We have repeatedly seen the buildfarm reach the Assert(false) in
SyncRepGetSyncStandbysPriority. This apparently is due to failing to
consider the possibility that the sync_standby_priority values in
shared memory might be inconsistent; but they will be whenever only
some of the walsenders have updated their values after a change in
the synchronous_standby_names setting. That function is vastly too
complex for what it does, anyway, so rewriting it seems better than
trying to apply a band-aid fix.
Furthermore, the API of SyncRepGetSyncStandbys is broken by design:
it returns a list of WalSnd array indexes, but there is nothing
guaranteeing that the contents of the WalSnd array remain stable.
Thus, if some walsender exits and then a new walsender process
takes over that WalSnd array slot, a caller might make use of
WAL position data that it should not, potentially leading to
incorrect decisions about whether to release transactions that
are waiting for synchronous commit.
To fix, replace SyncRepGetSyncStandbys with a new function
SyncRepGetCandidateStandbys that copies all the required data
from shared memory while holding the relevant mutexes. If the
associated walsender process then exits, this data is still safe to
make release decisions with, since we know that that much WAL *was*
sent to a valid standby server. This incidentally means that we no
longer need to treat sync_standby_priority as protected by the
SyncRepLock rather than the per-walsender mutex.
SyncRepGetSyncStandbys is no longer used by the core code, so remove
it entirely in HEAD. However, it seems possible that external code is
relying on that function, so do not remove it from the back branches.
Instead, just remove the known-incorrect Assert. When the bug occurs,
the function will return a too-short list, which callers should treat
as meaning there are not enough sync standbys, which seems like a
reasonably safe fallback until the inconsistent state is resolved.
Moreover it's bug-compatible with what has been happening in non-assert
builds. We cannot do anything about the walsender-replacement race
condition without an API/ABI break.
The bogus assertion exists back to 9.6, but 9.6 is sufficiently
different from the later branches that the patch doesn't apply at all.
I chose to just remove the bogus assertion in 9.6, feeling that the
probability of a bad outcome from the walsender-replacement race
condition is too low to justify rewriting the whole patch for 9.6.
Discussion: https://postgr.es/m/21519.1585272409@sss.pgh.pa.us
Apparently in some language versions of Visual Studio nmake outputs some
material after the version number and before the end of the line. This
has been seen in Chinese versions. Therefore, we no longer demand that
the version string comes at the end of a line.
Per complaint from Cuiping Lin.
Backpatch to all live branches.
The result of the query used to retrieve the WAL segment size from the
backend was not getting freed in two code paths. Both pg_basebackup and
pg_receivewal exit immediately if a failure happened on this query, so
this was not an actual problem, but it could be an issue if this code
gets used for other tools in different ways, be they future tools in
this code tree or external, existing, ones.
Oversight in commit fc49e24, so backpatch down to 11.
Author: Jie Zhang
Discussion: https://postgr.es/m/970ad9508461469b9450b64027842331@G08CNEXMBPEKD06.g08.fujitsu.local
Backpatch-through: 11
fixup_whole_row_references() did the wrong thing with a dropped column,
resulting in a commit-time warning about a cache reference leak.
I (tgl) added a test case exercising this, but back-patched the test
only as far as v10; the patch didn't apply cleanly to 9.6 and it
didn't seem worth the trouble to adapt it. The bug is pretty old
though, so apply the code change all the way back.
Michael Luo, with cosmetic improvements by me
Discussion: https://postgr.es/m/BYAPR08MB5606D1453D7F50E2AF4D2FD29AD80@BYAPR08MB5606.namprd08.prod.outlook.com
ExecReScanHashJoin will destroy the join's hash table if it expects
that the inner relation will produce different rows on rescan.
Up to now it's not bothered to clear the additional pointer to that
hash table that exists in the child HashState node. However, it's
possible for the query to terminate without building a fresh hash
table (this happens if the outer relation is found to be empty
during the final rescan). So we can end with a dangling pointer
to a deleted hash table. That was harmless originally, but since
9.0 EXPLAIN ANALYZE has used that pointer to print hash table
statistics. In debug builds this reproducibly results in garbage
statistics. In non-debug builds there's frequently no ill effects,
but in principle one could get wrong EXPLAIN ANALYZE output, or
perhaps even a crash if free() has released the hashtable memory
back to the OS.
