Allow using multiple worker processes to build BRIN index, which until
now was supported only for BTREE indexes. For large tables this often
results in significant speedup when the build is CPU-bound.
The work is split in a simple way - each worker builds BRIN summaries on
a subset of the table, determined by the regular parallel scan used to
read the data, and feeds them into a shared tuplesort which sorts them
by blkno (start of the range). The leader then reads this sorted stream
of ranges, merges duplicates (which may happen if the parallel scan does
not align with BRIN pages_per_range), and adds the resulting ranges into
the index.
The number of duplicate results produced by workers (requiring merging
in the leader process) should be fairly small, thanks to how parallel
scans assign chunks to workers. The likelihood of duplicate results may
increase for higher pages_per_range values, but then there are fewer
page ranges in total. In any case, we expect the merging to be much
cheaper than summarization, so this should be a win.
Most of the parallelism infrastructure is a simplified copy of the code
used by BTREE indexes, omitting the parts irrelevant for BRIN indexes
(e.g. uniqueness checks).
This also introduces a new index AM flag amcanbuildparallel, determining
whether to attempt to start parallel workers for the index build.
Original patch by me, with reviews and substantial reworks by Matthias
van de Meent, certainly enough to make him a co-author.
Author: Tomas Vondra, Matthias van de Meent
Reviewed-by: Matthias van de Meent
Discussion: https://postgr.es/m/c2ee7d69-ce17-43f2-d1a0-9811edbda6e6%40enterprisedb.com
When building BRIN indexes, the brinbuildCallback only advances to the
next page range when seeing a tuple that doesn't belong to the current
one. This means that the index may end up missing ranges at the end of
the table, if those pages do not contain any indexable tuples.
We tend not to have completely empty pages at the end of a relation, but
this also applies to partial indexes, where the tuples may simply not
match the index predicate. This results in inefficient scans using the
affected BRIN index - without the summaries, the page ranges have to be
read and processed, which consumes I/O and possibly also CPU time.
The existing code already added empty ranges for earlier parts of the
table, this commit makes sure we add them for the ranges at the end of
the table too.
Patch by Matthias van de Meent, with review/improvements by me.
Author: Matthias van de Meent
Reviewed-by: Tomas Vondra
Discussion: https://postgr.es/m/CAEze2WiMsPZg%3DxkvSF_jt4%3D69k6K7gz5B8V2wY3gCGZ%2B1BzCbQ%40mail.gmail.com
While checking if a record could fit in the circular WAL decoding
buffer, the coding from commit 3f1ce973 used arithmetic that could
overflow. 64 bit systems were unaffected for various technical reasons,
which probably explains the lack of problem reports. Likewise for 32
bit systems running known 32 bit kernels. The systems at risk of
problems appear to be 32 bit processes running on 64 bit kernels, with
unlucky placement in memory.
Per complaint from GCC -fsanitize=undefined -m32, while testing
variations of 039_end_of_wal.pl.
Back-patch to 15.
Reviewed-by: Nathan Bossart <nathandbossart@gmail.com>
Reviewed-by: Robert Haas <robertmhaas@gmail.com>
Discussion: https://postgr.es/m/CA%2BhUKGKH0oRPOX7DhiQ_b51sM8HqcPp2J3WA-Oen%3DdXog%2BAGGQ%40mail.gmail.com
tts_virtual_copyslot() contained an Assert that checked that the srcslot
contained <= attributes than the dstslot. This seems to be backwards as
if the srcslot contained fewer attributes then the dstslot could be left
with stale Datum values from the previously stored tuple. It might make
more sense to allow the source to contain additional attributes and only
copy the leading ones that also exist in the destination, however, that's
not what we're doing here.
Here we just remove the Assert from tts_virtual_copyslot() and add an
Assert to ExecCopySlot() to verify the attribute counts match. There
does not seem to be any places where the destination contains fewer
attributes, so instead of going to the trouble of making the code
properly handle this, let's just ensure the attribute counts match. If
this Assert fails then that will indicate that we do have cases that
require us to handle the srcslot with more attributes than the dstslot.
It seems better to only write this code if there's a genuine requirement
for it rather than write it now only to leave it untested.
