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
Sami complained that there's a discrepancy between n_mod_since_analyze
and n_ins_since_vacuum, as the former only accounts for committed changes
and the latter tracks committed and aborted inserts. Nobody seemed
overly concerned that this would cause any concerning issues. The
repercussions, from what I can tell, are limited to causing an
autovacuum to trigger for inserts sooner than it otherwise might. For
typical ratios of commits to aborts, it's unlikely to ever be noticed.
Fixing things to make it so n_ins_since_vacuum only displays committed
inserts would require an additional field in PgStat_TableCounts, which
does not quite seem worthwhile at this stage. This commit just adds a
comment with some details to mention that we know about it, which will
hopefully prevent repeat discussions.
Reported-by: Sami Imseih <samimseih@gmail.com>
Author: David Rowley <drowleyml@gmail.com>
Reviewed-by: Sami Imseih <samimseih@gmail.com>
Discussion: https://postgr.es/m/CAApHDvpgV3a-R2EGmPOh0L-x3pHbZpM3y4dySWfy+UqUazwDQA@mail.gmail.com
It's been twenty years since we generated constraint names that
look like "$N". So this advice about double-quoting such names
is well past its sell-by date, and now it merely seems confusing.
Reported-by: Yaroslav Saburov <y.saburov@gmail.com>
Author: "David G. Johnston" <david.g.johnston@gmail.com>
Discussion: https://postgr.es/m/174393459040.678.17810152410419444783@wrigleys.postgresql.org
Backpatch-through: 13
By inspection, ip_addrsize() can't return a negative result.
(If it could, we'd have way bigger problems elsewhere.)
So delete useless check in network_send(). Most C compilers
are probably perfectly capable of removing this code by
themselves, but it's confusing/misleading.
Bug: #18889
Reported-by: Daniel Elishakov <dan-eli@mail.ru>
Discussion: https://postgr.es/m/18889-73d4f19e953a629e@postgresql.org
Instead of two separate (and different) implementations, refactor to use
a single common routine.
Along the way, remove use of a hardcoded file permissions constant in
favor of the common project setting for directory creation.
Author: Mahendra Singh Thalor <mahi6run@gmail.com>
Discussion: https://postgr.es/m/CAKYtNApihL8X1h7XO-zOjznc8Ca66Aevgvhc9zOTh6DBh2iaeA@mail.gmail.com
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
d69d45a5a changed how em_is_child members are stored in
EquivalenceClasses. Children are no longer stored in the ec_members
list. optimizer/README mentioned that most operations "should ignore
child members", but that felt a little untrue now since child members
are now stored in a separate place, they simply won't be found by the
normal means of looking (a foreach loop over ec_members), and if you don't
find them, there's technically no need to "ignore" them.
Here we tweak the wording slightly to reflect the new storage location
for child members.
Reported-by: Amit Langote <amitlangote09@gmail.com>
Author: Amit Langote <amitlangote09@gmail.com>
Author: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/CA+HiwqE8v=EuAP_3F_A2xn8zWx+nG_etW_Fe_DvKO-Fkx=+DdQ@mail.gmail.com
This moves/renames some of the functions defined in pg_numa.c:
* pg_numa_get_pagesize() is renamed to pg_get_shmem_pagesize(), and
moved to src/backend/storage/ipc/shmem.c. The new name better reflects
that the page size is not related to NUMA, and it's specifically about
the page size used for the main shared memory segment.
* move pg_numa_available() to src/backend/storage/ipc/shmem.c, i.e. into
the backend (which more appropriate for functions callable from SQL).
While at it, improve the comment to explain what page size it returns.
* remove unnecessary includes from src/port/pg_numa.c, adding
unnecessary dependencies (src/port should be suitable for frontent).
These were either leftovers or unnecessary thanks to the other changes
in this commit.
This eliminates unnecessary dependencies on backend symbols, which we
don't want in src/port.
Reported-by: Kirill Reshke <reshkekirill@gmail.com>
Reviewed-by: Andres Freund <andres@anarazel.de>
https://postgr.es/m/CALdSSPi5fj0a7UG7Fmw2cUD1uWuckU_e8dJ+6x-bJEokcSXzqA@mail.gmail.com
With debug_discard_caches = 1, the runtime of this test script
increased by about a factor of 10 after commit 0dca5d68d. That's
causing some of our buildfarm animals to fail with a timeout.