To fix, just make sure we clear the additional pointer when destroying
the hash table. In problematic cases, EXPLAIN ANALYZE will then print
no hashtable statistics (reverting to its pre-9.0 behavior). This isn't
ideal, but since the problem manifests only in unusual corner cases,
it's hard to justify taking any risks to do better in the back
branches. A follow-on patch will improve matters in HEAD.
Konstantin Knizhnik and Tom Lane, per diagnosis by Thomas Munro
of a trouble report from Alvaro Herrera.
Discussion: https://postgr.es/m/20200323165059.GA24950@alvherre.pgsql
The docs explained that a SHARE ROW EXCLUSIVE lock is needed on the
referenced table, but failed to say the same about the table being
altered. Since the page says that ACCESS EXCLUSIVE lock is taken
unless otherwise stated, this left readers with the wrong conclusion.
Discussion: https://postgr.es/m/834603375.3470346.1586482852542@mail.yahoo.com
CREATE GROUP is an exact alias for CREATE ROLE, and CREATE USER is
almost an exact alias, as can easily be confirmed by checking the
code. So the man page syntax descriptions ought to match up. The
last few additions of role options seem to have forgotten to update
create_group.sgml, though. Fix that, and add a naggy reminder to
create_role.sgml in hopes of not forgetting again.
Discussion: https://postgr.es/m/158647836143.655.9853963229391401576@wrigleys.postgresql.org
Suppress a probably-meaningless uninitialized-variable warning
(induced by my previous patch, I'm sorry to say).
Improve mark_hl_fragments()'s test for overlapping cover strings:
it failed to consider the possibility that the current string is
strictly within another one. That's unlikely given the preceding
splitting into MaxWords fragments, but I don't think it's impossible.
Discussion: https://postgr.es/m/16345-2e0cf5cddbdcd3b4@postgresql.org
This code could produce very poor results when asked to highlight a
string based on a query using phrase-match operators. The root cause
is that hlCover(), which is supposed to find a minimal substring that
matches the query, was written assuming that word position is not
significant. I'm only 95% convinced that its algorithm was correct even
for plain AND/OR queries; but it definitely fails completely for phrase
matches, causing it to possibly not identify a cover string at all.
Hence, rewrite hlCover() with a less-tense algorithm that just tries
all the possible substrings, earlier and shorter ones first. (This is
not as bad as it sounds performance-wise, because all of the string
matching has been done already: the repeated tsquery match checks
boil down to pointer comparisons.)
Unfortunately, since that approach produces more candidate cover
strings than before, it also exposes that there were bugs in the
heuristics in mark_hl_words() for selecting a best cover string.
Fixes there include:
* Do not apply the ShortWord filter to words that appear in the query.
* Remove a misguided optimization for quickly rejecting a cover.
* Fix order-of-operation bug that could cause computation of a
wrong figure of merit (poslen) when shortening a cover.
* Change the preference rule so that candidate headlines that do not
include their whole cover string (after MaxWords trimming) are lowest
priority, since they may not actually satisfy the user's query.
This results in some changes in existing regression test cases,
but they all seem reasonable. Note in particular that the tests
involving strings like "1 2 3" were previously being affected by
the ShortWord filter, masking the normal matching behavior.
Per bug #16345 from Augustinas Jokubauskas; the new test cases are
based on that example. Back-patch to 9.6 where phrase search was
added to tsquery.
Discussion: https://postgr.es/m/16345-2e0cf5cddbdcd3b4@postgresql.org
This code was woefully unreadable and under-commented. Try to improve
matters by adding comments, using some macros to make complicated
if-tests more readable, using boolean type where appropriate, etc.
There are a couple of tiny coding improvements too, but this commit
includes (I hope) no behavioral change.
Nonetheless, back-patch as far as 9.6, because a followup bug-fixing
commit depends on this.
Discussion: https://postgr.es/m/16345-2e0cf5cddbdcd3b4@postgresql.org
Currently, we don't account for buffer usage incurred by parallel workers
for parallel create index. This commit allows each worker to record the
buffer usage stats and leader backend to accumulate that stats at the
end of the operation. This will allow pg_stat_statements to display
correct buffer usage stats for (parallel) create index command.