Thanks to Andres Freund for helping with the analysis of this.
Discussion: https://postgr.es/m/CAApHDvpMAvBL0T+TRORquyx1iqFQKMVTXtqNocOw0Pa2uh1heg@mail.gmail.com
An invalid index is skipped when doing REINDEX CONCURRENTLY at table
level, with INDEX_CORRUPTED used as errcode. This is confusing,
because an invalid index could exist after an interruption. The errcode
is switched to OBJECT_NOT_IN_PREREQUISITE_STATE instead, as per a
suggestion from Andres Freund.
While on it, the error messages are reworded, and a hint is added,
telling how to rebuild an invalid index in this case. This has been
suggested by Noah Misch.
Discussion: https://postgr.es/m/20231118230958.4fm3fhk4ypshxopa@awork3.anarazel.de
The commit 29d0a77fa6 labelled binary_upgrade_logical_slot_has_caught_up()
as a non-strict function to allow providing a better error message to callers
in case the passed slot_name is NULL. On further discussion, it seems that
it is not helpful to have a different error message for NULL input in this
function, so this patch marks the function as strict.
This patch also removes the explicit permission check to use replication
slots as this function is invoked only by superusers and instead adds an
Assert.
Reported-by: Masahiko Sawada
Author: Hayato Kuroda
Reviewed-by: Vignesh C
Discussion: https://postgr.es/m/CAD21AoDSyiBKkMXBxN_gUayZZUCOgyHnG8Ge8rcPXNP3Tf6B4g@mail.gmail.com
A REINDEX CONCURRENTLY run on a table with no indexes would always pop
the topmost snapshot from the active snapshot stack, making the snapshot
handling inconsistent between the multiple-relation and single-relation
cases. This commit slightly changes the snapshot stack handling so as a
snapshot is popped only ReindexMultipleInternal() in this case after a
relation has been reindexed, fixing a problem where an event trigger
function may need a snapshot but does not have one. This also keeps the
places where PopActiveSnapshot() is called closer to each other.
While on it, this expands the existing tests to cover all the cases that
could be faced with REINDEX commands and such event triggers, for one or
more relations, with or without indexes.
This behavior is inconsistent since 5dc92b844e, but we've never had a
need for an active snapshot at the end of a REINDEX until now.
Thanks also to Jian He for the input.
Reported-by: Alexander Lakhin
Discussion: https://postgr.es/m/cb538743-484c-eb6a-a8c5-359980cd3a17@gmail.com
When preparing commit 98e675ed7a I accidentally forgot to run
pgindent, which did produce a diff. Fix by adding the required
whitespace per the koel buildfarm failure.
Commit 5a991ef869 accidentally reversed the order of the tuples
and fields parameters, making the error message incorrectly refer
to 3 tuples with 1 field when IDENTIFY_SYSTEM returns 1 tuple and
3 or 4 fields. Fix by changing the order of the parameters. This
also adds a comment describing why we check for < 3 when postgres
since 9.4 has been sending 4 fields.
Backpatch all the way since the bug is almost a decade old.
Author: Tomonari Katsumata <t.katsumata1122@gmail.com>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Bug: #18224
Backpatch-through: v12
Presently, pg_convert() allocates a new bytea and copies the result
regardless of whether any conversion actually happened. This
commit adjusts this function to return the source pointer as-is if
no conversion occurred. This optimization isn't expected to make a
tremendous difference, but it still seems worthwhile to avoid
unnecessary memory allocations.
Author: Yurii Rashkovskii
Reviewed-by: Bertrand Drouvot
Discussion: https://postgr.es/m/CA%2BRLCQyknBPSWXRBQGOi6aYEcdQ9RpH9Kch4GjoeY8dQ3D%2Bvhw%40mail.gmail.com
After commit fd5e8b440d, InitProcess() is called later in the
EXEC_BACKEND startup sequence, so it's enough to set the
am_autovacuum_[launcher|worker] variables at the same place as in the
!EXEC_BACKEND case.
This commit adds support for REINDEX in event triggers, making this
command react for the events ddl_command_start and ddl_command_end. The
indexes rebuilt are collected with the ReindexStmt emitted by the
caller, for the concurrent and non-concurrent paths.