The reason for the increased time is that now we are re-planning
some intentionally-non-inlineable SQL functions on every execution,
where the previous coding held onto the original plans throughout
the outer query. The previous behavior was arguably quite buggy,
so I don't think 0dca5d68d deserves blame here. But we would
like this test script to not take so long.
To fix, instead of forcing a "parallel safe" label via a
non-inlineable SQL function, apply it directly to the advisory-lock
functions by making internal-language aliases for them. A small
problem is that the advisory-lock functions return void but this
test would really like them to return integer 1. I cheated here by
declaring the aliases as returning "int". That's perhaps undue
familiarity with the implementation of PG_RETURN_VOID(), but that
hasn't changed in twenty years and is unlikely to do so in the next
twenty. That gets us an integer 0 result, and then an inline-able
wrapper to convert that to an integer 1 allows the rest of the script
to remain unchanged.
For me, this reduces the runtime with debug_discard_caches = 1
by about 100x, making the test comfortably faster than before
instead of slower.
Discussion: https://postgr.es/m/136163.1744179562@sss.pgh.pa.us
This spec fails ~3% of my Valgrind runs, and the spec has failed on Valgrind
buildfarm member skink at a similar rate. Two problems contributed to that:
- A competing buffer pin triggered VACUUM's lazy_scan_noprune() path, causing
"tuples missed: 1 dead from 1 pages not removed due to cleanup lock
contention". FREEZE fixes that.
- The spec ran lazy VACUUM immediately after VACUUM FULL. The spec implicitly
assumed lazy VACUUM prunes the one tuple that VACUUM FULL made dead. First
wait for old snapshots, making that assumption reliable.
This also adds two forms of defense in depth:
- Wait for snapshots using shared catalog pruning rules (VISHORIZON_SHARED).
This avoids the removable cutoff moving backward when an XID-bearing
autoanalyze process runs in another database. That may never happen in this
test, but it's cheap insurance.
- Use lazy VACUUM option DISABLE_PAGE_SKIPPING. Commit
c2dc1a7976 did this for a related requirement
in other tests, but I suspect FREEZE is necessary and sufficient in all
these tests.
Back-patch to v17, where the test first appeared.
Reported-by: Andres Freund <andres@anarazel.de>
Discussion: https://postgr.es/m/sv3taq4e6ea4qckimien3nxp3sz4b6cw6sfcy4nhwl52zpur4g@h6i6tohxmizu
Backpatch-through: 17
Change MyCancelKeyLength's type from uint8 to int. While it always
fits in a uint8, plain int is less surprising, as there's no
particular reason for it to be uint8.
Fix one ProcSignalInit caller that passed 'false' instead of NULL for
the pointer argument.
Author: Peter Eisentraut <peter@eisentraut.org>
Discussion: https://www.postgresql.org/message-id/61be9e31-7b7d-49d5-bc11-721800d89d64@eisentraut.org
These examples fail to account for join clauses generated by
EquivalenceClasses, but since we haven't mentioned EquivalenceClasses
yet it seems like it'd just add confusion to make them fully accurate.
Instead, parenthetically note that they're oversimplified.
Reported-by: Zeyuan Hu <ferrishu3886@gmail.com>
Co-authored-by: David Rowley <dgrowleyml@gmail.com>
Co-authored-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/CACvHWmYFo+60yMqKJajDDvKN5EM41YHrCT3oxukwXmGAqpWvyw@mail.gmail.com
It's weird to have the core regression tests depending on contrib
code, and coverage testing shows that those test queries add nothing
to the core-code coverage of the core tests. So pull those test bits
out and put them into ordinary test scripts inside contrib/spi/,
making that more like other contrib modules.
Aside from being structurally nicer, anything we can take out of the
core tests (which are executed multiple times per check-world run)
and put into tests executed only once should be a win. It doesn't
look like this change will buy a whole lot of milliseconds, but a
cycle saved is a cycle earned.
Also, there is some discussion around possibly removing refint and/or
autoinc altogether. I don't know if that will happen, but we'd
certainly need to decouple them from the core tests to do so.
The tests for autoinc were quite intertwined with the undocumented
"ttdummy" trigger in regress.c. That made the tests very hard to
understand and contributed nothing to autoinc's testing either.
So I just deleted ttdummy and rewrote the autoinc tests without it.
I realized while doing this that the description of autoinc in
the SGML docs is not a great description of what the function
actually does, so the patch includes some updates to those docs.
