As currently designed, with a callback registered in a ERROR_CLEANUP
block, the shutdown callback would get called twice when updating
archive_library on SIGHUP, which is something that we want to avoid to
ease the life of extension writers.
Anyway, an ERROR in the archiver process is treated as a FATAL, stopping
it immediately, hence there is no need for a ERROR_CLEANUP block.
Instead of that, the shutdown callback is not called upon
before_shmem_exit(), giving to the modules the opportunity to do any
cleanup actions before the server shuts down its subsystems.
While on it, this commit adds some testing coverage for the shutdown
callback. Neither shell_archive nor basic_archive have been using it,
and one is added to shell_archive, whose trigger is checked in a TAP
test through a shutdown sequence.
Author: Nathan Bossart, Bharath Rupireddy
Reviewed-by: Kyotaro Horiguchi, Michael Paquier
Discussion: https://postgr.es/m/20221015221328.GB1821022@nathanxps13
Backpatch-through: 15
AuthToken gains a regular expression, and IdentLine is changed so as it
uses an AuthToken rather than tracking separately the ident user string
used for the regex compilation and its generated regex_t. In the case
of pg_ident.conf, a set of AuthTokens is built in the pre-parsing phase
of the file, and an extra regular expression is compiled when building
the list of IdentLines, after checking the sanity of the fields in a
pre-parsed entry.
The logic in charge of computing and executing regular expressions is
now done in a new set of routines called respectively
regcomp_auth_token() and regexec_auth_token() that are wrappers around
pg_regcomp() and pg_regexec(), working on AuthTokens. While on it, this
patch adds a routine able to free an AuthToken, free_auth_token(), to
simplify a bit the logic around the requirement of using a specific free
routine for computed regular expressions. Note that there are no
functional or behavior changes introduced by this commit.
The goal of this patch is to ease the use of regular expressions with
more items of pg_hba.conf (user list, database list, potentially
hostnames) where AuthTokens are used extensively. This will be tackled
later in a separate patch.
Author: Bertrand Drouvot, Michael Paquier
Discussion: https://postgr.es/m/fff0d7c1-8ad4-76a1-9db3-0ab6ec338bf7@amazon.com
Preprocessing of the HAVING clause will reduce havingQual to NIL
if the clause is constant-TRUE. This is one case where that
convention is rather unfortunate, because "HAVING TRUE" is not at all
the same as not having any HAVING clause at all. (Per the SQL spec,
it still forces the query to be grouped.) The planner deals with this
by having a boolean hasHavingQual that records whether havingQual was
originally nonempty; places that just want to check whether HAVING
was specified are supposed to consult that.
I found three places that got that wrong. Fortunately, these could
only affect cost estimates not correctness. It'd be hard even
to demonstrate the errors; for example, the one in allpaths.c would
only matter in a query that has HAVING TRUE but no GROUP BY and no
aggregates, which would require a completely variable-free SELECT
list, making the case probably of only academic interest. Hence,
while these are worth fixing before someone copies the incorrect
coding somewhere more critical, they don't seem worth back-patching.
I didn't bother trying to devise regression tests, either.
Discussion: https://postgr.es/m/2503888.1666042643@sss.pgh.pa.us
The original hint says to use SET PUBLICATION when really ADD/DROP
PUBLICATION is called for, so this is arguably a bug fix.
Also, a very similar message elsewhere was using an inconsistent
SQLSTATE.
While at it, unwrap some strings.
Backpatch to 15.
Author: Hou zj <houzj.fnst@fujitsu.com>
Discussion: https://postgr.es/m/OS0PR01MB57160AD0E7386547BA978EB394299@OS0PR01MB5716.jpnprd01.prod.outlook.com
These routines have been renamed in a19e5ce. There is no need to keep
the compatibility declarations on HEAD, as once an extension moves to
the new routine name when compiling with v16~ the code would work the
same way when recompiled on v15. No backpatch to v15 for this one,
because ABI compatibility has to be maintained there.
Discussion: https://postgr.es/m/20221013194820.ciktb2sbbpw7cljm@awork3.anarazel.de
Per discussion, the existing routine name able to initialize a SRF
function with materialize mode is unpopular, so rename it. Equally, the
flags of this function are renamed, as of:
- SRF_SINGLE_USE_EXPECTED -> MAT_SRF_USE_EXPECTED_DESC
- SRF_SINGLE_BLESS -> MAT_SRF_BLESS
The previous function and flags introduced in 9e98583 are kept around
for compatibility purposes, so as any extension code already compiled
with v15 continues to work as-is. The declarations introduced here for
compatibility will be removed from HEAD in a follow-up commit.
