Tom Lane [Fri, 11 Jul 2014 23:12:48 +0000 (19:12 -0400)]
Fix bug with whole-row references to append subplans.
ExecEvalWholeRowVar incorrectly supposed that it could "bless" the source
TupleTableSlot just once per query. But if the input is coming from an
Append (or, perhaps, other cases?) more than one slot might be returned
over the query run. This led to "record type has not been registered"
errors when a composite datum was extracted from a non-blessed slot.
This bug has been there a long time; I guess it escaped notice because when
dealing with subqueries the planner tends to expand whole-row Vars into
RowExprs, which don't have the same problem. It is possible to trigger
the problem in all active branches, though, as illustrated by the added
regression test.
Tom Lane [Tue, 8 Jul 2014 18:03:26 +0000 (14:03 -0400)]
Don't assume a subquery's output is unique if there's a SRF in its tlist.
While the x output of "select x from t group by x" can be presumed unique,
this does not hold for "select x, generate_series(1,10) from t group by x",
because we may expand the set-returning function after the grouping step.
(Perhaps that should be re-thought; but considering all the other oddities
involved with SRFs in targetlists, it seems unlikely we'll change it.)
Put a check in query_is_distinct_for() so it's not fooled by such cases.
Tom Lane [Wed, 2 Jul 2014 18:20:44 +0000 (14:20 -0400)]
Add some errdetail to checkRuleResultList().
This function wasn't originally thought to be really user-facing,
because converting a table to a view isn't something we expect people
to do manually. So not all that much effort was spent on the error
messages; in particular, while the code will complain that you got
the column types wrong it won't say exactly what they are. But since
we repurposed the code to also check compatibility of rule RETURNING
lists, it's definitely user-facing. It now seems worthwhile to add
errdetail messages showing exactly what the conflict is when there's
a mismatch of column names or types. This is prompted by bug #10836
from Matthias Raffelsieper, which might have been forestalled if the
error message had reported the wrong column type as being "record".
Per Alvaro's advice, back-patch to branches before 9.4, but resist
the temptation to rephrase any existing strings there. Adding new
strings is not really a translation degradation; anyway having the
info presented in English is better than not having it at all.
Tom Lane [Tue, 1 Jul 2014 15:22:56 +0000 (11:22 -0400)]
Fix inadequately-sized output buffer in contrib/unaccent.
The output buffer size in unaccent_lexize() was calculated as input string
length times pg_database_encoding_max_length(), which effectively assumes
that replacement strings aren't more than one character. While that was
all that we previously documented it to support, the code actually has
always allowed replacement strings of arbitrary length; so if you tried
to make use of longer strings, you were at risk of buffer overrun. To fix,
use an expansible StringInfo buffer instead of trying to determine the
maximum space needed a-priori.
This would be a security issue if unaccent rules files could be installed
by unprivileged users; but fortunately they can't, so in the back branches
the problem can be labeled as improper configuration by a superuser.
Nonetheless, a memory stomp isn't a nice way of reacting to improper
configuration, so let's back-patch the fix.
Tom Lane [Thu, 26 Jun 2014 17:41:10 +0000 (10:41 -0700)]
Back-patch "Fix EquivalenceClass processing for nested append relations".
When we committed a87c729153e372f3731689a7be007bc2b53f1410, we somehow
failed to notice that it didn't merely improve plan quality for expression
indexes; there were very closely related cases that failed outright with
"could not find pathkey item to sort". The failing cases seem to be those
where the planner was already capable of selecting a MergeAppend plan,
and there was inheritance involved: the lack of appropriate eclass child
members would prevent prepare_sort_from_pathkeys() from succeeding on the
MergeAppend's child plan nodes for inheritance child tables.
Accordingly, back-patch into 9.1 through 9.3, along with an extra
regression test case covering the problem.
Fujii Masao [Thu, 26 Jun 2014 05:27:27 +0000 (14:27 +0900)]
Remove obsolete example of CSV log file name from log_filename document.
7380b63 changed log_filename so that epoch was not appended to it
when no format specifier is given. But the example of CSV log file name
with epoch still left in log_filename document. This commit removes
such obsolete example.
This commit also documents the defaults of log_directory and
log_filename.
The syntax doesn't let you specify "WITH OIDS" for foreign tables, but it
was still possible with default_with_oids=true. But the rest of the system,
including pg_dump, isn't prepared to handle foreign tables with OIDs
properly.
Backpatch down to 9.1, where foreign tables were introduced. It's possible
that there are databases out there that already have foreign tables with
OIDs. There isn't much we can do about that, but at least we can prevent
them from being created in the future.
Patch by Etsuro Fujita, reviewed by Hadi Moshayedi.
Kevin Grittner [Sat, 21 Jun 2014 14:17:52 +0000 (09:17 -0500)]
Fix documentation template for CREATE TRIGGER.
By using curly braces, the template had specified that one of
"NOT DEFERRABLE", "INITIALLY IMMEDIATE", or "INITIALLY DEFERRED"
was required on any CREATE TRIGGER statement, which is not
accurate. Change to square brackets makes that optional.
Tom Lane [Fri, 20 Jun 2014 02:13:54 +0000 (22:13 -0400)]
Avoid leaking memory while evaluating arguments for a table function.
