupts() is
gone, it works. As stated in that commit, this also leads to
libpqwalreceiver reserving file descriptors for libpq conncetions,
which is nice.
Author: Andres Freund <andres@anarazel.de> (the earlier commit)
Author: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Reviewed-by: Fujii Masao <masao.fujii@gmail.com>
Reviewed-by: Yura Sokolov <y.sokolov@postgrespro.ru>
Previously, invalidated logical and physical replication slots could
be copied using the pg_copy_logical_replication_slot and
pg_copy_physical_replication_slot functions. Replication slots that
were invalidated for reasons other than WAL removal retained their
restart_lsn. This meant that a new slot copied from an invalidated
slot could have a restart_lsn pointing to a WAL segment that might
have already been removed.
This commit restricts the copying of invalidated replication slots.
Backpatch to v16, where slots could retain their restart_lsn when
invalidated for reasons other than WAL removal.
For v15 and earlier, this check is not required since slots can only
be invalidated due to WAL removal, and existing checks already handle
this issue.
Author: Shlok Kyal <shlok.kyal.oss@gmail.com>
Reviewed-by: vignesh C <vignesh21@gmail.com>
Reviewed-by: Zhijie Hou <houzj.fnst@fujitsu.com>
Reviewed-by: Peter Smith <smithpb2250@gmail.com>
Reviewed-by: Masahiko Sawada <sawada.mshk@gmail.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Discussion: https://postgr.es/m/CANhcyEU65aH0VYnLiu%3DOhNNxhnhNhwcXBeT-jvRe1OiJTo_Ayg%40mail.gmail.com
Backpatch-through: 16
The issue is that the transactions prepared before two-phase decoding is
enabled can fail to replicate to the subscriber after being committed on a
promoted standby following a failover. This is because the two_phase_at
field of a slot, which tracks the LSN from which two-phase decoding
starts, is not synchronized to standby servers. Without two_phase_at, the
logical decoding might incorrectly identify prepared transaction as
already replicated to the subscriber after promotion of standby server,
causing them to be skipped.
To address the issue on HEAD, the two_phase_at field of the slot is
exposed by the pg_replication_slots view and allows the slot
synchronization to copy this value to the corresponding synced slot on the
standby server.
This bug is likely to occur if the user toggles the two_phase option to
true after initial slot creation. Given that altering the two_phase option
of a replication slot is not allowed in PostgreSQL 17, this bug is less
likely to occur. We can't change the view/function definition in
backbranch so we can't push the same fix but we are brainstorming an
appropriate solution for PG17.
Author: Zhijie Hou <houzj.fnst@fujitsu.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Reviewed-by: Masahiko Sawada <sawada.mshk@gmail.com>
Discussion: https://postgr.es/m/TYAPR01MB5724CC7C288535BBCEEE65DA94A72@TYAPR01MB5724.jpnprd01.prod.outlook.com
check_createrole_self_grant and check_synchronized_standby_slots
were allocating memory on a LOG elevel without checking if the
allocation succeeded or not, which would have led to a segfault
on allocation failure.
On top of that, a number of callsites were using the ERROR level,
relying on erroring out rather than returning false to allow the
GUC machinery handle it gracefully. Other callsites used WARNING
instead of LOG. While neither being not wrong, this changes all
check_ functions do it consistently with LOG.
init_custom_variable gets a promoted elevel to FATAL to keep
the guc_malloc error handling in line with the rest of the
error handling in that function which already call FATAL. If
we encounter an OOM in this callsite there is no graceful
handling to be had, better to error out hard.
Backpatch the fix to check_createrole_self_grant down to v16
and the fix to check_synchronized_standby_slots down to v17
where they were introduced.
Author: Daniel Gustafsson <daniel@yesql.se>
Reported-by: Nikita <pm91.arapov@gmail.com>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Bug: #18845
Discussion: https://postgr.es/m/18845-582c6e10247377ec@postgresql.org
Backpatch-through: 16
It seems potentially useful to label our shared libraries with version
information, now that a facility exists for retrieving that. This
patch labels them with the PG_VERSION string. There was some
discussion about using semantic versioning conventions, but that
doesn't seem terribly helpful for modules with no SQL-level presence;
and for those that do have SQL objects, we typically expect them
to support multiple revisions of the SQL definitions, so it'd still
not be very helpful.