The new names have been suggested by Andres Freund and Melanie
Plageman.
Discussion: https://postgr.es/m/20221013194820.ciktb2sbbpw7cljm@awork3.anarazel.de
Backpatch-through: 15
When creating a cast that uses a conversion function, we've
historically allowed the input and result types to be
binary-compatible with the function's input and result types,
rather than necessarily being identical. This means that the new
cast is logically dependent on the binary-compatible cast or casts
that it references: if those are defined by pg_cast entries, and you
try to restore the new cast without having defined them, it'll fail.
Hence, we should make pg_depend entries to record these dependencies
so that pg_dump knows that there is an ordering requirement.
This is not the only place where we allow such shortcuts; aggregate
functions for example are similarly lax, and in principle should gain
similar dependencies. However, for now it seems sufficient to fix
the cast-versus-cast case, as pg_dump's other ordering heuristics
should keep it out of trouble for other object types.
Per report from David Turoň; thanks also to Robert Haas for
preliminary investigation. I considered back-patching, but
seeing that this issue has existed for many years without
previous reports, it's not clear it's worth the trouble.
Moreover, back-patching wouldn't be enough to ensure that the
new pg_depend entries exist in existing databases anyway.
Discussion: https://postgr.es/m/OF0A160F3E.578B15D1-ONC12588DA.003E4857-C12588DA.0045A428@notes.linuxbox.cz
DefineQueryRewrite() has long required that ON SELECT rules be named
"_RETURN". But we overlooked the converse case: we should forbid
non-ON-SELECT rules that are named "_RETURN". In particular this
prevents using CREATE OR REPLACE RULE to overwrite a view's _RETURN
rule with some other kind of rule, thereby breaking the view.
Per bug #17646 from Kui Liu. Back-patch to all supported branches.
Discussion: https://postgr.es/m/17646-70c93cfa40365776@postgresql.org
The executor will dump core if it's asked to execute a seqscan on
a relation having no table AM, such as a view. While that shouldn't
really happen, it's possible to get there via catalog corruption,
such as a missing ON SELECT rule. It seems worth installing a defense
against that. There are multiple plausible places for such a defense,
but I picked the planner's get_relation_info().
Per discussion of bug #17646 from Kui Liu. Back-patch to v12 where
the tableam APIs were introduced; in older versions you won't get a
SIGSEGV, so it seems less pressing.
Discussion: https://postgr.es/m/17646-70c93cfa40365776@postgresql.org
If the non-recursive term of a SEARCH BREADTH FIRST recursive
query has only constants in its target list, the planner will
fold the starting RowExpr added by rewrite into a simple Const
of type RECORD. The executor doesn't have any problem with
that --- but EXPLAIN VERBOSE will encounter the Const as the
ultimate source of truth about what the field names of the
SET column are, and it didn't know what to do with that.
Fortunately, we can pull the identifying typmod out of the
Const, in much the same way that record_out would.
For reasons that remain a bit obscure to me, this only fails
with SEARCH BREADTH FIRST, not SEARCH DEPTH FIRST or CYCLE.
But I added regression test cases for both of those options
too, just to make sure we don't break it in future.
Per bug #17644 from Matthijs van der Vleuten. Back-patch
to v14 where these constructs were added.
Discussion: https://postgr.es/m/17644-3bd1f3036d6d7a16@postgresql.org
In the latest version of Apple's macOS SDK, <sys/socket.h>
fails to compile if "REF" is #define'd as something.
Apple may or may not agree that this is a bug, and even if
they do accept the bug report I filed, they probably won't
fix it very quickly. In the meantime, our back branches will all
fail to compile gram.y. v15 and HEAD currently escape the problem
thanks to the refactoring done in 98e93a1fc, but that's purely
accidental. Moreover, since that patch removed a widely-visible
inclusion of <netdb.h>, back-patching it seems too likely to break
third-party code.
Instead, change the token's code name to REF_P, following our usual
convention for naming parser tokens that are likely to have symbol
conflicts. The effects of that should be localized to the grammar
and immediately surrounding files, so it seems like a safer answer.
Per project policy that we want to keep recently-out-of-support
branches buildable on modern systems, back-patch all the way to 9.2.
