There are two reasons for doing so:
1) It is generally faster to perform checks in a batched fashion and making
sequential scans faster is nice.
2) We would like to stop setting hint bits while pages are being written
out. The necessary locking becomes visible for page mode scans, if done for
every tuple. With batching, the overhead can be amortized to only happen
once per page.
There are substantial further optimization opportunities along these
lines:
- Right now HeapTupleSatisfiesMVCCBatch() simply uses the single-tuple
HeapTupleSatisfiesMVCC(), relying on the compiler to inline it. We could
instead write an explicitly optimized version that avoids repeated xid
tests.
- Introduce batched version of the serializability test
- Introduce batched version of HeapTupleSatisfiesVacuum
Reviewed-by: Melanie Plageman <melanieplageman@gmail.com>
Discussion: https://postgr.es/m/6rgb2nvhyvnszz4ul3wfzlf5rheb2kkwrglthnna7qhe24onwr@vw27225tkyar
To be able to guarantee that we can set the hint bit, acquire an exclusive
lock on the old buffer. This is required as a future commit will only allow
hint bits to be set with a new lock level, which is acquired as-needed in a
non-blocking fashion.
We need the hint bits, set in heapam_relation_copy_for_cluster() ->
HeapTupleSatisfiesVacuum(), to be set, as otherwise reform_and_rewrite_tuple()
-> rewrite_heap_tuple() will get confused. Specifically, rewrite_heap_tuple()
checks for HEAP_XMAX_INVALID in the old tuple to determine whether to check
the old-to-new mapping hash table.
It'd be better if we somehow could avoid setting hint bits on the old page. A
common reason to use VACUUM FULL is very bloated tables - rewriting most of
the old table during VACUUM FULL doesn't exactly help.
Reviewed-by: Heikki Linnakangas <heikki.linnakangas@iki.fi>
Reviewed-by: Kirill Reshke <reshkekirill@gmail.com>
Discussion: https://postgr.es/m/4wggb7purufpto6x35fd2kwhasehnzfdy3zdcu47qryubs2hdz@fa5kannykekr
Before this commit fsm_vacuum_page() modified the page without any lock on the
page. Historically that was kind of ok, as we didn't rely on the freespace to
really stay consistent and we did not have checksums. But these days pages are
checksummed and there are ways for FSM pages to be included in WAL records,
even if the FSM itself is still not WAL logged. If a FSM page ever were
modified while a WAL record referenced that page, we'd be in trouble, as the
WAL CRC could end up getting corrupted.
The reason to address this right now is a series of patches with the goal to
only allow modifications of pages with an appropriate lock level. Obviously
not having any lock is not appropriate :)
Reviewed-by: Kirill Reshke <reshkekirill@gmail.com>
Discussion: https://postgr.es/m/4wggb7purufpto6x35fd2kwhasehnzfdy3zdcu47qryubs2hdz@fa5kannykekr
Discussion: https://postgr.es/m/e6a8f734-2198-4958-a028-aba863d4a204@iki.fi
Doing this meant that those two headers, which are supposed to be
internal to their corresponding index AMs, were being included pretty
much universally, because tuplesort.h is included by execnodes.h which
is very widely used. Stop that, and fix fallout.
We also change indexing.h to no longer include execnodes.h (tuptable.h
is sufficient), and relscan.h to no longer include buf.h (pointless
since c2fe139c20).
Author: Mario González <gonzalemario@gmail.com>
Discussion: https://postgr.es/m/CAFsReFUcBFup=Ohv_xd7SNQ=e73TXi8YNEkTsFEE2BW7jS1noQ@mail.gmail.com
Previously the instrumentation logic always converted to seconds, only for
many of the callers to do unnecessary division to get to milliseconds. As an
upcoming refactoring will split the Instrumentation struct, utilize instrtime
always to keep things simpler. It's also a bit faster to not have to first
convert to a double in functions like InstrEndLoop(), InstrAggNode().