I did not label any of src/test/modules/. It seems unnecessary since
we don't install those, and besides there ought to be someplace that
still provides test coverage for the original PG_MODULE_MAGIC macro.
Author: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/dd4d1b59-d0fe-49d5-b28f-1e463b68fa32@gmail.com
Introduce a new conflict type, multiple_unique_conflicts, to handle cases
where an incoming row during logical replication violates multiple UNIQUE
constraints.
Previously, the apply worker detected and reported only the first
encountered key conflict (insert_exists/update_exists), causing repeated
failures as each constraint violation needs to be handled one by one
making the process slow and error-prone.
With this patch, the apply worker checks all unique constraints upfront
once the first key conflict is detected and reports
multiple_unique_conflicts if multiple violations exist. This allows users
to resolve all conflicts at once by deleting all conflicting tuples rather
than dealing with them individually or skipping the transaction.
In the future, this will also allow us to specify different resolution
handlers for such a conflict type.
Add the stats for this conflict type in pg_stat_subscription_stats.
Author: Nisha Moond <nisha.moond412@gmail.com>
Author: Zhijie Hou <houzj.fnst@fujitsu.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Reviewed-by: Peter Smith <smithpb2250@gmail.com>
Reviewed-by: Dilip Kumar <dilipbalaut@gmail.com>
Discussion: https://postgr.es/m/CABdArM7FW-_dnthGkg2s0fy1HhUB8C3ELA0gZX1kkbs1ZZoV3Q@mail.gmail.com
This commit introduces a new GUC option max_active_replication_origins
to control the maximum number of active replication
origins. Previously, this was controlled by
'max_replication_slots'. Having a separate GUC option provides better
flexibility for setting up subscribers, as they may not require
replication slots (for cascading replication) but always require
replication origins.
Author: Euler Taveira <euler@eulerto.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Reviewed-by: Masahiko Sawada <sawada.mshk@gmail.com>
Reviewed-by: Peter Eisentraut <peter@eisentraut.org>
Reviewed-by: vignesh C <vignesh21@gmail.com>
Discussion: https://postgr.es/m/b81db436-8262-4575-b7c4-bc0c1551000b@app.fastmail.com
This commit just does the minimal wiring up of the AIO subsystem, added in the
next commit, to the rest of the system. The next commit contains more details
about motivation and architecture.
This commit is kept separate to make it easier to review, separating the
changes across the tree, from the implementation of the new subsystem.
We discussed squashing this commit with the main commit before merging AIO,
but there has been a mild preference for keeping it separate.
Reviewed-by: Heikki Linnakangas <hlinnaka@iki.fi>
Reviewed-by: Noah Misch <noah@leadboat.com>
Discussion: https://postgr.es/m/uvrtrknj4kdytuboidbhwclo4gxhswwcpgadptsjvjqcluzmah%40brqs62irg4dt
The problem is that ALTER SUBSCRIPTION ... SET PUBLICATION ... will lead
to restarting of apply worker and after the restart, the apply worker will
use the existing slot and replication origin corresponding to the
subscription. Now, it is possible that before the restart, the origin has
not been updated, and the WAL start location points to a location before
where PUBLICATION pointed to by SET PUBLICATION doesn't exist, and that
can lead to an error like: "ERROR: publication "pub1" does not exist".
Once this error occurs, apply worker will never be able to proceed and
will always return the same error.
We decided to skip loading the publication if the publication does not
exist. The publication is loaded later and updates the relation entry when
the publication gets created.
We decided not to backpatch this as this is a behaviour change, and we don't
see field reports. This problem has been found by intermittent buildfarm
failures.
Author: vignesh C <vignesh21@gmail.com>
Reviewed-by: Dilip Kumar <dilipbalaut@gmail.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Reviewed-by: Masahiko Sawada <sawada.mshk@gmail.com>
Discussion: https://postgr.es/m/flat/CALDaNm0-n8FGAorM%2BbTxkzn%2BAOUyx5%3DL_XmnvOP6T24%2B-NcBKg%40mail.gmail.com
Discussion: https://postgr.es/m/CAA4eK1+T-ETXeRM4DHWzGxBpKafLCp__5bPA_QZfFQp7-0wj4Q@mail.gmail.com
We want to support a "noreturn" decoration on more compilers besides
just GCC-compatible ones, but for that we need to move the decoration
in front of the function declaration instead of either behind it or
wherever, which is the current style afforded by GCC-style attributes.