Author: Tom Lane <tgl@sss.pgh.pa.us>
Reviewed-by: Heikki Linnakangas <hlinnaka@iki.fi>
Discussion: https://postgr.es/m/3872677.1744077559@sss.pgh.pa.us
During development the third argument to pg_get_process_memory_contexts
was a retry count, but it was changed to a timeout instead. The param
name was accidentally left in pg_proc.dat though. Fix by renaming to
the correct parameter name.
Author: Fujii Masao <masao.fujii@oss.nttdata.com>
Reviewed-by: Daniel Gustafsson <daniel@yesql.se>
Discussion: https://postgr.es/m/3eb40b3e-45c7-426a-b7f8-81f7d05a9b53@oss.nttdata.com
This test was leaving files like delete_old_cluster.{sh,bat} in the
source directory for VPATH and meson builds. To fix, change the
directory to tmp_check before running the test, as was done in
commits 15b6d21553, 8af917be6b, and c462b054ba.
Oversight in commit af0d4901c1.
Reported-by: Andrew Dunstan <andrew@dunslane.net> (on Discord)
Reviewed-by: Michael Paquier <michael@paquier.xyz>
Reviewed-by: Andrew Dunstan <andrew@dunslane.net>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/Z_RHkG770w3SE0yU%40nathan
Commit b2bdb972c0 added MBUILD_TARGET to ensure that meson builds
the tests before running them, this adds MBUILD_TARGET to OpenBSD
and NetBSD builds as well where it was missing.
No backpatching since OpenBSD and NetBSD support does not exist
in the backbranch CI.
Author: Nazir Bilal Yavuz <byavuz81@gmail.com>
Reviewed-by: Daniel Gustafsson <daniel@yesql.se>
Discussion: https://postgr.es/m/CAN55FZ2LNnRrtL+cpSdEg44fQcLPq_GjJjfNa0vz+xqEdq=ZHw@mail.gmail.com
The page_num was defined as integer, which should be sufficient for the
near future (with 4K pages it's 8TB). But it's virtually free to return
bigint, and get a wider range. This was agreed on the thread, but I
forgot to tweak this in ba2a3c2302.
While at it, make the data types in CREATE VIEW a bit more consistent.
Discussion: https://postgr.es/m/CAKZiRmxh6KWo0aqRqvmcoaX2jUxZYb4kGp3N%3Dq1w%2BDiH-696Xw%40mail.gmail.co
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
It can be set to either COPY (the default) or CLONE if the system
supports it. CLONE causes callers of copydir(), currently CREATE
DATABASE ... STRATEGY=FILE_COPY and ALTER DATABASE ... SET TABLESPACE =
..., to use copy_file_range (Linux, FreeBSD) or copyfile (macOS) to copy
files instead of a read-write loop over the contents.
CLONE gives the kernel the opportunity to share block ranges on
copy-on-write file systems and push copying down to storage on others,
depending on configuration. On some systems CLONE can be used to clone
large databases quickly with CREATE DATABASE ... TEMPLATE=source
STRATEGY=FILE_COPY.
Other operating systems could be supported; patches welcome.
Co-authored-by: Nazir Bilal Yavuz <byavuz81@gmail.com>
Reviewed-by: Robert Haas <robertmhaas@gmail.com>
Reviewed-by: Ranier Vilela <ranier.vf@gmail.com>
Discussion: https://postgr.es/m/CA%2BhUKGLM%2Bt%2BSwBU-cHeMUXJCOgBxSHLGZutV5zCwY4qrCcE02w%40mail.gmail.com
This adds a function for retrieving memory context statistics
and information from backends as well as auxiliary processes.
The intended usecase is cluster debugging when under memory
pressure or unanticipated memory usage characteristics.
When calling the function it sends a signal to the specified
process to submit statistics regarding its memory contexts
into dynamic shared memory. Each memory context is returned
in detail, followed by a cumulative total in case the number
of contexts exceed the max allocated amount of shared memory.
Each process is limited to use at most 1Mb memory for this.
A summary can also be explicitly requested by the user, this
will return the TopMemoryContext and a cumulative total of
all lower contexts.
In order to not block on busy processes the caller specifies
the number of seconds during which to retry before timing out.
In the case where no statistics are published within the set
timeout, the last known statistics are returned, or NULL if
no previously published statistics exist. This allows dash-
board type queries to continually publish even if the target
process is temporarily congested. Context records contain a
timestamp to indicate when they were submitted.