Author: Lukas Fittl <lukas@fittl.com>
Reviewed-by: Andres Freund <andres@anarazel.de>
Discussion: https://postgr.es/m/CAP53PkzZ3UotnRrrnXWAv=F4avRq9MQ8zU+bxoN9tpovEu6fGQ@mail.gmail.com
This formerly said "unique constraint must ...", which was accurate
enough when it only applied to UNIQUE and PRIMARY KEY constraints.
However, now we use it for exclusion constraints too, and in that
case it's a tad confusing. Do what we already did in the errdetail
message: print the constraint_type, so that it looks like "UNIQUE
constraint ...", "EXCLUDE constraint ...", etc.
Author: jian he <jian.universality@gmail.com>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/CACJufxH6VhAf65Vghg4T2q315gY=Rt4BUfMyunkfRj0n2S9n-g@mail.gmail.com
Since commit bd15b7db48, pg_dump uses pg_get_sequence_data() (née
pg_sequence_read_tuple()) to gather all sequence data in a single
query as opposed to a query per sequence. Two related bugs have
been identified:
* If the user lacks appropriate privileges on the sequence, pg_dump
generates a setval() command with garbage values instead of
failing as expected.
* pg_dump can fail due to a concurrently dropped sequence, even if
the dropped sequence's data isn't part of the dump.
This commit fixes the above issues by 1) teaching
pg_get_sequence_data() to return nulls instead of erroring for a
missing sequence and 2) teaching pg_dump to fail if it tries to
dump the data of a sequence for which pg_get_sequence_data()
returned nulls. Note that pg_dump may still fail due to a
concurrently dropped sequence, but it should now only do so when
the sequence data is part of the dump. This matches the behavior
before commit bd15b7db48.
Bug: #19365
Reported-by: Paveł Tyślacki <pavel.tyslacki@gmail.com>
Suggested-by: Tom Lane <tgl@sss.pgh.pa.us>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://postgr.es/m/19365-6245240d8b926327%40postgresql.org
Discussion: https://postgr.es/m/2885944.1767029161%40sss.pgh.pa.us
Backpatch-through: 18
This is important for Postgres extensions that are written in C++,
such as pg_duckdb, which uses PGXS as the build system currently. In
the autotools build, C++ is not coupled to LLVM. If the autotools
build is configured without --with-llvm, the C++ compiler and the
various flags get persisted into the Makefile.global.
Author: Tristan Partin <tristan@partin.io>
Author: Jelte Fennema-Nio <postgres@jeltef.nl>
Discussion: https://www.postgresql.org/message-id/flat/D98JHQF7H2A8.VSE3I4CJBTAB%40partin.io
When creating a partition for a RANGE partitioned table, the reporting
of errors relating to converting the specified range values into
constant values for the partition key's type could display the name of a
previous partition key column when an earlier range was specified as
MINVALUE or MAXVALUE.
This was caused by the code not correctly incrementing the index that
tracks which partition key the foreach loop was working on after
processing MINVALUE/MAXVALUE ranges.
Fix by using foreach_current_index() to ensure the index variable is
always set to the List element being worked on.
Author: myzhen <zhenmingyang@yeah.net>
Reviewed-by: zhibin wang <killerwzb@gmail.com>
Discussion: https://postgr.es/m/273cab52.978.19b96fc75e7.Coremail.zhenmingyang@yeah.net
Backpatch-through: 14
This patch completes the transition to making inet_ops be default for
inet/cidr columns, rather than btree_gist's opclasses. Once we do
that, though, pg_upgrade has a big problem. A dump from an older
version will see btree_gist's opclasses as being default, so it will
not mention the opclass explicitly in CREATE INDEX commands, which
would cause the restore to create the indexes using inet_ops. Since
that's not compatible with what's actually in the files, havoc would
ensue.