Also rename the macro to "pg_noreturn" to be similar to the C11
standard "noreturn".
pg_noreturn is now supported on all compilers that support C11 (using
_Noreturn), as well as GCC-compatible ones (using __attribute__, as
before), as well as MSVC (using __declspec). (When PostgreSQL
requires C11, the latter two variants can be dropped.)
Now, all supported compilers effectively support pg_noreturn, so the
extra code for !HAVE_PG_ATTRIBUTE_NORETURN can be dropped.
This also fixes a possible problem if third-party code includes
stdnoreturn.h, because then the current definition of
#define pg_attribute_noreturn() __attribute__((noreturn))
would cause an error.
Note that the C standard does not support a noreturn attribute on
function pointer types. So we have to drop these here. There are
only two instances at this time, so it's not a big loss. In one case,
we can make up for it by adding the pg_noreturn to a wrapper function
and adding a pg_unreachable(), in the other case, the latter was
already done before.
Reviewed-by: Dagfinn Ilmari Mannsåker <ilmari@ilmari.org>
Reviewed-by: Andres Freund <andres@anarazel.de>
Discussion: https://www.postgresql.org/message-id/flat/pxr5b3z7jmkpenssra5zroxi7qzzp6eswuggokw64axmdixpnk@zbwxuq7gbbcw
On Publication rename, we need to only invalidate the RelationSyncCache
entries corresponding to relations that are part of the publication being
renamed.
As part of this patch, we introduce a new invalidation message to
invalidate the cache maintained by the logical decoding output plugin. We
can't use existing relcache invalidation for this purpose, as that would
unnecessarily cause relcache invalidations in other backends.
This will improve performance by building fewer relation cache entries
during logical replication.
Author: Hayato Kuroda <kuroda.hayato@fujitsu.com>
Author: Shlok Kyal <shlok.kyal.oss@gmail.com>
Reviewed-by: Hou Zhijie <houzj.fnst@fujitsu.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Discussion: https://postgr.es/m/OSCPR01MB14966C09AA201EFFA706576A7F5C92@OSCPR01MB14966.jpnprd01.prod.outlook.com
There used to be bespoken pools for these structs to reduce the
palloc/pfree overhead, but that was ripped out a long time ago and
replaced with the generic, cheaper generational memory allocator
(commit a4ccc1cef5). The Get/Return terminology made sense with the
pools, as you "got" an object from the pool and "returned" it later,
but now it just looks weird. Rename to Alloc/Free.
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://www.postgresql.org/message-id/c9e43d2d-8e83-444f-b111-430377368989@iki.fi
Previously, pg_logicalinspect functions were too trusting of their
input and blindly passed it to SnapBuildRestoreSnapshot(). If the
input pointed to a directory, the server could a PANIC error while
attempting to fsync_fname() with isdir=false on a directory.
This commit adds validation checks for input filenames and passes the
LSN extracted from the filename to SnapBuildRestoreSnapshot() instead
of the filename itself. It also adds regression tests for various
input patterns and permission checks.
Bug: #18828
Reported-by: Robins Tharakan <tharakan@gmail.com>
Co-authored-by: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
Co-authored-by: Masahiko Sawada <sawada.mshk@gmail.com>
Discussion: https://postgr.es/m/18828-0f4701c635064211@postgresql.org
On change of publication via ALTER PUBLICATION ... SET/ADD/DROP commands,
we were invalidating all the relations present in relation sync cache
maintained by pgoutput. We need to invalidate only the relation entries
that are changed as part of publication DDL.
We have ensured that the publication DDL execution generated the
invalidations required to invalidate impacted relation sync entries in
RelationSyncCache.
This improves the performance by avoiding building the cache entries for
the cases where a publication has many tables but only one of them is
dropped.
Author: Shlok Kyal <shlok.kyal.oss@gmail.com>
Author: Hayato Kuroda <kuroda.hayato@fujitsu.com>
Reviewed-by: Hou Zhijie <houzj.fnst@fujitsu.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Discussion: https://postgr.es/m/OSCPR01MB14966C09AA201EFFA706576A7F5C92@OSCPR01MB14966.jpnprd01.prod.outlook.com
This commit adds two improvements related to the monitoring of WAL
writes for the WAL receiver.