Thanks to that, it is possible to know a full list of the indexes that a
single REINDEX command has worked on.
Author: Garrett Thornburg, Jian He
Reviewed-by: Jim Jones, Michael Paquier
Discussion: https://postgr.es/m/CAEEqfk5bm32G7sbhzHbES9WejD8O8DCEOaLkxoBP7HNWxjPpvg@mail.gmail.com
The order of process initialization steps is now more consistent
between !EXEC_BACKEND and EXEC_BACKEND modes. InitProcess() is called
at the same place in either mode. We can now also move the
AttachSharedMemoryStructs() call into InitProcess() itself. This
reduces the number of "#ifdef EXEC_BACKEND" blocks.
Reviewed-by: Tristan Partin, Andres Freund, Alexander Lakhin
Discussion: https://www.postgresql.org/message-id/7a59b073-5b5b-151e-7ed3-8b01ff7ce9ef@iki.fi
The BackgroundWorker struct is now passed the same way as the Port
struct. Seems more consistent.
This makes it possible to move InitProcess later in SubPostmasterMain
(in next commit), as we no longer need to access shared memory to read
background worker entry.
Reviewed-by: Tristan Partin, Andres Freund, Alexander Lakhin
Discussion: https://www.postgresql.org/message-id/7a59b073-5b5b-151e-7ed3-8b01ff7ce9ef@iki.fi
For clarity, have separate functions for *creating* the shared memory
and semaphores at postmaster or single-user backend startup, and
for *attaching* to existing shared memory structures in EXEC_BACKEND
case. CreateSharedMemoryAndSemaphores() is now called only at
postmaster startup, and a new AttachSharedMemoryStructs() function is
called at backend startup in EXEC_BACKEND mode.
Reviewed-by: Tristan Partin, Andres Freund
Discussion: https://www.postgresql.org/message-id/7a59b073-5b5b-151e-7ed3-8b01ff7ce9ef@iki.fi
The padding bytes written to the backend params file were
uninitialized. That's harmless because we don't access the padding
bytes when we read the file back in, but Valgrind doesn't know
that. In any case, clear the padding bytes to make Valgrind happy.
Reported-by: Alexander Lakhin
Discussion: https://www.postgresql.org/message-id/014768ed-8b39-c44f-b07c-098c87b1644c@gmail.com
When creating a partitioned index, the partition key must be a subset
of the index's columns. But this currently doesn't check that the
collations between the partition key and the index definition match.
So you can construct a unique index that fails to enforce uniqueness.
(This would most likely involve a nondeterministic collation, so it
would have to be crafted explicitly and is not something that would
just happen by accident.)
This patch adds the required collation check. As a result, any
previously allowed unique index that has a collation mismatch would no
longer be allowed to be created.
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://www.postgresql.org/message-id/flat/3327cb54-f7f1-413b-8fdb-7a9dceebb938%40eisentraut.org
Commit 861f86beea changed hash_xlog_squeeze_page() to start reading
the write buffer conditionally but forgot to initialize it leading to an
uninitialized access.
Reported-by: Alexander Lakhin
Author: Hayato Kuroda
Reviewed-by: Alexander Lakhin, Amit Kapila
Discussion: http://postgr.es/m/62ed1a9f-746a-8e86-904b-51b9b806a1d9@gmail.com
Commit 7389aad6 removed the notion of backends started from inside a
signal handler. A stray comment still referred to them, while
explaining the need for a call to set_stack_base(). That leads to the
question of whether we still need the call in !EXEC_BACKEND builds.
There doesn't seem to be much point in suppressing it now, as it doesn't
hurt and probably helps to measure the stack base from the exact same
place in EXEC_BACKEND and !EXEC_BACKEND builds.
Back-patch to 16.