Reported-by: Julien Rouhaud
Author: Sawada Masahiko
Reviewed-by: Dilip Kumar, Julien Rouhaud and Amit Kapila
Backpatch-through: 11, where this was introduced
Discussion: https://postgr.es/m/20200328151721.GB12854@nol
Repair an oversight in commit 8728b2c70: if we're postponing restore
of event triggers to the end, we must also postpone restoring any
comments on them, since of course we cannot create the comments first.
(This opens yet another opportunity for an event trigger to bollix
the restore, but there's no help for that.)
Per bug #16346 from Alexander Lakhin.
Like the previous commit, back-patch to all supported branches.
Hamid Akhtar and Tom Lane
Discussion: https://postgr.es/m/16346-6210ad7a0ea81be1@postgresql.org
Our documentation describes four allowed input syntaxes for circles,
but the regression tests tried only three ... with predictable
consequences. Remarkably, this has been wrong since the circle
datatype was added in 1997, but nobody noticed till now.
David Zhang, with some help from me
Discussion: https://postgr.es/m/332c47fa-d951-7574-b5cc-a8f7f7201202@highgo.ca
Since the existing bit number argument can't exceed INT32_MAX, it's
not possible for these functions to manipulate bits beyond the first
256MB of a bytea value. However, it'd be good if they could do at
least that much, and not fall over entirely for longer bytea values.
Adjust the comparisons to be done in int64 arithmetic so that works.
Also tweak the error reports to show sane values in case of overflow.
Also add some test cases to improve the miserable code coverage
of these functions.
Apply patch to back branches only; HEAD has a better solution
as of commit 26a944cf2.
Extracted from a much larger patch by Movead Li
Discussion: https://postgr.es/m/20200312115135445367128@highgo.ca
A table rewritten by ALTER TABLE would lose tracking of an index usable
for CLUSTER. This setting is tracked by pg_index.indisclustered and is
controlled by ALTER TABLE, so some extra work was needed to restore it
properly. Note that ALTER TABLE only marks the index that can be used
for clustering, and does not do the actual operation.
Author: Amit Langote, Justin Pryzby
Reviewed-by: Ibrar Ahmed, Michael Paquier
Discussion: https://postgr.es/m/20200202161718.GI13621@telsasoft.com
Backpatch-through: 9.5
There's a very low risk that RecentGlobalXmin could be far enough in
the past to be older than relfrozenxid, or even wrapped
around. Luckily the consequences of that having happened wouldn't be
too bad - the page wouldn't be pruned for a while.
Avoid that risk by using TransactionXmin instead. As that's announced
via MyPgXact->xmin, it is protected against wrapping around (see code
comments for details around relfrozenxid).
Author: Andres Freund
Discussion: https://postgr.es/m/20200328213023.s4eyijhdosuc4vcj@alap3.anarazel.de
Backpatch: 9.5-
entryGetItem()'s three code paths each contained bugs associated
with filtering the entries for gin_fuzzy_search_limit.
The posting-tree path failed to advance "advancePast" after having
decided to filter an item. If we ran out of items on the current
page and needed to advance to the next, what would actually happen
is that entryLoadMoreItems() would re-load the same page. Eventually,
the random dropItem() test would accept one of the same items it'd
previously rejected, and we'd move on --- but it could take awhile
with small gin_fuzzy_search_limit. To add insult to injury, this
case would inevitably cause entryLoadMoreItems() to decide it needed
to re-descend from the root, making things even slower.
The posting-list path failed to implement gin_fuzzy_search_limit
filtering at all, so that all entries in the posting list would
be returned.
The bitmap-result path used a "gotitem" variable that it failed to
update in the one place where it'd actually make a difference, ie
at the one "continue" statement. I think this was unreachable in
practice, because if we'd looped around then it shouldn't be the
case that the entries on the new page are before advancePast.
Still, the "gotitem" variable was contributing nothing to either
clarity or correctness, so get rid of it.
Refactor all three loops so that the termination conditions are
more alike and less unreadable.