First, write counts and timings are now counted in pg_stat_io for the
WAL receiver. These have been discarded from pg_stat_wal in
ff99918c62 due to performance concerns, related to the fact that we
still relied on an on-disk file for the stats back then, even with
track_wal_io_timing to avoid the overhead of the timestamp calculations.
This implementation is simpler than the original proposal as it is
possible to rely on the APIs of pgstat_io.c to do the job. Like the
fsync and read data, track_wal_io_timing needs to be enabled to track
the timings.
Second, a wait event is added around the pg_pwrite() call in charge of
the writes, using the exiting WAIT_EVENT_WAL_WRITE. This is useful as
the WAL receiver data is tracked in pg_stat_activity.
Reviewed-by: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
Discussion: https://postgr.es/m/Z8gFnH4o3jBm5BRz@ip-10-97-1-34.eu-west-3.compute.internal
The usual pattern for handling a signal is that the signal handler
sets a flag and calls SetLatch(MyLatch), and CHECK_FOR_INTERRUPTS() or
other code that is part of a wait loop calls another function to deal
with it. The naming of the functions involved was a bit inconsistent,
however. CHECK_FOR_INTERRUPTS() calls ProcessInterrupts() to do the
heavy-lifting, but the analogous functions in aux processes were
called HandleMainLoopInterrupts(), HandleStartupProcInterrupts(),
etc. Similarly, most subroutines of ProcessInterrupts() were called
Process*(), but some were called Handle*().
To make things less confusing, rename all the functions that are part
of the overall signal/interrupt handling system but are not executed
in a signal handler to e.g. ProcessSomething(), rather than
HandleSomething(). The "Process" prefix is now consistently used in
the non-signal-handler functions, and the "Handle" prefix in functions
that are part of signal handlers, except for some completely unrelated
functions that clearly have nothing to do with signal or interrupt
handling.
Reviewed-by: Nathan Bossart <nathandbossart@gmail.com>
Discussion: https://www.postgresql.org/message-id/8a384b26-1499-41f6-be33-64b801fb98b8@iki.fi
The WAL receiver and WAL summarizer processes gain each one a call to
pgstat_report_wal(), to make sure that they report their WAL statistics
to pgstats, gathering data for pg_stat_io.
In the WAL receiver, the stats reports are timed with status updates sent
to the primary, that depend on wal_receiver_status_interval and
wal_receiver_timeout. This is a conservative choice, but perhaps we
could be more aggressive with the frequency of the stats reports. An
interesting historical fact is that the WAL receiver does writes and
syncs of WAL, but it has never reported its statistics to pgstats in
pg_stat_wal.
In the WAL summarizer, the stats reports are done each time the process
waits for WAL.
While on it, pg_stat_io is adjusted so as these two processes do not
report any rows when IOObject is not WAL, making the view easier to use
with less rows.
Two tests are added in TAP, checking statistics for the WAL summarizer
and the WAL receiver. Status updates in the WAL receiver are currently
possible in the recovery test 001_stream_rep.pl.
Reviewed-by: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
Discussion: https://postgr.es/m/Z8UKZyVSHUUQJHNb@paquier.xyz
For AIO, and also some other recent patches, we need the ability to call
relpath() in a critical section. Until now that was not feasible, as it
allocated memory.
The fact that relpath() allocated memory also made it awkward to use in log
messages because we had to take care to free the memory afterwards. Which we
e.g. didn't do for when zeroing out an invalid buffer.
We discussed other solutions, e.g. filling a pre-allocated buffer that's
passed to relpath(), but they all came with plenty downsides or were larger
projects. The easiest fix seems to be to make relpath() return the path by
value.
To be able to return the path by value we need to determine the maximum length
of a relation path. This patch adds a long #define that computes the exact
maximum, which is verified to be correct in a regression test.
As this change the signature of relpath(), extensions using it will need to
adapt their code. We discussed leaving a backward-compat shim in place, but
decided it's not worth it given the use of relpath() doesn't seem widespread.
Discussion: https://postgr.es/m/xeri5mla4b5syjd5a25nok5iez2kr3bm26j2qn4u7okzof2bmf@kwdh2vf7npra
When a standby replays an XLOG_PARAMETER_CHANGE record that lowers
wal_level below logical, we invalidate all logical slots in hot
standby mode. However, if this record was replayed while not in hot
standby mode, logical slots could remain valid even after promotion,
potentially causing an assertion failure during WAL record decoding.