Discussion: https://postgr.es/m/1803927.1665938411@sss.pgh.pa.us
While directly targetting a foreign table with MERGE was already
expressly forbidden, we failed to catch the case of a partitioned table
that has a foreign table as a partition; and the result if you try is an
incomprehensible error. Fix that by adding a specific check.
Backpatch to 15.
Reported-by: Tatsuhiro Nakamori <bt22nakamorit@oss.nttdata.com>
Discussion: https://postgr.es/m/bt22nakamorit@oss.nttdata.com
It can be useful to know when a relation has last been used, e.g., when
evaluating whether an index is still required. It was already possible to
infer the time of the last usage by tracking, e.g.,
pg_stat_all_indexes.idx_scan over time. But far from everybody does so.
To make it easier to detect the last time a relation has been scanned, track
that time in each relation's pgstat entry. To minimize overhead a) the
timestamp is updated only when the backend pending stats entry is flushed to
shared stats b) the last transaction's stop timestamp is used as the
timestamp.
Bumps catalog and stats format versions.
Author: Dave Page <dpage@pgadmin.org>
Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Bruce Momjian <bruce@momjian.us>
Reviewed-by: Vik Fearing <vik@postgresfriends.org>
Discussion: https://postgr.es/m/CA+OCxozrVHNFVEPkweUHMZje+t1tfY816d9MZYc6eZwOOusOaQ@mail.gmail.com
Previously GetCurrentTransactionStopTimestamp() computed a new timestamp
whenever xactStopTimestamp was unset and xactStopTimestamp was only set when a
commit or abort record was written.
An upcoming patch will add additional calls to
GetCurrentTransactionStopTimestamp() from pgstats. To avoid computing
timestamps multiple times, set xactStopTimestamp in
GetCurrentTransactionStopTimestamp() if not already set.
Author: Dave Page <dpage@pgadmin.org>
Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Vik Fearing <vik@postgresfriends.org>
Discussion: https://postgr.es/m/20220906155325.an3xesq5o3fq36gt%40awork3.anarazel.de
The previous patch made addition of new GUCs cheap, but other GUC
operations aren't improved and indeed get a bit slower, because
hash_seq_search() is slower than just scanning a pointer array.
However, most performance-critical GUC operations only need
to touch a relatively small fraction of the GUCs; especially
so for AtEOXact_GUC(). We can improve matters at the cost
of a bit more space by adding dlist or slist links to the
GUC data structures. This patch invents lists that track
(1) all GUCs with non-default "source";
(2) all GUCs with nonempty state stack (implying they've
been changed in the current transaction);
(3) all GUCs due for reporting to the client.
All of guc.c's performance-critical cases can make use of one or
another of these lists to avoid searching the whole hash table.
In particular, the stack list means that transaction end
doesn't take time proportional to the number of GUCs, but
only to the number changed in the current transaction.
Discussion: https://postgr.es/m/2982579.1662416866@sss.pgh.pa.us
This gets rid of bsearch() in favor of hashed lookup. The main
advantage is that it becomes far cheaper to add new GUCs, since
we needn't re-sort the pointer array. Adding N new GUCs had
been O(N^2 log N), but now it's closer to O(N). We need to
sort only in SHOW ALL and equivalent functions, which are
hopefully not performance-critical to anybody.
Also, merge GetNumConfigOptions() into get_guc_variables(),
because in a world where the set of GUCs isn't fairly static
you really want to consider those two results as tied together
not independent.
Discussion: https://postgr.es/m/2982579.1662416866@sss.pgh.pa.us
The only real argument for using malloc directly was that we needed
the ability to not throw error on OOM; but mcxt.c grew that feature
awhile ago.
Keeping the data in a memory context improves accountability and
debuggability --- for example, without this it's almost impossible
to detect memory leaks in the GUC code with anything less costly
than valgrind. Moreover, the next patch in this series will add a
hash table for GUC lookup, and it'd be pretty silly to be using
palloc-dependent hash facilities alongside malloc'd storage of the
underlying data.
This is a bit invasive though, in particular causing an API break
for GUC check hooks that want to modify the GUC's value or use an
"extra" data structure. They must now use guc_malloc() and
guc_free() instead of malloc() and free(). Failure to change
affected code will result in assertion failures or worse; but
thanks to recent effort in the mcxt infrastructure, it shouldn't
be too hard to diagnose such oversights (at least in assert-enabled
builds).
One note is that this changes ParseLongOption() to return short-lived
palloc'd not malloc'd data. There wasn't any caller for which the
previous definition was better.