The code coverage report showed that we had no coverage at all for
the re-descend-from-root code path in entryLoadMoreItems(), which
seems like a very bad thing, so add a test case that exercises it.
We also had exactly no coverage for gin_fuzzy_search_limit, so add a
simplistic test case that at least hits those code paths a little bit.
Back-patch to all supported branches.
Adé Heyward and Tom Lane
Discussion: https://postgr.es/m/CAEknJCdS-dE1Heddptm7ay2xTbSeADbkaQ8bU2AXRCVC2LdtKQ@mail.gmail.com
contrib/lo's lo_manage() thought it could use
trigdata->tg_trigger->tgname in its error message about
not being called as a trigger. That naturally led to a core dump.
unique_key_recheck() figured it could Assert that fcinfo->context
is a TriggerData node in advance of having checked that it's
being called as a trigger. That's harmless in production builds,
and perhaps not that easy to reach in any case, but it's logically
wrong.
The first of these per bug #16340 from William Crowell;
the second from manual inspection of other CALLED_AS_TRIGGER
call sites.
Back-patch the lo.c change to all supported branches, the
other to v10 where the thinko crept in.
Discussion: https://postgr.es/m/16340-591c7449dc7c8c47@postgresql.org
We require the partition key to be a subset of the set of columns
being made unique, so that physically-separate indexes on the different
partitions are sufficient to enforce the uniqueness constraint.
The existing code checked that the listed columns appear, but did not
inquire into the index semantics, which is a serious oversight given
that different index opclasses might enforce completely different
notions of uniqueness.
Ideally, perhaps, we'd just match the partition key opfamily to the
index opfamily. But hash partitioning uses hash opfamilies which we
can't directly match to btree opfamilies. Hence, look up the equality
operator in each family, and accept if it's the same operator. This
should be okay in a fairly general sense, since the equality operator
ought to precisely represent the opfamily's notion of uniqueness.
A remaining weak spot is that we don't have a cross-index-AM notion of
which opfamily member is "equality". But we know which one to use for
hash and btree AMs, and those are the only two that are relevant here
at present. (Any non-core AMs that know how to enforce equality are
out of luck, for now.)
Back-patch to v11 where this feature was introduced.
Guancheng Luo, revised a bit by me
Discussion: https://postgr.es/m/D9C3CEF7-04E8-47A1-8300-CA1DCD5ED40D@gmail.com
Buildfarm experience shows that this function can fail with ENOENT
if some other process unlinks a file between when we read the directory
entry and when we try to stat() it. The problem is old but we had
not noticed it until 085b6b667 added regression test coverage.
To fix, just ignore ENOENT failures. There is one other case that
this might hide: a symlink that points to nowhere. That seems okay
though, at least better than erroring.
Back-patch to v10 where this function was added, since the regression
test cases were too.
Discussion: https://postgr.es/m/20200308173103.GC1357@telsasoft.com
Experimentation shows that it's not hard at all to drive the
old implementation of "ltree ~ lquery" match to stack overflow,
so throw in a check_stack_depth() call, as I just did in HEAD.
I wasn't able to make it take a long time, because all the
pathological cases I tried hit stack overflow first; but
I bet there are some others that do take a long time, so add
CHECK_FOR_INTERRUPTS() too.
Discussion: https://postgr.es/m/CAP_rww=waX2Oo6q+MbMSiZ9ktdj6eaJj0cQzNu=Ry2cCDij5fw@mail.gmail.com
GetLocaleInfoEx() can fail on strings that setlocale() was perfectly
happy with. A common way for that to happen is if the locale string
is actually a Unix-style string, say "et_EE.UTF-8". In that case,
what's after the dot is an encoding name, not a Windows codepage number;
blindly treating it as a codepage number led to failure, with a fairly
silly error message. Hence, check to see if what's after the dot is
all digits, and if not, treat it as a literal encoding name rather than
a codepage number. This will do the right thing with many Unix-style
locale strings, and produce a more sensible error message otherwise.
Somewhat independently of that, treat a zero (CP_ACP) result from
GetLocaleInfoEx() as meaning that we must use UTF-8 encoding.
Back-patch to all supported branches.