Reported-by: Heikki Linnakangas <hlinnaka@iki.fi>
Reported-by: Tristan Partin <tristan@neon.tech>
Reported-by: Andres Freund <andres@anarazel.de>
Discussion: https://postgr.es/m/CA%2BhUKG%2BEJHcevNGNOxVWxTNFbuB%3Dvjf4U68%2B85rAC_Sxvy2zEQ%40mail.gmail.com
InitAuxiliaryProcess() closely resembles InitProcess(), but it didn't
call InitLWLockAccess(). But because InitLWLockAccess() is a no-op
unless compiled with LWLOCK_STATS, and everything works even if it's
not called, the only consequence was that the stats were not printed
for aux processes.
This was an oversight in commit 1c6821be31, in version 9.5, so it is
missing in all supported branches. But since it only affects
developers using LWLOCK_STATS and no one has complained, no
backpatching.
Discussion: https://www.postgresql.org/message-id/20231130202648.7k6agmuizdilufnv@awork3.anarazel.de
It's not necessary to fill the key field in most cases, since
hash_search has already done that. Some existing call sites have an
assert or comment that this contract has been fulfilled, but those
are quite old and that practice seems unnecessary here.
While at it, remove a nearby redundant assignment that a smart compiler
will elide anyway.
Zhao Junwang, with some adjustments by me
Reviewed by Nathan Bossart, with additional feedback from Tom Lane
Discussion: http://postgr.es/m/CAEG8a3%2BUPF%3DR2QGPgJMF2mKh8xPd1H2TmfH77zPuVUFdBpiGUA%40mail.gmail.com
Quotes are applied to GUCs in a very inconsistent way across the code
base, with a mix of double quotes or no quotes used. This commit
removes double quotes around all the GUC names that are obviously
referred to as parameters with non-English words (use of underscore,
mixed case, etc).
This is the result of a discussion with Álvaro Herrera, Nathan Bossart,
Laurenz Albe, Peter Eisentraut, Tom Lane and Daniel Gustafsson.
Author: Peter Smith
Discussion: https://postgr.es/m/CAHut+Pv-kSN8SkxSdoHano_wPubqcg5789ejhCDZAcLFceBR-w@mail.gmail.com
Switch from using TransactionId to FullTransactionId in naming of 2PC files.
Transaction state file in the pg_twophase directory now have extra 8 bytes in
the name to address an epoch of a given xid.
Author: Maxim Orlov, Aleksander Alekseev, Alexander Korotkov, Teodor Sigaev
Author: Nikita Glukhov, Pavel Borisov, Yura Sokolov
Reviewed-by: Jacob Champion, Heikki Linnakangas, Alexander Korotkov
Reviewed-by: Japin Li, Pavel Borisov, Tom Lane, Peter Eisentraut, Andres Freund
Reviewed-by: Andrey Borodin, Dilip Kumar, Aleksander Alekseev
Discussion: https://postgr.es/m/CACG%3DezZe1NQSCnfHOr78AtAZxJZeCvxrts0ygrxYwe%3DpyyjVWA%40mail.gmail.com
Discussion: https://postgr.es/m/CAJ7c6TPDOYBYrnCAeyndkBktO0WG2xSdYduTF0nxq%2BvfkmTF5Q%40mail.gmail.com
This avoids the wraparound in async.c and removes the corresponding code
complexity. The maximum amount of allocated SLRU pages for NOTIFY / LISTEN
queue is now determined by the max_notify_queue_pages GUC. The default
value is 1048576. It allows to consume up to 8 GB of disk space which is
exactly the limit we had previously.
Author: Maxim Orlov, Aleksander Alekseev, Alexander Korotkov, Teodor Sigaev
Author: Nikita Glukhov, Pavel Borisov, Yura Sokolov
Reviewed-by: Jacob Champion, Heikki Linnakangas, Alexander Korotkov
Reviewed-by: Japin Li, Pavel Borisov, Tom Lane, Peter Eisentraut, Andres Freund
Reviewed-by: Andrey Borodin, Dilip Kumar, Aleksander Alekseev
Discussion: https://postgr.es/m/CACG%3DezZe1NQSCnfHOr78AtAZxJZeCvxrts0ygrxYwe%3DpyyjVWA%40mail.gmail.com
Discussion: https://postgr.es/m/CAJ7c6TPDOYBYrnCAeyndkBktO0WG2xSdYduTF0nxq%2BvfkmTF5Q%40mail.gmail.com
We've had repeated bugs in the area of handling SLRU wraparound in the past,
some of which have caused data loss. Switching to an indexing system for SLRUs
that does not wrap around should allow us to get rid of a whole bunch
of problems and improve the overall reliability of the system.