Discussion: https://postgr.es/m/2982579.1662416866@sss.pgh.pa.us
We lack a version of repalloc() that supports MCXT_ALLOC_NO_OOM
semantics, so invent repalloc_extended() with the usual set of
flags. repalloc_huge() becomes a legacy wrapper for that.
Also, fix dynahash.c so that it can support HASH_ENTER_NULL
requests when using the default palloc-based allocator.
The only reason it didn't do that already was the lack of the
MCXT_ALLOC_NO_OOM option when that code was written, ages ago.
While here, simplify a few overcomplicated tests in mcxt.c.
Discussion: https://postgr.es/m/2982579.1662416866@sss.pgh.pa.us
Most code prints PIDs as %d, but some code tried to print them as long
or unsigned long. While this is in theory allowed, the fact that PIDs
fit into int is deeply baked into all PostgreSQL code, so these random
deviations don't accomplish anything except confusion.
Note that we still need casts from pid_t to int, because on 64-bit
MinGW, pid_t is long long int. (But per above, actually supporting
that range in PostgreSQL code would be major surgery and probably not
useful.)
Discussion: https://www.postgresql.org/message-id/289c2e45-c7d9-5ce4-7eff-a9e2a33e1580@enterprisedb.com
The comment talked about some Asserts which did not exist and also a
variable name which seems to have long since disappeared.
Rewrite the comment in a way that will hopefully stand the test of
time and inform people why we always write "INSERT 0 <nrows>" instead of
"INSERT <nrows>" in the command completion tag for INSERT.
Reviewed-by: Mark Dilger
Discussion: https://postgr.es/m/CAApHDvpiUg09AvvGAVopNAKemA9z-kCmt7Fi6HKauc32bKzx4w@mail.gmail.com
Commit 3d956d956 allowed the COPY, but it's done by inserting individual
rows to the foreign table, so it can be inefficient due to the overhead
caused by each round-trip to the foreign server. To improve performance
of the COPY in such a case, this patch allows batch insertion, by
extending the multi-insert machinery in CopyFrom() to the foreign-table
case so that we insert multiple rows to the foreign table at once using
the FDW callback routine added by commit b663a4136. This patch also
allows this for postgres_fdw. It is enabled by the "batch_size" option
added by commit b663a4136, which is disabled by default.
When doing batch insertion, we update progress of the COPY command after
performing the FDW callback routine, to count rows not suppressed by the
FDW as well as a BEFORE ROW INSERT trigger. For consistency, this patch
changes the timing of updating it for plain tables: previously, we
updated it immediately after adding each row to the multi-insert buffer,
but we do so only after writing the rows stored in the buffer out to the
table using table_multi_insert(), which I think would be consistent even
with non-batching mode, because in that mode we update it after writing
each row out to the table using table_tuple_insert().
Andrey Lepikhov, heavily revised by me, with review from Ian Barwick,
Andrey Lepikhov, and Zhihong Yu.
Discussion: https://postgr.es/m/bc489202-9855-7550-d64c-ad2d83c24867%40postgrespro.ru
Contrary to what is documented in src/backend/access/transam/README,
ginHeapTupleFastInsert() had a few ordering issues with the way it does
its WAL operations when inserting items in its fast path.
First, when using a separate list, XLogBeginInsert() was being always
called before START_CRIT_SECTION(), and in this case a second thing was
wrong when merging lists, as an exclusive lock was taken on the tail
page *before* calling XLogBeginInsert(). Finally, when inserting items
into a tail page, the order of XLogBeginInsert() and
START_CRIT_SECTION() was reversed. This commit addresses all these
issues by moving the calls of XLogBeginInsert() after all the pages
logged are locked and pinned, within a critical section.
Author: Matthias van de Meent, Zhang Mingli
Discussion: https://postgr.es/m/CAEze2WhL8uLMqynnnCu1LAPwxD5RKEo0nHV+eXGg_N6ELU88HQ@mail.gmail.com
An LSN was calculated from a segment number, a segment size and a
position offset, matching exactly the LSN given by the caller of
XLogReaderValidatePageHeader(). This change removes the extra LSN
calculation, relying only on the LSN given by the function caller
instead.