Author: Rahila Syed <rahilasyed90@gmail.com>
Reviewed-by: Daniel Gustafsson <daniel@yesql.se>
Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Tomas Vondra <tomas@vondra.me>
Reviewed-by: Atsushi Torikoshi <torikoshia@oss.nttdata.com>
Reviewed-by: Fujii Masao <masao.fujii@oss.nttdata.com>
Reviewed-by: Alexander Korotkov <aekorotkov@gmail.com>
Discussion: https://postgr.es/m/CAH2L28v8mc9HDt8QoSJ8TRmKau_8FM_HKS41NeO9-6ZAkuZKXw@mail.gmail.com
Before, BAS_BULKREAD was always of size 256kB. With the default
io_combine_limit of 16, that only allowed 1-2 IOs to be in flight -
insufficient even on very low latency storage.
We don't just want to increase the size to a much larger hardcoded value, as
very large rings (10s of MBs of of buffers), appear to have negative
performance effects when reading in data that the OS has cached (but not when
actually needing to do IO).
To address this, increase the size of BAS_BULKREAD to allow for
io_combine_limit * effective_io_concurrency buffers getting read in. To
prevent the ring being much larger than useful, limit the increased size with
GetPinLimit().
The formula outlined above keeps the ring size to sizes for which we have not
observed performance regressions, unless very large effective_io_concurrency
values are used together with large shared_buffers setting.
Reviewed-by: Thomas Munro <thomas.munro@gmail.com>
Discussion: https://postgr.es/m/lqwghabtu2ak4wknzycufqjm5ijnxhb4k73vzphlt2a3wsemcd@gtftg44kdim6
Discussion: https://postgr.es/m/uvrtrknj4kdytuboidbhwclo4gxhswwcpgadptsjvjqcluzmah@brqs62irg4dt
In addition to the added functions, the pg_buffercache_evict() function now
shows whether the buffer was flushed.
pg_buffercache_evict_relation(): Evicts all shared buffers in a
relation at once.
pg_buffercache_evict_all(): Evicts all shared buffers at once.
Both functions provide mechanism to evict multiple shared buffers at
once. They are designed to address the inefficiency of repeatedly calling
pg_buffercache_evict() for each individual buffer, which can be time-consuming
when dealing with large shared buffer pools. (e.g., ~477ms vs. ~2576ms for
16GB of fully populated shared buffers).
These functions are intended for developer testing and debugging
purposes and are available to superusers only.
Minimal tests for the new functions are included. Also, there was no test for
pg_buffercache_evict(), test for this added too.
No new extension version is needed, as it was already increased this release
by ba2a3c2302.
Author: Nazir Bilal Yavuz <byavuz81@gmail.com>
Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Aidar Imamov <a.imamov@postgrespro.ru>
Reviewed-by: Joseph Koshakow <koshy44@gmail.com>
Discussion: https://postgr.es/m/CAN55FZ0h_YoSqqutxV6DES1RW8ig6wcA8CR9rJk358YRMxZFmw%40mail.gmail.com
When planning queries to partitioned tables, we clone all
EquivalenceMembers belonging to the partitioned table into em_is_child
EquivalenceMembers for each non-pruned partition. For partitioned tables
with large numbers of partitions, this meant the ec_members list could
become large and code searching that list would become slow. Effectively,
the more partitions which were present, the more searches needed to be
performed for operations such as find_ec_member_matching_expr() during
create_plan() and the more partitions present, the longer these searches
would take, i.e., a quadratic slowdown.
To fix this, here we adjust how we store EquivalenceMembers for
em_is_child members. Instead of storing these directly in ec_members,
these are now stored in a new array of Lists in the EquivalenceClass,
which is indexed by the relid. When we want to find EquivalenceMembers
belonging to a certain child relation, we can narrow the search to the
array element for that relation.
To make EquivalenceMember lookup easier and to reduce the amount of code
change, this commit provides a pair of functions to allow iteration over
the EquivalenceMembers of an EC which also handles finding the child
members, if required. Callers that never need to look at child members
can remain using the foreach loop over ec_members, which will now often
be faster due to only parent-level members being stored there.
The actual performance increases here are highly dependent on the number
of partitions and the query being planned. Performance increases can be
visible with as few as 8 partitions, but the speedup is marginal for
such low numbers of partitions. The speedups become much more visible
with a few dozen to hundreds of partitions. With some tested queries
using 56 partitions, the planner was around 3x faster than before. For
use cases with thousands of partitions, these are likely to become
significantly faster. Some testing has shown planner speedups of 60x or
more with 8192 partitions.