Juan José Santamaría Flecha
Discussion: https://postgr.es/m/24905.1585445371@sss.pgh.pa.us
The documentation says that the max length is 255 bytes, but
code inspection says it's actually 255 characters; and relevant
lengths are stored as uint16 so that that works.
These uint16 fields could be overflowed by excessively long input,
producing strange results. Complain for invalid input.
Likewise check for out-of-range values of the repeat counts in lquery.
(We don't try too hard on that one, notably not bothering to detect
if atoi's result has overflowed.)
Also detect length overflow in ltree_concat.
In passing, be more consistent about whether "syntax error" messages
include the type name. Also, clarify the documentation about what
the size limit is.
This has been broken for a long time, so back-patch to all supported
branches.
Nikita Glukhov, reviewed by Benjie Gillam and Tomas Vondra
Discussion: https://postgr.es/m/CAP_rww=waX2Oo6q+MbMSiZ9ktdj6eaJj0cQzNu=Ry2cCDij5fw@mail.gmail.com
In 9.4 I added support to use a historical snapshot in
ScanPgRelation(), while adding logical decoding. Unfortunately a
conflict with the concurrent removal of SnapshotNow was incorrectly
resolved, leading to an unregistered snapshot being used.
It is not correct to use an unregistered (or non-active) snapshot for
anything non-trivial, because catalog invalidations can cause the
snapshot to be invalidated.
Luckily it seems unlikely to actively cause problems in practice, as
ScanPgRelation() requires that we already have a lock on the relation,
we only look for a single row, and we don't appear to rely on the
result's tid to be correct. It however is clearly wrong and potential
negative consequences would likely be hard to find. So it seems worth
backpatching the fix, even without a concrete hazard.
Discussion: https://postgr.es/m/20200229052459.wzhqnbhrriezg4v2@alap3.anarazel.de
Backpatch: 9.5-
plpgsql_xact_cb ought to treat events XACT_EVENT_PARALLEL_COMMIT and
XACT_EVENT_PARALLEL_ABORT like XACT_EVENT_COMMIT and XACT_EVENT_ABORT
respectively, since its goal is to do process-local cleanup. This
oversight caused plpgsql's end-of-transaction cleanup to not get done
in parallel workers. Since a parallel worker will exit just after the
transaction cleanup, the effects of this are limited. I couldn't find
any case in the core code with user-visible effects, but perhaps there
are some in extensions. In any case it's wrong, so let's fix it before
it bites us not after.
In passing, add some comments around the handling of expression
evaluation resources in DO blocks. There's no live bug there, but it's
quite unobvious what's happening; at least I thought so. This isn't
related to the other issue, except that I found both things while poking
at expression-evaluation performance.
Back-patch the plpgsql_xact_cb fix to 9.5 where those event types
were introduced, and the DO-block commentary to v11 where DO blocks
gained the ability to issue COMMIT/ROLLBACK.
Discussion: https://postgr.es/m/10353.1585247879@sss.pgh.pa.us
When SaveSlotToPath() is called with elevel=LOG, the early exits didn't
release the slot's io_in_progress_lock.
This could result in a walsender being stuck on the lock forever. A
possible way to get into this situation is if the offending code paths
are triggered in a low disk space situation.
Author: Pavan Deolasee <pavan.deolasee@2ndquadrant.com>
Reported-by: Craig Ringer <craig@2ndquadrant.com>
Discussion: https://www.postgresql.org/message-id/flat/56a138c5-de61-f553-7e8f-6789296de785%402ndquadrant.com
During the reporting of constraint violations for partitioned tables,
ExecPartitionCheckEmitError(), ExecConstraints(),
ExecWithCheckOptions() set the slot descriptor of the input slot to
the root partition's tuple desc. That's generally problematic when
the slot could be used by other routines, but can cause crashes after
the introduction of slots with "fixed" tuple descriptors in
ad7dbee368.
The problem likely escaped detection so far for two reasons: First,
currently the only known way that these routines are used with a
partitioned table that is not "owned" by partitioning code is when
"fast defaults" are used for the child partition. Second, as an error
is raised afterwards, an "external" slot that had its descriptor
changed, is very unlikely to continue being used.