Author: Bharath Rupireddy
Reviewed-by: Richard Guo, Álvaro Herrera, Kyotaro Horiguchi
Discussion: https://postgr.es/m/CALj2ACXuh4Ms9j9sxMYdtHEe=5sFcyrs-GAHyADu_A_G71kZTg@mail.gmail.com
The postmaster is not supposed to do anything that depends
fundamentally on shared memory contents, because that creates
the risk that a backend crash that trashes shared memory will
take the postmaster down with it, preventing automatic recovery.
In commit 969d7cd43 I lost sight of this principle and coded
AssignPostmasterChildSlot() in such a way that it could fail
or even crash if the shared PMSignalState structure became
corrupted. Remarkably, we've not seen field reports of such
crashes; but I managed to induce one while testing the recent
changes around palloc chunk headers.
To fix, make a semi-duplicative state array inside the postmaster
so that we need consult only local state while choosing a "child
slot" for a new backend. Ensure that other postmaster-executed
routines in pmsignal.c don't have critical dependencies on the
shared state, either. Corruption of PMSignalState might now
lead ReleasePostmasterChildSlot() to conclude that backend X
failed, when actually backend Y was the one that trashed things.
But that doesn't matter, because we'll force a cluster-wide reset
regardless.
Back-patch to all supported branches, since this is an old bug.
Discussion: https://postgr.es/m/3436789.1665187055@sss.pgh.pa.us
DEFAULT markers appearing in an INSERT on an updatable view
could be mis-processed if they were in a multi-row VALUES clause.
This would lead to strange errors such as "cache lookup failed
for type NNNN", or in older branches even to crashes.
The cause is that commit 41531e42d tried to re-use rewriteValuesRTE()
to remove any SetToDefault nodes (that hadn't previously been replaced
by the view's own default values) appearing in "product" queries,
that is DO ALSO queries. That's fundamentally wrong because the
DO ALSO queries might not even be INSERTs; and even if they are,
their targetlists don't necessarily match the view's column list,
so that almost all the logic in rewriteValuesRTE() is inapplicable.
What we want is a narrow focus on replacing any such nodes with NULL
constants. (That is, in this context we are interpreting the defaults
as being strictly those of the view itself; and we already replaced
any that aren't NULL.) We could add still more !force_nulls tests
to further lobotomize rewriteValuesRTE(); but it seems cleaner to
split out this case to a new function, restoring rewriteValuesRTE()
to the charter it had before.
Per bug #17633 from jiye_sw. Patch by me, but thanks to
Richard Guo and Japin Li for initial investigation.
Back-patch to all supported branches, as the previous fix was.
Discussion: https://postgr.es/m/17633-98cc85e1fa91e905@postgresql.org
Make a common replication origin name formatting function to replace
multiple snprintf() expressions. This also includes logic previously done
by ReplicationOriginNameForTablesync().
This makes the code to generate the origin name consistent among apply
worker and tablesync worker.
Author: Peter Smith
Reviewed-By: Aleksander Alekseev
Discussion: https://postgr.es/m/CAHut%2BPsa8hhfSE6ozUK-ih7GkQziAVAf4f3bqiXEj2nQiu-43g%40mail.gmail.com
This is useful as a way for extensions to process COPY TO rows in the
way they see fit (say auditing, analytics, backend, etc.) without the
need to invoke an external process running as the OS user running the
backend through PROGRAM that requires superuser rights. COPY FROM
already provides a similar callback for logical replication. For COPY
TO, the callback is triggered when we are ready to send a row in
CopySendEndOfRow(), which is the same code path as when sending a row
to a frontend or a pipe/file.
A small test module, test_copy_callbacks, is added to provide some
coverage for this facility.
Author: Bilva Sanaba, Nathan Bossart
Discussion: https://postgr.es/m/253C21D1-FCEB-41D9-A2AF-E6517015B7D7@amazon.com
Before commit c6e0fe1f2, functions such as AllocSetFree could pretty
safely presume that they were given a valid chunk pointer for their
own type of context, because the indirect call through a memory
context object and method struct would be very unlikely to work
otherwise. But now, if pfree() is mistakenly invoked on a pointer
to garbage, we have three chances in eight of ending up at one of
these functions. That means we need to take extra measures to
verify that we are looking at what we're supposed to be looking at,
especially in debug builds.
Hence, add code to verify that the chunk's back-link to a block header
leads to a memory context object that satisfies the right sort of
IsA() check. This is still a bit weaker than what we did before,
but for the moment assume that an IsA() check is sufficient.
As a compromise between speed and safety, implement these checks
as Asserts when dealing with small chunks but plain test-and-elogs
when dealing with large (external) chunks. The latter case should
not be too performance-critical, but the former case probably is.