To fix this issue, this commit adds a check for hot_standby status
when restoring a logical replication slot on standbys. This check
ensures that logical slots are invalidated when they become
incompatible due to insufficient wal_level during recovery.
Backpatch to v16 where logical decoding on standby was introduced.
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Reviewed-by: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
Discussion: https://postgr.es/m/CAD21AoABoFwGY_Rh2aeE6tEq3HkJxf0c6UeOXn4VV9v6BAQPSw%40mail.gmail.com
Backpatch-through: 16
The type argument wasn't actually really necessary. It was a remnant
of converting the API of the gist strategy translation from using
opclass to using opfamily+opcintype (commits c09e5a6a01,
622f678c10). For looking up the gist translation function, we used
the convention "amproclefttype = amprocrighttype = opclass's
opcintype" (see pg_amproc.h). But each operator family should only
have one translation function, and getting the right type for the
lookup is sometimes cumbersome and fragile, so this is all
unnecessarily complicated.
To simplify this, change the gist stategy support procedure to take
"any", "any" as argument. (This is arbitrary but seems intuitive.
The alternative of using InvalidOid as argument(s) upsets various DDL
commands, so it's not practical.) Then we don't need opcintype for
the lookup, and we can remove it from all the API layers introduced by
commit c09e5a6a01.
This also adds some more documentation about the correct signature of
the gist support function and adds more checks in gistvalidate().
This was previously underspecified. (It relied implicitly on
convention mentioned above.)
Discussion: https://www.postgresql.org/message-id/flat/E72EAA49-354D-4C2E-8EB9-255197F55330@enterprisedb.com
Logical replication crashes if the subscriber's partitioned table
has a BRIN index. There are two independently blamable causes,
and this patch fixes both:
1. brininsertcleanup fails if called twice for the same IndexInfo,
because it half-destroys its BrinInsertState but leaves it still
linked from ii_AmCache. brininsert would also fail in that state,
so it's pretty hard to see any advantage to this coding. Fully
remove the BrinInsertState, instead, so that a new brininsert
call would create a new cache.
2. A logical replication subscriber sometimes does ExecOpenIndices
twice on the same ResultRelInfo, followed by doing ExecCloseIndices
twice; the second call reaches the brininsertcleanup bug. Quite
aside from tickling unexpected cases in aminsertcleanup methods,
this seems very wasteful, because the IndexInfos built in the
first ExecOpenIndices call are just lost during the second call,
and have to be rebuilt at possibly-nontrivial cost. We should
establish a coding rule that you don't do that.
The problematic coding is that when the target table is partitioned,
apply_handle_tuple_routing calls ExecFindPartition which does
ExecOpenIndices (and expects that ExecCleanupTupleRouting will
close the indexes again). Using the ResultRelInfo made by
ExecFindPartition, it calls apply_handle_delete_internal or
apply_handle_insert_internal, both of which think they need to do
ExecOpenIndices/ExecCloseIndices for themselves. They do in the main
non-partitioned code paths, but not here. The simplest fix is to pull
their ExecOpenIndices/ExecCloseIndices calls out and put them in the
call sites for the non-partitioned cases. (We could have refactored
apply_handle_update_internal similarly, but I did not do so today
because there's no bug there: the partitioned code path doesn't
call it.)
Also, remove the always-duplicative open/close calls within
apply_handle_tuple_routing itself.
Since brininsertcleanup and indeed the whole aminsertcleanup mechanism
are new in v17, there's no observable bug in older branches. A case
could be made for trying to avoid these duplicative open/close calls
in the older branches, but for now it seems not worth the trouble and
risk of new bugs.
Bug: #18815
Reported-by: Sergey Belyashov <sergey.belyashov@gmail.com>
Discussion: https://postgr.es/m/18815-2a0407cc7f40b327@postgresql.org
Backpatch-through: 17
This commit introduces idle_replication_slot_timeout GUC that allows
inactive slots to be invalidated at the time of checkpoint. Because
checkpoints happen checkpoint_timeout intervals, there can be some lag
between when the idle_replication_slot_timeout was exceeded and when the
slot invalidation is triggered at the next checkpoint. To avoid such lags,
users can force a checkpoint to promptly invalidate inactive slots.