This particular patch however only changes the indexing and doesn't address
the wraparound per se. This is going to be done in the following patches.
Author: Maxim Orlov, Aleksander Alekseev, Alexander Korotkov, Teodor Sigaev
Author: Nikita Glukhov, Pavel Borisov, Yura Sokolov
Reviewed-by: Jacob Champion, Heikki Linnakangas, Alexander Korotkov
Reviewed-by: Japin Li, Pavel Borisov, Tom Lane, Peter Eisentraut, Andres Freund
Reviewed-by: Andrey Borodin, Dilip Kumar, Aleksander Alekseev
Discussion: https://postgr.es/m/CACG%3DezZe1NQSCnfHOr78AtAZxJZeCvxrts0ygrxYwe%3DpyyjVWA%40mail.gmail.com
Discussion: https://postgr.es/m/CAJ7c6TPDOYBYrnCAeyndkBktO0WG2xSdYduTF0nxq%2BvfkmTF5Q%40mail.gmail.com
Assigning a precedence to a keyword that isn't a kind of expression
operator is rather dangerous, because it might mask grammar
ambiguities that we'd rather know about. It's much safer to attach
explicit precedences to individual rules, which will affect the
behavior of only that one rule. Moreover, when we do have to give
a precedence to a non-operator keyword, we should try to give it the
same precedence as IDENT, thereby reducing the risk of surprising
side-effects.
Apply this hard-won knowledge to SET (which I misassigned ages ago
in commit 2647ad658) and some SQL/JSON-related productions
(from commits 6ee30209a, 71bfd1543).
Patch HEAD only, since there's no evidence of actual bugs here.
Discussion: https://postgr.es/m/CADT4RqBPdbsZW7HS1jJP319TMRHs1hzUiP=iRJYR6UqgHCrgNQ@mail.gmail.com
We should have done it this way all along, but we accidentally got
away with using the wrong BIO field up until OpenSSL 3.2. There,
the library's BIO routines that we rely on use the "data" field
for their own purposes, and our conflicting use causes assorted
weird behaviors up to and including core dumps when SSL connections
are attempted. Switch to using the approved field for the purpose,
i.e. app_data.
While at it, remove our configure probes for BIO_get_data as well
as the fallback implementation. BIO_{get,set}_app_data have been
there since long before any OpenSSL version that we still support,
even in the back branches.
Also, update src/test/ssl/t/001_ssltests.pl to allow for a minor
change in an error message spelling that evidently came in with 3.2.
Tristan Partin and Bo Andreson. Back-patch to all supported branches.
Discussion: https://postgr.es/m/CAN55FZ1eDDYsYaL7mv+oSLUij2h_u6hvD4Qmv-7PK7jkji0uyQ@mail.gmail.com
If the tuple being updated is not visible to the crosscheck snapshot,
we return TM_Updated but the assertions would not hold in that case.
Move them to before the cross-check.
Fixes bug #17893. Backpatch to all supported versions.
Author: Alexander Lakhin
Backpatch-through: 12
Discussion: https://www.postgresql.org/message-id/17893-35847009eec517b5%40postgresql.org
00b41463c adjusted Bitmapset so that an empty set is always represented
as NULL. This makes checking for empty sets far cheaper than it used
to be.
There were various places in the code where we'd call bms_membership()
to handle the 3 possible BMS_Membership values. For the BMS_SINGLETON
case, we'd also call bms_singleton_member() to find the single set member.
This can now be done in a more optimal way by first checking if the set is
NULL and then not bothering with bms_membership() and simply call
bms_get_singleton_member() instead to find the single member. This
function will return false if there are multiple members in the set.
Here we also tidy up some logic in examine_variable() for the single
member case. There's now no need to call bms_is_member() as we've
already established that we're working with a singleton Bitmapset, so we
can just check if varRelid matches the singleton member.