In slab.c, all chunks are small; but nonetheless use a plain test
in SlabRealloc, because that is certainly not performance-critical,
indeed we should be suspicious that it's being called in error.
In aset.c, additionally add some assertions that the "value" field
of the chunk header is within the small range allowed for freelist
indexes. Without that, we might find ourselves trying to wipe
most of memory when CLOBBER_FREED_MEMORY is enabled, or scribbling
on a "freelist header" that's far away from the context object.
Eventually, field experience might show us that it's smarter for
these tests to be active always, but for now we'll try to get
away with just having them as assertions.
While at it, also be more uniform about asserting that context
objects passed as parameters are of the type we expect. Some
places missed that altogether, and slab.c was for no very good
reason doing it differently from the other allocators.
Discussion: https://postgr.es/m/3578387.1665244345@sss.pgh.pa.us
Remove the Trap and TrapMacro macros, which were nearly unused
and confusingly had the opposite condition polarity from the
otherwise-functionally-equivalent Assert macros.
Having done that, it's very hard to justify carrying the errorType
argument of ExceptionalCondition, so drop that too, and just
let it assume everything's an Assert. This saves about 64K
of code space as of current HEAD.
Discussion: https://postgr.es/m/3928703.1665345117@sss.pgh.pa.us
Instead of Abs() for int64, use the C standard functions labs() or
llabs() as appropriate. Define a small wrapper around them that
matches our definition of int64. (labs() is C90, llabs() is C99.)
Reviewed-by: Zhang Mingli <zmlpostgres@gmail.com>
Reviewed-by: Tom Lane <tgl@sss.pgh.pa.us>
Discussion: https://www.postgresql.org/message-id/flat/4beb42b5-216b-bce8-d452-d924d5794c63%40enterprisedb.com
Previously PgStat_StatReplSlotEntry contained the slotname, which was mainly
used when writing out the stats during shutdown, to identify the slot in the
serialized data (at runtime the index in ReplicationSlotCtl->replication_slots
is used, but that can change during a restart). Unfortunately the slotname was
overwritten when the slot's stats were reset.
That turned out to only cause "real" problems if the slot was active during
the reset, triggering an assertion failure at the next
pgstat_report_replslot(). In other paths the stats were re-initialized during
pgstat_acquire_replslot().
Fix this by removing slotname from PgStat_StatReplSlotEntry. Instead we can
get the slot's name from the slot itself. Besides fixing a bug, this also is
architecturally cleaner (a name is not really statistics). This is safe
because stats, for a slot removed while shut down, will not be restored at
startup.
In 15 the slotname is not removed, but renamed, to avoid changing the stats
format. In master, bump PGSTAT_FILE_FORMAT_ID.
This commit does not contain a test for the fix. I think this can only be
tested by a tap test starting pg_recvlogical in the background and checking
pg_recvlogical's output. That type of test is notoriously hard to be reliable,
so committing it shortly before the release is wrapped seems like a bad idea.
Reported-by: Jaime Casanova <jcasanov@systemguards.com.ec>
Author: Andres Freund <andres@anarazel.de>
Reviewed-by: Masahiko Sawada <sawada.mshk@gmail.com>
Reviewed-by: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Discussion: https://postgr.es/m/YxfagaTXUNa9ggLb@ahch-to
Backpatch: 15-, where the bug was introduced in 5891c7a8ed
There are a number of bugs in this area. Two of them are fixed here,
namely:
1. get_relation_idx_constraint_oid does not restrict the type of
constraint that's returned, so with sufficient bad luck it can
return the OID of a foreign key constraint. This has the effect that
a primary key in a partition can end up as a child of a foreign key,
which makes no sense (it needs to be the child of the equivalent
primary key.)
Change the API contract so that only index-backed constraints are
returned, mimicking get_constraint_index().
2. Both CloneFkReferenced and CloneFkReferencing clone a
self-referencing foreign key, so the partition ends up with
a duplicate foreign key. Change the former function to ignore such
constraints.
Add some tests to verify that things are better now. (However, these
new tests show some additional misbehavior that will be fixed later --
namely that there's a constraint marked NOT VALID.)
Backpatch to 12, where these constraints are possible at all.