This isn't readily fixable, because the CREATE INDEX command strings
are built by the older server's pg_get_indexdef() function; pg_dump
hasn't nearly enough knowledge to modify those strings successfully.
Even if we cared to put in the work to make that happen in pg_dump,
it would be counterproductive because the end goal here is to get
people off of these opclasses. Allowing such indexes to persist
through pg_upgrade wouldn't advance that goal.
Therefore, this patch just adds code to pg_upgrade to detect indexes
that would be problematic and refuse to upgrade.
There's another issue too: even without any indexes to worry about,
pg_dump in binary-upgrade mode will reproduce the "CREATE OPERATOR
CLASS ... DEFAULT" commands for btree_gist's opclasses, and those
will fail because now we have a built-in opclass that provides a
conflicting default. We could ask users to drop the btree_gist
extension altogether before upgrading, but that would carry very
severe penalties. It would affect perfectly-valid indexes for other
data types, and it would drop operators that might be relied on in
views or other database objects. Instead, put a hack in DefineOpClass
to ignore the DEFAULT clauses for these opclasses when in
binary-upgrade mode. This will result in installing a version of
btree_gist that isn't quite the version it claims to be, but that can
be fixed by issuing ALTER EXTENSION UPDATE afterwards.
Since we don't apply that hack when not in binary-upgrade mode,
it is now impossible to install any version of btree_gist
less than 1.9 via CREATE EXTENSION.
Author: Tom Lane <tgl@sss.pgh.pa.us>
Reviewed-by: Matthias van de Meent <boekewurm+postgres@gmail.com>
Reviewed-by: Heikki Linnakangas <hlinnaka@iki.fi>
Discussion: https://postgr.es/m/2483812.1754072263@sss.pgh.pa.us
The only user-visible change is the fix in the "malformed
pg_dependencies" error detail. That one is new in commit e1405aa5e3,
so no backpatching required.
When repurposing a standby to a logical replica, pg_createsubscriber
uses for the new replica a set of configuration parameters saved into
postgresql.auto.conf, to force recovery patterns when the physical
replica is promoted.
While not wrong in practice, this approach can cause issues when forcing
again recovery on a logical replica or its base backup as the recovery
parameters are not reset on the target server once pg_createsubscriber
is done with the node.
This commit aims at improving the situation, by changing the way
recovery parameters are saved on the target node. Instead of writing
all the configuration to postgresql.auto.conf, this file now uses an
include_if_exists, that points to a pg_createsubscriber.conf. This new
file contains all the recovery configuration, and is renamed to
pg_createsubscriber.conf.disabled when pg_createsubscriber exits. This
approach resets the recovery parameters, and offers the benefit to keep
a trace of the setup used when the target node got promoted, for
debugging purposes. If pg_createsubscriber.conf cannot be renamed
(unlikely scenario), a warning is issued to inform users that a manual
intervention may be required to reset this configuration.
This commit includes a test case to demonstrate the problematic case: a
standby node created from a base backup of what was the target node of
pg_createsubscriber does not get confused when started. If removing
this new logic, the test fails with the standby not able to start due
to an incorrect recovery target setup, where the startup process fails
quickly with a FATAL.
I have provided the design idea for the patch, that Alyona has written
(with some code adjustments from me). This could be considered as a
bug, but after discussion this is put into the bucket for improvements.
Redesigning pg_createsubscriber would not be acceptable in the stable
branches anyway.
Author: Alyona Vinter <dlaaren8@gmail.com>
Reviewed-by: Ilyasov Ian <ianilyasov@outlook.com>
Reviewed-by: Alexander Korotkov <aekorotkov@gmail.com>
Reviewed-by: Hayato Kuroda <kuroda.hayato@fujitsu.com>
Reviewed-by: Michael Paquier <michael@paquier.xyz>
Reviewed-by: Andrey Rudometov <unlimitedhikari@gmail.com>
Discussion: https://postgr.es/m/CAGWv16K6L6Pzm99i1KiXLjFWx2bUS3DVsR6yV87-YR9QO7xb3A@mail.gmail.com
During catcache searches, the most-recently searched entries are kept at
the head of the list to speed up subsequent searches, keeping the
"freshest" entries at its beginning. A rehash of the catcache was doing
the opposite: fresh entries were moved to the tail of the newly-created
buckets, causing a rehash to slow down a bit.