Even though the issue currently is only known to cause a crash for
11 (as that has both fast defaults and "fixed" slot descriptors), it
seems worth applying the fix to 10 too. Potentially changing random
slots is hazardous.
Regression tests will be added in a separate commit, as it seems best
to add them for master and 12 too.
Reported-By: Daniel WM
Author: Andres Freund
Bug: #16293
Discussion: https://postgr.es/m/16293-26f5777d10143a66@postgresql.org
Backpatch: 11, 10 only
src/port/getopt_long.c failed on such an argument, always seeing it
as an unrecognized switch. This is unhelpful; better is to treat such
an item as a non-switch argument. That behavior is what we find in
GNU's getopt_long(); it's what src/port/getopt.c does; and it is
required by POSIX for getopt(), which getopt_long() ought to be
generally a superset of. Moreover, it's expected by ecpg, which
intends an argument of "-" to mean "read from stdin". So fix it.
Also add some documentation about ecpg's behavior in this area, since
that was miserably underdocumented. I had to reverse-engineer it
from the code.
Per bug #16304 from James Gray. Back-patch to all supported branches,
since this has been broken forever.
Discussion: https://postgr.es/m/16304-c662b00a1322db7f@postgresql.org
Until now, only selected bulk operations (e.g. COPY) did this. If a
given relfilenode received both a WAL-skipping COPY and a WAL-logged
operation (e.g. INSERT), recovery could lose tuples from the COPY. See
src/backend/access/transam/README section "Skipping WAL for New
RelFileNode" for the new coding rules. Maintainers of table access
methods should examine that section.
To maintain data durability, just before commit, we choose between an
fsync of the relfilenode and copying its contents to WAL. A new GUC,
wal_skip_threshold, guides that choice. If this change slows a workload
that creates small, permanent relfilenodes under wal_level=minimal, try
adjusting wal_skip_threshold. Users setting a timeout on COMMIT may
need to adjust that timeout, and log_min_duration_statement analysis
will reflect time consumption moving to COMMIT from commands like COPY.
Internally, this requires a reliable determination of whether
RollbackAndReleaseCurrentSubTransaction() would unlink a relation's
current relfilenode. Introduce rd_firstRelfilenodeSubid. Amend the
specification of rd_createSubid such that the field is zero when a new
rel has an old rd_node. Make relcache.c retain entries for certain
dropped relations until end of transaction.
Back-patch to 9.5 (all supported versions). This introduces a new WAL
record type, XLOG_GIST_ASSIGN_LSN, without bumping XLOG_PAGE_MAGIC. As
always, update standby systems before master systems. This changes
sizeof(RelationData) and sizeof(IndexStmt), breaking binary
compatibility for affected extensions. (The most recent commit to
affect the same class of extensions was
089e4d405d0f3b94c74a2c6a54357a84a681754b.)
Kyotaro Horiguchi, reviewed (in earlier, similar versions) by Robert
Haas. Heikki Linnakangas and Michael Paquier implemented earlier
designs that materially clarified the problem. Reviewed, in earlier
designs, by Andrew Dunstan, Andres Freund, Alvaro Herrera, Tom Lane,
Fujii Masao, and Simon Riggs. Reported by Martijn van Oosterhout.
Discussion: https://postgr.es/m/20150702220524.GA9392@svana.org
Back-patch a subset of commit 9155580fd5
to v11, v10, 9.6, and 9.5. Include the latest repairs to this function.
Use a new XLOG_FPI_MULTI value instead of reusing XLOG_FPI. That way,
if an older server reads WAL from this function, that server will PANIC
instead of applying just one page of the record. The next commit adds a
call to this function.
Discussion: https://postgr.es/m/20200304.162919.898938381201316571.horikyota.ntt@gmail.com
swap_relation_files() calls toast_get_valid_index() to find and lock
this index, just before swapping with the rebuilt TOAST index. The
latter function releases the lock before returning. Potential for
mischief is low; a concurrent session can issue ALTER INDEX ... SET
(fillfactor = ...), which is not alarming. Nonetheless, changing
pg_class.relfilenode without a lock is unconventional. Back-patch to
9.5 (all supported versions), because another fix needs this.
Discussion: https://postgr.es/m/20191226001521.GA1772687@rfd.leadboat.com