Author: Yuya Watari <watari.yuya@gmail.com>
Co-authored-by: David Rowley <dgrowleyml@gmail.com>
Reviewed-by: David Rowley <dgrowleyml@gmail.com>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Reviewed-by: Andrey Lepikhov <a.lepikhov@postgrespro.ru>
Reviewed-by: Alena Rybakina <lena.ribackina@yandex.ru>
Reviewed-by: Dmitry Dolgov <9erthalion6@gmail.com>
Reviewed-by: Amit Langote <amitlangote09@gmail.com>
Reviewed-by: Ashutosh Bapat <ashutosh.bapat.oss@gmail.com>
Tested-by: Thom Brown <thom@linux.com>
Tested-by: newtglobal postgresql_contributors <postgresql_contributors@newtglobalcorp.com>
Discussion: https://postgr.es/m/CAJ2pMkZNCgoUKSE%2B_5LthD%2BKbXKvq6h2hQN8Esxpxd%2Bcxmgomg%40mail.gmail.com
Some tests try to invalidate logical slots on the standby server by
running VACUUM on the primary. The problem is that xl_running_xacts was
getting generated and replayed before the VACUUM command, leading to the
advancement of the active slot's catalog_xmin. Due to this, active slots
were not getting invalidated, leading to test failures.
We fix it by skipping the generation of xl_running_xacts for the required
tests with the help of injection points. As the required interface for
injection points was not present in back branches, we fixed the failing
tests in them by disallowing the slot to become active for the required
cases (where rows_removed conflict could be generated).
Author: Hayato Kuroda <kuroda.hayato@fujitsu.com>
Reviewed-by: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Backpatch-through: 16, where it was introduced
Discussion: https://postgr.es/m/Z6oQXc8LmiTLfwLA@ip-10-97-1-34.eu-west-3.compute.internal
A PG 17 optimization allowed columns with NOT NULL constraints to skip
table scans for IS NULL queries, and to skip IS NOT NULL checks for IS
NOT NULL queries. This didn't work for domain types, since domain types
don't follow the IS NULL/IS NOT NULL constraint logic. To fix, disable
this optimization for domains for PG 17+.
Reported-by: Jan Behrens
Diagnosed-by: Tom Lane
Discussion: https://postgr.es/m/Z37p0paENWWUarj-@momjian.us
Backpatch-through: 17
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
Introduces a new view pg_buffercache_numa, showing NUMA memory nodes
for individual buffers. For each buffer the view returns an entry for
each memory page, with the associated NUMA node.
The database blocks and OS memory pages may have different size - the
default block size is 8KB, while the memory page is 4K (on x86). But
other combinations are possible, depending on configure parameters,
platform, etc. This means buffers may overlap with multiple memory
pages, each associated with a different NUMA node.
To determine the NUMA node for a buffer, we first need to touch the
memory pages using pg_numa_touch_mem_if_required, otherwise we might get
status -2 (ENOENT = The page is not present), indicating the page is
either unmapped or unallocated.
The view may be relatively expensive, especially when accessed for the
first time in a backend, as it touches all memory pages to get reliable
information about the NUMA node. This may also force allocation of the
shared memory.
Author: Jakub Wartak <jakub.wartak@enterprisedb.com>
Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
Reviewed-by: Tomas Vondra <tomas@vondra.me>
Discussion: https://postgr.es/m/CAKZiRmxh6KWo0aqRqvmcoaX2jUxZYb4kGp3N%3Dq1w%2BDiH-696Xw%40mail.gmail.com
Introduce new pg_shmem_alloctions_numa view with information about how
shared memory is distributed across NUMA nodes. For each shared memory
segment, the view returns one row for each NUMA node backing it, with
the total amount of memory allocated from that node.
The view may be relatively expensive, especially when executed for the
first time in a backend, as it has to touch all memory pages to get
reliable information about the NUMA node. This may also force allocation
of the shared memory.
Unlike pg_shmem_allocations, the view does not show anonymous shared
memory allocations. It also does not show memory allocated using the
dynamic shared memory infrastructure.
Author: Jakub Wartak <jakub.wartak@enterprisedb.com>
Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
Reviewed-by: Tomas Vondra <tomas@vondra.me>
Discussion: https://postgr.es/m/CAKZiRmxh6KWo0aqRqvmcoaX2jUxZYb4kGp3N%3Dq1w%2BDiH-696Xw%40mail.gmail.com
Add basic NUMA awareness routines, using a minimal src/port/pg_numa.c
portability wrapper and an optional build dependency, enabled by
--with-libnuma configure option. For now this is Linux-only, other
platforms may be supported later.