When a rehash is done, this commit switches the code to use
dlist_push_tail() instead of dlist_push_head(), so as fresh entries are
kept at the head of the lists, not their tail.
Author: ChangAo Chen <cca5507@qq.com>
Reviewed-by: John Naylor <johncnaylorls@gmail.com>
Reviewed-by: Michael Paquier <michael@paquier.xyz>
Discussion: https://postgr.es/m/tencent_9EA10D8512B5FE29E7323F780A0749768708@qq.com
This new identifier type provides support for 64-bit unsigned values,
to be used in catalogs, like OIDs. An advantage of a new data type is
that it becomes easier to grep for it in the code when assigning this
type to a catalog attribute, linking it to dedicated APIs and internal
structures.
The following operators are added in this commit, with dedicated tests:
- Casts with integer types and OID.
- btree and hash operators
- min/max functions.
- C type with related macros and defines, named around "Oid8".
This has been mentioned as useful on its own on the thread to add
support for 64-bit TOAST values, so as it becomes possible to attach
this data type to the TOAST code and catalog definitions. However, as
this concept can apply to many more areas, it is implemented as its own
independent change. This is based on a discussion with Andres Freund
and Tom Lane.
Bump catalog version.
Author: Michael Paquier <michael@paquier.xyz>
Reviewed-by: Greg Burd <greg@burd.me>
Reviewed-by: Nikhil Kumar Veldanda <veldanda.nikhilkumar17@gmail.com>
Discussion: https://postgr.es/m/1891064.1754681536@sss.pgh.pa.us
In a7f107df2 I changed subplan param evaluation to happen within the
containing expression. As part of that, ExecInitSubPlanExpr() was changed to
evaluate parameters via a new EEOP_PARAM_SET expression step. These parameters
were temporarily stored into ExprState->resvalue/resnull, with some reasoning
why that would be fine. Unfortunately, that analysis was wrong -
ExecInitSubscriptionRef() evaluates the input array into "resv"/"resnull",
which will often point to ExprState->resvalue/resnull. This means that the
EEOP_PARAM_SET, if inside an array subscript, would overwrite the input array
to array subscript.
The fix is fairly simple - instead of evaluating into
ExprState->resvalue/resnull, store the temporary result of the subplan in the
subplan's return value.
Bug: #19370
Reported-by: Zepeng Zhang <redraiment@gmail.com>
Diagnosed-by: Tom Lane <tgl@sss.pgh.pa.us>
Diagnosed-by: Andres Freund <andres@anarazel.de>
Discussion: https://postgr.es/m/19370-7fb7a5854b7618f1@postgresql.org
Backpatch-through: 18
The new test 002_worker_terminate relies on the generation of a LOG
entry to check that a worker has been started, but missed the fact that
a node set with log_error_verbosity = verbose would add an error code.
The regexp used for the matching check did not take this case into
account, making the test fail on a timeout. The regexp is now fixed to
handle the verbose case correctly.
Per buildfarm member prion, that uses log_error_verbosity = verbose.
The error was reproducible by setting this GUC the same way in the test.
Oversight in f1e251be80.
Background workers gain a new flag, called BGWORKER_INTERRUPTIBLE, that
offers the possibility to terminate the workers when these are connected
to a database that is involved in one of the following commands:
ALTER DATABASE RENAME TO
ALTER DATABASE SET TABLESPACE
CREATE DATABASE
DROP DATABASE
This is useful to give background workers the same behavior as backends
and autovacuum workers, which are stopped when these commands are
executed. The default behavior, that exists since 9.3, is still to
never terminate bgworkers connected to the database involved in any of
these commands. The new flag has to be set to terminate the workers.