Reviewed-by: Richard Guo
Discussion: https://postgr.es/m/CAApHDvqW+CxNPcY245GaWiuqkkqgTudtG2ncGvvSjGn2wdTZLA@mail.gmail.com
Fix a bug introduced by c1ec02be1d. It may happen that the executor
opens indexes on the result relation, but no rows end up being inserted.
Then the index_insert_cleanup still gets executed, but passes down NULL
to the AM callback. The AM callback may not expect this, as is the case
of brininsertcleanup, leading to a crash.
Fixed by only calling the cleanup callback if (ii_AmCache != NULL). This
way the AM can simply assume to only see a valid cache.
Reported-by: Richard Guo
Discussion: https://postgr.es/m/CAMbWs4-w9qC-o9hQox9UHvdVZAYTp8OrPQOKtwbvzWaRejTT=Q@mail.gmail.com
XLogSetAsyncXactLSN(), called at asynchronous commit, would wake up
walwriter every time the LSN advances, but walwriter doesn't actually
do anything unless it has at least 'wal_writer_flush_after' full
blocks of WAL to write. Repeatedly waking up walwriter to do nothing
is a waste of CPU cycles in both walwriter and the backends doing the
wakeups. To fix, apply the same logic in XLogSetAsyncXactLSN() to
decide whether to wake up walwriter, as walwriter uses to determine if
it has any work to do.
In the passing, rename misleadingly named 'flushbytes' local variable
to 'flushblocks'.
Author: Andres Freund, Heikki Linnakangas
Discussion: https://www.postgresql.org/message-id/20231024230929.vsc342baqs7kmbte@awork3.anarazel.de
52e4f0cd4 introduced a bug in pgoutput in which missing values in tuples
were incorrectly filled in with NULL. The problem was the use of
CreateTupleDescCopy where CreateTupleDescCopyConstr was required, as the
former drops the constraints in the tuple description (specifically, the
default value constraint) on the floor.
The bug could result in incorrectness when a table replicated via
`REPLICA IDENTITY FULL` underwent a schema change that added a column
with a default value. The problem is that in such cases updates fill NULL
values in old tuples for missing columns for default values. Then on the
subscriber, we failed to find a matching tuple and missed updating the
required row.
Author: Nikhil Benesch
Reviewed-by: Hou Zhijie, Amit Kapila
Backpatch-through: 15
Discussion: http://postgr.es/m/CAPWqQZTEpZQamYsGMn6ZDRvVywwpVPiKH6OY4KSgA+NmeqFNzA@mail.gmail.com
Values corresponding to STATISTIC_KIND_RANGE_LENGTH_HISTOGRAM and
STATISTIC_KIND_BOUNDS_HISTOGRAM were not exposed to pg_stats when these
slot kinds were introduced in 918eee0c49.
This commit adds the missing fields to pg_stats.
Catversion is bumped.
Discussion: https://postgr.es/m/flat/b67d8b57-9357-7e82-a2e7-f6ce6eaeec67@postgrespro.ru
Author: Egor Rogov, Soumyadeep Chakraborty
Reviewed-by: Tomas Vondra, Justin Pryzby, Jian He
The brininsert code used to initialize (and destroy) BrinDesc and
BrinRevmap for each tuple, which is not free. This patch initializes
these structures only once, and reuses them for all inserts in the same
command. The data is passed through indexInfo->ii_AmCache.
This also introduces an optional AM callback "aminsertcleanup" that
allows performing custom cleanup in case simply pfree-ing ii_AmCache is
not sufficient (which is the case when the cache contains TupleDesc,
Buffers, and so on).
Author: Soumyadeep Chakraborty
Reviewed-by: Alvaro Herrera, Matthias van de Meent, Tomas Vondra
Discussion: https://postgr.es/m/CAE-ML%2B9r2%3DaO1wwji1sBN9gvPz2xRAtFUGfnffpd0ZqyuzjamA%40mail.gmail.com
Tomas Vondra
Heikki Linnakangas
Thomas Munro
David Rowley
Michael Paquier
Amit Kapila
Daniel Gustafsson
Jeff Davis
Nathan Bossart
Peter Eisentraut
Alexander Korotkov
John Naylor
Tom Lane
Bruce Momjian