A built-in SQL function pg_numa_available() allows checking NUMA
support, i.e. that the server was built/linked with the NUMA library.
The main function introduced is pg_numa_query_pages(), which allows
determining the NUMA node for individual memory pages. Internally the
function uses move_pages(2) syscall, as it allows batching, and is more
efficient than get_mempolicy(2).
Author: Jakub Wartak <jakub.wartak@enterprisedb.com>
Co-authored-by: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Álvaro Herrera <alvherre@alvh.no-ip.org>
Reviewed-by: Tomas Vondra <tomas@vondra.me>
Discussion: https://postgr.es/m/CAKZiRmxh6KWo0aqRqvmcoaX2jUxZYb4kGp3N%3Dq1w%2BDiH-696Xw%40mail.gmail.com
check_foreign_key incorrectly used a single cache entry for its saved
plans for a 'c' (cascade) trigger, although there are two different
queries to execute depending on whether it fires for an update or a
delete. This caused the wrong things to be done if both types of
event occur in one session. (This was indeed visible in the triggers
regression test, but apparently nobody ever questioned it.) To fix,
add the operation type to the cache key.
Its debug log output failed to distinguish update from delete
events, too.
Also, change the intended trigger usage from BEFORE ROW to AFTER ROW,
and add checks insisting on that usage. BEFORE is really rather
unsafe, since if there are other BEFORE triggers they might change or
cancel the operation we are trying to check. AFTER triggers are the
standard way to propagate changes to other rows, so we should follow
that way here.
In passing, remove a useless duplicate lookup of the cache entry.
This code is mostly intended as a documentation example, so we
won't consider a back-patch.
Author: Dmitrii Bondar <d.bondar@postgrespro.ru>
Reviewed-by: Paul Jungwirth <pj@illuminatedcomputing.com>
Reviewed-by: Lilian Ontowhee <ontowhee@gmail.com>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/79755a2b18ed4fe5e29da6a87a1e00d1@postgrespro.ru
In some edge cases valgrind flags issues with the memory referenced by
IOs. All of the cases addressed in this change are false positives.
Most of the false positives are caused by UnpinBuffer[NoOwner] marking buffer
data as inaccessible. This happens even though the AIO subsystem still holds a
pin. That's good, there shouldn't be accesses to the buffer outside of AIO
related code until it is pinned by "user" code again. But it requires some
explicit work - if the buffer is not pinned by the current backend, we need to
explicitly mark the buffer data accessible/inaccessible while executing
completion callbacks.
That however causes a cascading issue in IO workers: After the completion
callbacks for a buffer is executed, the page is marked as inaccessible. If
subsequently the same worker is executing IO targeting the same buffer, we
would get an error, as the memory is still marked inaccessible. To avoid that,
we need to explicitly mark the memory as accessible in IO workers.
Another issue is that IO executed in workers or via io_uring will not mark
memory as DEFINED. In the case of workers that is because valgrind does not
track memory definedness across processes. For io_uring that is because
valgrind does not understand io_uring, and therefore its IOs never mark memory
as defined, whether the completions are processed in the defining process or
in another context. It's not entirely clear how to best solve that. The
current user of AIO is not affected, as it explicitly marks buffers as DEFINED
& NOACCESS anyway. Defer solving this issue until we have a user with
different needs.
Per buildfarm animal skink.
Reviewed-by: Noah Misch <noah@leadboat.com>
Co-authored-by: Noah Misch <noah@leadboat.com>
Discussion: https://postgr.es/m/3pd4322mogfmdd5nln3zphdwhtmq3rzdldqjwb2sfqzcgs22lf@ok2gletdaoe6
This mirrors 1e0dfd166b (+ 46ef520b95), for temporary table buffers. This
is mainly interesting right now because the AIO work currently triggers
spurious valgrind errors, and the fix for that is cleaner if temp buffers
behave the same as shared buffers.
This requires one change beyond the annotations themselves, namely to pin
local buffers while writing them out in FlushRelationBuffers().
Reviewed-by: Noah Misch <noah@leadboat.com>
Co-authored-by: Noah Misch <noah@leadboat.com>
Discussion: https://postgr.es/m/3pd4322mogfmdd5nln3zphdwhtmq3rzdldqjwb2sfqzcgs22lf@ok2gletdaoe6
Peter Eisentraut