Note that the idle timeout invalidation mechanism is not applicable for
slots that do not reserve WAL or for slots on the standby server that are
synced from the primary server (i.e., standby slots having 'synced' field
'true'). Synced slots are always considered to be inactive because they
don't perform logical decoding to produce changes.
The slots can become inactive for a long period if a subscriber is down
due to a system error or inaccessible because of network issues. If such a
situation persists, it might be more practical to recreate the subscriber
rather than attempt to recover the node and wait for it to catch up which
could be time-consuming.
Then, external tools could create replication slots (e.g., for migrations
or upgrades) that may fail to remove them if an error occurs, leaving
behind unused slots that take up space and resources. Manually cleaning
them up can be tedious and error-prone, and without intervention, these
lingering slots can cause unnecessary WAL retention and system bloat.
As the duration of idle_replication_slot_timeout is in minutes, any test
using that would be time-consuming. We are planning to commit a follow up
patch for tests by using the injection point framework.
Author: Nisha Moond <nisha.moond412@gmail.com>
Author: Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>
Reviewed-by: Peter Smith <smithpb2250@gmail.com>
Reviewed-by: Hayato Kuroda <kuroda.hayato@fujitsu.com>
Reviewed-by: Vignesh C <vignesh21@gmail.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Reviewed-by: Hou Zhijie <houzj.fnst@fujitsu.com>
Reviewed-by: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
Discussion: https://postgr.es/m/CALj2ACW4aUe-_uFQOjdWCEN-xXoLGhmvRFnL8SNw_TZ5nJe+aw@mail.gmail.com
Discussion: https://postgr.es/m/OS0PR01MB5716C131A7D80DAE8CB9E88794FC2@OS0PR01MB5716.jpnprd01.prod.outlook.com
RBTXN_PREPARE flag and rbtxn_prepared macro could be misinterpreted as
either indicating the transaction type (e.g. a prepared transaction or
a normal transaction) or its currentstate (e.g. skipped or its prepare
message is sent), especially after commit 072ee847ad introduced the
RBTXN_SENT_PREPARE flag and the rbtxn_sent_prepare macro.
The RBTXN_PREPARE flag (and its corresponding macro) have been renamed
to RBTXN_IS_PREPARE to explicitly indicate the transaction
type. Therefore, this commit also adds the RBTXN_IS_PREPARE flag to
the transaction that is a prepared transaction and has been skipped,
which previously had only the RBTXN_SKIPPED_PREPARE flag.
Reviewed-by: Amit Kapila, Peter Smith
Discussion: https://postgr.es/m/CAA4eK1KgNmBsG%3D155E7QQ6TX9RoWnM4z5Z20SvsbwxSe_QXYsg%40mail.gmail.com
Previously, transaction aborts were detected concurrently only during
system catalog scans while replaying a transaction in streaming mode.
This commit adds an additional CLOG lookup to check the transaction
status, allowing the logical decoding to skip changes also when it
doesn't touch system catalogs, if the transaction is already
aborted. This optimization enhances logical decoding performance,
especially for large transactions that have already been rolled back,
as it avoids unnecessary disk or network I/O.
To avoid potential slowdowns caused by frequent CLOG lookups for small
transactions (most of which commit), the CLOG lookup is performed only
for large transactions before eviction. The performance benchmark
results showed there is not noticeable performance regression due to
CLOG lookups.
Reviewed-by: Amit Kapila, Peter Smith, Vignesh C, Ajin Cherian
Reviewed-by: Dilip Kumar, Andres Freund
Discussion: https://postgr.es/m/CAD21AoDht9Pz_DFv_R2LqBTBbO4eGrpa9Vojmt5z5sEx3XwD7A@mail.gmail.com
This adds a new variant of generated columns that are computed on read
(like a view, unlike the existing stored generated columns, which are
computed on write, like a materialized view).
The syntax for the column definition is
... GENERATED ALWAYS AS (...) VIRTUAL
and VIRTUAL is also optional. VIRTUAL is the default rather than
STORED to match various other SQL products. (The SQL standard makes
no specification about this, but it also doesn't know about VIRTUAL or
STORED.) (Also, virtual views are the default, rather than
materialized views.)