A couple of tests are added to worker_spi to track the commands that
impact the termination of the workers. There is a test case for a
non-interruptible worker, additionally, that relies on an injection
point to make the wait time in CountOtherDBBackends() reduced from 5s to
0.3s for faster test runs. The tests rely on the contents of the server
logs to check if a worker has been started or terminated:
- LOG generated by worker_spi_main() at startup, once connection to
database is done.
- FATAL in bgworker_die() when terminated.
A couple of tests run in the CI have showed that this method is stable
enough. The safe_psql() calls that scan pg_stat_activity could be
replaced with some poll_query_until() for more stability, if the current
method proves to be an issue in the buildfarm.
Author: Aya Iwata <iwata.aya@fujitsu.com>
Reviewed-by: Hayato Kuroda <kuroda.hayato@fujitsu.com>
Reviewed-by: Chao Li <li.evan.chao@gmail.com>
Reviewed-by: Ryo Matsumura <matsumura.ryo@fujitsu.com>
Reviewed-by: Michael Paquier <michael@paquier.xyz>
Reviewed-by: Peter Smith <smithpb2250@gmail.com>
Discussion: https://postgr.es/m/OS7PR01MB11964335F36BE41021B62EAE8EAE4A@OS7PR01MB11964.jpnprd01.prod.outlook.com
When processing the "publish" options of an ALTER PUBLICATION command,
we call SplitIdentifierString() to split the options into a List of
strings. Since SplitIdentifierString() modifies the delimiter
character and puts NULs in their place, this would overwrite the memory
of the AlterPublicationStmt. Later in AlterPublicationOptions(), the
modified AlterPublicationStmt is copied for event triggers, which would
result in the event trigger only seeing the first "publish" option
rather than all options that were specified in the command.
To fix this, make a copy of the string before passing to
SplitIdentifierString().
Here we also adjust a similar case in the pgoutput plugin. There's no
known issues caused by SplitIdentifierString() here, so this is being
done out of paranoia.
Thanks to Henson Choi for putting together an example case showing the
ALTER PUBLICATION issue.
Author: sunil s <sunilfeb26@gmail.com>
Reviewed-by: Henson Choi <assam258@gmail.com>
Reviewed-by: zengman <zengman@halodbtech.com>
Backpatch-through: 14
Prior to v15, GUC settings supplied in the CONNECTION clause of
CREATE SUBSCRIPTION were correctly passed through to
the publisher's walsender. For example:
CREATE SUBSCRIPTION mysub
CONNECTION 'options=''-c wal_sender_timeout=1000'''
PUBLICATION ...
would cause wal_sender_timeout to take effect on the publisher's walsender.
However, commit f3d4019da5 changed the way logical replication
connections are established, forcing the publisher's relevant
GUC settings (datestyle, intervalstyle, extra_float_digits) to
override those provided in the CONNECTION string. As a result,
from v15 through v18, GUC settings in the CONNECTION string were
always ignored.
This regression prevented per-connection tuning of logical replication.
For example, using a shorter timeout for walsender connecting
to a nearby subscriber and a longer one for walsender connecting
to a remote subscriber.
This commit restores the intended behavior by ensuring that
GUC settings in the CONNECTION string are again passed through
and applied by the walsender, allowing per-connection configuration.
Backpatch to v15, where the regression was introduced.
Author: Fujii Masao <masao.fujii@gmail.com>
Reviewed-by: Chao Li <lic@highgo.com>
Reviewed-by: Kirill Reshke <reshkekirill@gmail.com>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Reviewed-by: Japin Li <japinli@hotmail.com>
Discussion: https://postgr.es/m/CAHGQGwGYV+-abbKwdrM2UHUe-JYOFWmsrs6=QicyJO-j+-Widw@mail.gmail.com
Backpatch-through: 15