Author: Jehan-Guillaume de Rorthais <jgdr@dalibo.com>
Discussion: https://postgr.es/m/20220603154232.1715b14c@karst
Commit c6e0fe1f2 was a shade too trusting that any pointer passed
to pfree, repalloc, etc will point at a valid chunk. Notably,
passing a pointer that was actually obtained from malloc tended
to result in obscure assertion failures, if not worse. (On FreeBSD
I've seen such mistakes take down the entire cluster, seemingly as
a result of clobbering shared memory.)
To improve matters, extend the mcxt_methods[] array so that it
has entries for every possible MemoryContextMethodID bit-pattern,
with the currently unassigned ID codes pointing to error-reporting
functions. Then, fiddle with the ID assignments so that patterns
likely to be associated with bad pointers aren't valid ID codes.
In particular, we should avoid assigning bit patterns 000 (zeroed
memory) and 111 (wipe_mem'd memory).
It turns out that on glibc (Linux), malloc uses chunk headers that
have flag bits in the same place we keep MemoryContextMethodID,
and that the bit patterns 000, 001, 010 are the only ones we'll
see as long as the backend isn't threaded. So we can have very
robust detection of pfree'ing a malloc-assigned block on that
platform, at least so long as we can refrain from using up those
ID codes. On other platforms, we don't have such a good guarantee,
but keeping 000 reserved will be enough to catch many such cases.
While here, make GetMemoryChunkMethodID() local to mcxt.c, as there
seems no need for it to be exposed even in memutils_internal.h.
Patch by me, with suggestions from Andres Freund and David Rowley.
Discussion: https://postgr.es/m/2910981.1665080361@sss.pgh.pa.us
This substantially speeds up building for windows, due to the vast amount of
headers included via windows.h. A cross build from linux targetting mingw goes
from
994.11user 136.43system 0:31.58elapsed 3579%CPU
to
422.41user 89.05system 0:14.35elapsed 3562%CPU
The wins on windows are similar-ish (but I don't have a system at hand just
now for actual numbers). Targetting other operating systems the wins are far
smaller (tested linux, macOS, FreeBSD).
For now precompiled headers are disabled by default, it's not clear how well
they work on all platforms. E.g. on FreeBSD gcc doesn't seem to have working
support, but clang does.
When doing a full build precompiled headers are only beneficial for targets
with multiple .c files, as meson builds a separate precompiled header for each
target (so that different compilation options take effect). This commit
therefore only changes target with at least two .c files to use precompiled
headers.
Because this commit adds b_pch=false to the default_options new build
directories will have precompiled headers disabled by default, however
existing build directories will continue use the default value of b_pch, which
is true.
Note that using precompiled headers with ccache requires setting
CCACHE_SLOPPINESS=pch_defines,time_macros to get hits.
Reviewed-by: Peter Eisentraut <peter.eisentraut@enterprisedb.com>
Reviewed-by: Justin Pryzby <pryzby@telsasoft.com>
Discussion: https://postgr.es/m/CA+hUKG+50eOUbN++ocDc0Qnp9Pvmou23DSXu=ZA6fepOcftKqA@mail.gmail.com
Discussion: https://postgr.es/m/c5736f70-bb6d-8d25-e35c-e3d886e4e905@enterprisedb.com
Discussion: https://postgr.es/m/20190826054000.GE7005%40paquier.xyz
Previously, the subscription stats entry was created when the first
stats, i.e., an error on apply worker or tablesync worker, were
reported. Therefore, the stats_reset field was not updated by
pg_stat_reset_subscription_stats() if the stats entry was not
populated yet, which was different behavior than other statistics.
This change creates the subscription stats entry and initializes it at
CREATE SUBSCRIPTION time.
Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Amit Kapila <amit.kapila16@gmail.com>
Author: Masahiko Sawada <sawada.mshk@gmail.com>
Discussion: https://postgr.es/m/CAAKRu_Zqd-e5imT_3-ZiQv1cfsWuy16OJTiUaCvqpq4V7GVdSg@mail.gmail.com
MemoryContextContains is no longer reliable in the wake of c6e0fe1f2,
because there's no longer very much redundancy in chunk headers.
(It wasn't *completely* reliable even before that, as there was a
chance of a false positive if you passed it something that didn't
point to an mcxt chunk at all. But it was generally good enough.)
Hence, remove it. There is no remaining core code that requires it.
Extensions that have been using it might be able to substitute a
test like "GetMemoryChunkContext(ptr) == context", recognizing that
this explicitly requires that the pointer point to some chunk.
Tom Lane and David Rowley
Discussion: https://postgr.es/m/1913788.1664898906@sss.pgh.pa.us
MemoryContextContains is no longer reliable in the wake of c6e0fe1f2,
so we need to get rid of these uses.