Virtual generated columns are stored in tuples as null values. (A
very early version of this patch had the ambition to not store them at
all. But so much stuff breaks or gets confused if you have tuples
where a column in the middle is completely missing. This is a
compromise, and it still saves space over being forced to use stored
generated columns. If we ever find a way to improve this, a bit of
pg_upgrade cleverness could allow for upgrades to a newer scheme.)
The capabilities and restrictions of virtual generated columns are
mostly the same as for stored generated columns. In some cases, this
patch keeps virtual generated columns more restricted than they might
technically need to be, to keep the two kinds consistent. Some of
that could maybe be relaxed later after separate careful
considerations.
Some functionality that is currently not supported, but could possibly
be added as incremental features, some easier than others:
- index on or using a virtual column
- hence also no unique constraints on virtual columns
- extended statistics on virtual columns
- foreign-key constraints on virtual columns
- not-null constraints on virtual columns (check constraints are supported)
- ALTER TABLE / DROP EXPRESSION
- virtual column cannot have domain type
- virtual columns are not supported in logical replication
The tests in generated_virtual.sql have been copied over from
generated_stored.sql with the keyword replaced. This way we can make
sure the behavior is mostly aligned, and the differences can be
visible. Some tests for currently not supported features are
currently commented out.
Reviewed-by: Jian He <jian.universality@gmail.com>
Reviewed-by: Dean Rasheed <dean.a.rasheed@gmail.com>
Tested-by: Shlok Kyal <shlok.kyal.oss@gmail.com>
Discussion: https://www.postgresql.org/message-id/flat/a368248e-69e4-40be-9c07-6c3b5880b0a6@eisentraut.org
This commit introduces changes to track unpruned relations explicitly,
making it possible for top-level plan nodes, such as ModifyTable and
LockRows, to avoid processing partitions pruned during initial
pruning. Scan-level nodes, such as Append and MergeAppend, already
avoid the unnecessary processing by accessing partition pruning
results directly via part_prune_index. In contrast, top-level nodes
cannot access pruning results directly and need to determine which
partitions remain unpruned.
To address this, this commit introduces a new bitmapset field,
es_unpruned_relids, which the executor uses to track the set of
unpruned relations. This field is referenced during plan
initialization to skip initializing certain nodes for pruned
partitions. It is initialized with PlannedStmt.unprunableRelids,
a new field that the planner populates with RT indexes of relations
that cannot be pruned during runtime pruning. These include relations
not subject to partition pruning and those required for execution
regardless of pruning.
PlannedStmt.unprunableRelids is computed during set_plan_refs() by
removing the RT indexes of runtime-prunable relations, identified
from PartitionPruneInfos, from the full set of relation RT indexes.
ExecDoInitialPruning() then updates es_unpruned_relids by adding
partitions that survive initial pruning.
To support this, PartitionedRelPruneInfo and PartitionedRelPruningData
now include a leafpart_rti_map[] array that maps partition indexes to
their corresponding RT indexes. The former is used in set_plan_refs()
when constructing unprunableRelids, while the latter is used in
ExecDoInitialPruning() to convert partition indexes returned by
get_matching_partitions() into RT indexes, which are then added to
es_unpruned_relids.
These changes make it possible for ModifyTable and LockRows nodes to
process only relations that remain unpruned after initial pruning.
ExecInitModifyTable() trims lists, such as resultRelations,
withCheckOptionLists, returningLists, and updateColnosLists, to
consider only unpruned partitions. It also creates ResultRelInfo
structs only for these partitions. Similarly, child RowMarks for
pruned relations are skipped.
By avoiding unnecessary initialization of structures for pruned
partitions, these changes improve the performance of updates and
deletes on partitioned tables during initial runtime pruning.
Due to ExecInitModifyTable() changes as described above, EXPLAIN on a
plan for UPDATE and DELETE that uses runtime initial pruning no longer
lists partitions pruned during initial pruning.
Reviewed-by: Robert Haas <robertmhaas@gmail.com> (earlier versions)
Reviewed-by: Tomas Vondra <tomas@vondra.me>
Discussion: https://postgr.es/m/CA+HiwqFGkMSge6TgC9KQzde0ohpAycLQuV7ooitEEpbKB0O_mg@mail.gmail.com
It is possible for the inactive_since value of an invalid replication slot
to be updated multiple times, which is unexpected behavior like during the
release of the slot or at the time of restart. This is harmless because
invalid slots are not allowed to be accessed but it is not prudent to
update invalid slots. We are planning to invalidate slots due to other
reasons like idle time and it will look odd that the slot's inactive_since
displays the recent time in this field after invalidated due to idle time.
So, this patch ensures that the inactive_since field of slots is not
updated for invalid slots.
In the passing, ensure to use the same inactive_since time for all the
slots at restart while restoring them from the disk.
Author: Nisha Moond <nisha.moond412@gmail.com>
Author: Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>
Reviewed-by: Vignesh C <vignesh21@gmail.com>
Reviewed-by: Peter Smith <smithpb2250@gmail.com>
Reviewed-by: Hou Zhijie <houzj.fnst@fujitsu.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Discussion: https://postgr.es/m/CABdArM7QdifQ_MHmMA=Cc4v8+MeckkwKncm2Nn6tX9wSCQ-+iw@mail.gmail.com
This turns GistTranslateCompareType() into a callback function of the
gist index AM instead of a standalone function. The existing callers
are changed to use IndexAmTranslateCompareType(). This then makes
that code not hardcoded toward gist.
This means in particular that the temporal keys code is now
independent of gist. Also, this generalizes commit 74edabce7a, so
other index access methods other than the previously hardcoded ones
could now work as REPLICA IDENTITY in a logical replication
subscriber.
Author: Mark Dilger <mark.dilger@enterprisedb.com>
Co-authored-by: Peter Eisentraut <peter@eisentraut.org>
Discussion: https://www.postgresql.org/message-id/flat/E72EAA49-354D-4C2E-8EB9-255197F55330@enterprisedb.com
Consistently use "Size" (or size_t, or in some places int64 or double)
as the type for variables holding memory allocation sizes. In most
places variables' data types were fine already, but we had an ancient
habit of computing bytes from kilobytes-units GUCs with code like
"work_mem * 1024L". That risks overflow on Win64 where they did not
make "long" as wide as "size_t". We worked around that by restricting
such GUCs' ranges, so you couldn't set work_mem et al higher than 2GB
on Win64. This patch removes that restriction, after replacing such
calculations with "work_mem * (Size) 1024" or variants of that.
It should be noted that this patch was constructed by searching
outwards from the GUCs that have MAX_KILOBYTES as upper limit.
So I can't positively guarantee there are no other places doing
memory-size arithmetic in int or long variables. I do however feel
pretty confident that increasing MAX_KILOBYTES on Win64 is safe now.
Also, nothing in our code should be dealing in multiple-gigabyte
allocations without authorization from a relevant GUC, so it seems
pretty likely that this search caught everything that could be at
risk of overflow.
Author: Vladlen Popolitov <v.popolitov@postgrespro.ru>
Co-authored-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/1a01f0-66ec2d80-3b-68487680@27595217
Once a replication slot is invalidated, it cannot be altered or used to
fetch changes. However, a process could still acquire an invalid slot and
fail later.
For example, if a process acquires a logical slot that was invalidated due
to wal_removed, it will eventually fail in CreateDecodingContext() when
attempting to access the removed WAL. Similarly, for physical replication
slots, even if the slot is invalidated and invalidation_reason is set to
wal_removed, the walsender does not currently check for invalidation when
starting physical replication. Instead, replication starts, and an error
is only reported later while trying to access WAL. Similarly, we prohibit
modifying slot properties for invalid slots but give the error for the
same after acquiring the slot.
This patch improves error handling by detecting invalid slots earlier at
the time of slot acquisition which is the first step. This also helped in
unifying different ERROR messages at different places and gave a
consistent message for invalid slots. This means that the message for
invalid slots will change to a generic message.
This will also be helpful for future patches where we are planning to
invalidate slots due to more reasons like idle_timeout because we don't
have to modify multiple places in such cases and avoid the chances of
missing out on a particular place.
Author: Nisha Moond <nisha.moond412@gmail.com>
Author: Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>
Reviewed-by: Vignesh C <vignesh21@gmail.com>
Reviewed-by: Peter Smith <smithpb2250@gmail.com>
Reviewed-by: Hayato Kuroda <kuroda.hayato@fujitsu.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Discussion: https://postgr.es/m/CABdArM6pBL5hPnSQ+5nEVMANcF4FCH7LQmgskXyiLY75TMnKpw@mail.gmail.com