It appears that there's no really good reason to force the result of
an aggregate's finalfn or serialfn to be allocated in the per-tuple
context. The only other plausible case is that the result points to
or into the aggregate's transition value, and that's fine because it
will last as long as we need it to. (This conclusion depends on the
assumption that finalfns are not allowed to scribble on the transition
value, but we've long required that.) So we can just drop the
MemoryContextContains plus datumCopy business, although we do need
to take care to not return a read-write pointer when the transition
value is an expanded datum.
Likewise, we don't really need to force the result of a window
function to be in the output context. In this case, the plausible
alternative is that it's pointing into the temporary tuple slot used
by WinGetFuncArgInPartition or WinGetFuncArgInFrame (since those
functions could return such a pointer, which might become the window
function's result). That will hold still for long enough, unless
there is another window function using the same WindowObject.
I'm content to always perform a datumCopy when there's more than one
such function.
On net, these changes should provide small speed improvements as well
as removing problematic code.
Tom Lane and David Rowley
Discussion: https://postgr.es/m/1913788.1664898906@sss.pgh.pa.us
The RecoveryLockLists data structure, which tracks all exclusive
locks that the startup process is holding on behalf of transactions
being replayed, did not have any provision for avoiding duplicate
entries for the same lock. Maybe that was okay when the code was
first written. However, modern practice is for checkpoints to
write fresh lists of all active exclusive locks into the WAL.
Thus, an exclusive lock that survives across multiple checkpoints
causes bloat in standbys' startup processes. If there are a lot
of such locks this can look like a memory leak, and it's even
possible to drive the startup process into a palloc failure from
an over-length List.
To fix, use a hash table instead of simple lists to track the
locks being held. Allowing for dynahash overhead, this requires
a little more space per lock than the old way (although it's the
same size as what we were allocating prior to c6e0fe1f2). It's
probably a shade slower too. However, testing indicates that the
penalty is negligible on ordinary workloads, so let's make this
change to improve robustness in extreme cases.
Patch by me, per report from Dmitriy Kuzmin. No back-patch
(for now anyway), since it seems that a significant improvement
would only occur in corner cases.
Discussion: https://postgr.es/m/CAHLDt=_ts0A7Agn=hCpUh+RCFkxd+G6uuT=kcTfqFtGur0dp=A@mail.gmail.com
ts_locale.c omitted support for "isalnum" tests, perhaps on the
grounds that there were initially no use-cases for that. However,
both ltree and pg_trgm need such tests, and we do also have one
use-case now in the core backend. The workaround of testing
isalpha and isdigit separately seems quite inefficient, especially
when dealing with multibyte characters; so let's fill in the
missing support.
Discussion: https://postgr.es/m/2548310.1664999615@sss.pgh.pa.us
This optional parameter can be specified in cases where there are nested
PG_TRY() statements within a function in order to stop the compiler from
issuing warnings about shadowed local variables when compiling with
-Wshadow. The optional parameter is used as a suffix on the variable
names declared within the PG_TRY(), PG_CATCH(), PG_FINALLY() and
PG_END_TRY() macros. The parameter, if specified, must be the same in
each component macro of the given PG_TRY() block.
This also adjusts the single case where we have nested PG_TRY() statements
to add a parameter to the inner-most PG_TRY().
This reduces the number of compiler warnings when compiling with
-Wshadow=compatible-local from 5 down to 1.
Author: David Rowley
Discussion: https://postgr.es/m/CAApHDvqWGMdB_pATeUqE=JCtNqNxObPOJ00jFEa2_sZ20j_Wvg@mail.gmail.com
The generated resource files aren't exactly the same ones as the old
buildsystems generate. Previously "InternalName" and "OriginalFileName" were
mostly wrong / not set (despite being required), but that was hard to fix in
at least the make build. Additionally, the meson build falls back to a
"auto-generated" description when not set, and doesn't set it in a few cases -
unlikely that anybody looks at these descriptions in detail.
Author: Andres Freund <andres@anarazel.de>
Author: Nazir Bilal Yavuz <byavuz81@gmail.com>
Reviewed-by: Peter Eisentraut <peter.eisentraut@enterprisedb.com>
Michael Paquier
Tom Lane
Alvaro Herrera
Andres Freund
Peter Eisentraut
David Rowley
Etsuro Fujita
Amit Kapila
John Naylor