ExecMakeTableFunctionResult evaluated the arguments for a function-in-FROM
in the query-lifespan memory context. This is insignificant in simple
cases where the function relation is scanned only once; but if the function
is in a sub-SELECT or is on the inside of a nested loop, any memory
consumed during argument evaluation can add up quickly. (The potential for
trouble here had been foreseen long ago, per existing comments; but we'd
not previously seen a complaint from the field about it.) To fix, create
an additional temporary context just for this purpose.
Per an example from MauMau. Back-patch to all active branches.
Noah Misch [Sat, 14 Jun 2014 14:52:25 +0000 (10:52 -0400)]
Make pqsignal() available to pg_regress of ECPG and isolation suites.
Commit 453a5d91d49e4d35054f92785d830df4067e10c1 made it available to the
src/test/regress build of pg_regress, but all pg_regress builds need the
same treatment. Patch 9.2 through 8.4; in 9.3 and later, pg_regress
gets pqsignal() via libpgport.
Noah Misch [Sat, 14 Jun 2014 13:41:13 +0000 (09:41 -0400)]
Secure Unix-domain sockets of "make check" temporary clusters.
Any OS user able to access the socket can connect as the bootstrap
superuser and proceed to execute arbitrary code as the OS user running
the test. Protect against that by placing the socket in a temporary,
mode-0700 subdirectory of /tmp. The pg_regress-based test suites and
the pg_upgrade test suite were vulnerable; the $(prove_check)-based test
suites were already secure. Back-patch to 8.4 (all supported versions).
The hazard remains wherever the temporary cluster accepts TCP
connections, notably on Windows.
As a convenient side effect, this lets testing proceed smoothly in
builds that override DEFAULT_PGSOCKET_DIR. Popular non-default values
like /var/run/postgresql are often unwritable to the build user.
Noah Misch [Sat, 14 Jun 2014 13:41:13 +0000 (09:41 -0400)]
Add mkdtemp() to libpgport.
This function is pervasive on free software operating systems; import
NetBSD's implementation. Back-patch to 8.4, like the commit that will
harness it.
Tom Lane [Fri, 13 Jun 2014 00:14:49 +0000 (20:14 -0400)]
Fix pg_restore's processing of old-style BLOB COMMENTS data.
Prior to 9.0, pg_dump handled comments on large objects by dumping a bunch
of COMMENT commands into a single BLOB COMMENTS archive object. With
sufficiently many such comments, some of the commands would likely get
split across bufferloads when restoring, causing failures in
direct-to-database restores (though no problem would be evident in text
output). This is the same type of issue we have with table data dumped as
INSERT commands, and it can be fixed in the same way, by using a mini SQL
lexer to figure out where the command boundaries are. Fortunately, the
COMMENT commands are no more complex to lex than INSERTs, so we can just
re-use the existing lexer for INSERTs.
Per bug #10611 from Jacek Zalewski. Back-patch to all active branches.
Tom Lane [Thu, 12 Jun 2014 20:51:14 +0000 (16:51 -0400)]
Remove inadvertent copyright violation in largeobject regression test.
Robert Frost is no longer with us, but his copyrights still are, so
let's stop using "Stopping by Woods on a Snowy Evening" as test data
before somebody decides to sue us. Wordsworth is more safely dead.
Tom Lane [Wed, 11 Jun 2014 02:48:16 +0000 (22:48 -0400)]
Fix ancient encoding error in hungarian.stop.
When we grabbed this file off the Snowball project's website, we mistakenly
supposed that it was in LATIN1 encoding, but evidently it was actually in
LATIN2. This resulted in ő (o-double-acute, U+0151, which is code 0xF5 in
LATIN2) being misconverted into õ (o-tilde, U+00F5), as complained of in
bug #10589 from Zoltán Sörös. We'd have messed up u-double-acute too,
but there aren't any of those in the file. Other characters used in the
file have the same codes in LATIN1 and LATIN2, which no doubt helped hide
the problem for so long.
The error is not only ours: the Snowball project also was confused about
which encoding is required for Hungarian. But dealing with that will
require source-code changes that I'm not at all sure we'll wish to
back-patch. Fixing the stopword file seems reasonably safe to back-patch
however.
Fujii Masao [Fri, 6 Jun 2014 09:46:32 +0000 (18:46 +0900)]
Fix breakages of hot standby regression test.
This commit changes HS regression test so that it uses
REPEATABLE READ transaction instead of SERIALIZABLE one
because SERIALIZABLE transaction isolation level is not
available in HS. Also this commit fixes VACUUM/ANALYZE
label mixup.
This was fixed in HEAD (commit 2985e16), but it should
have been back-patched to 9.1 which had introduced SSI
and forbidden SERIALIZABLE transaction in HS.
Tom Lane [Thu, 5 Jun 2014 15:31:18 +0000 (11:31 -0400)]
Add defenses against running with a wrong selection of LOBLKSIZE.
It's critical that the backend's idea of LOBLKSIZE match the way data has
actually been divided up in pg_largeobject. While we don't provide any
direct way to adjust that value, doing so is a one-line source code change
and various people have expressed interest recently in changing it. So,
just as with TOAST_MAX_CHUNK_SIZE, it seems prudent to record the value in
pg_control and cross-check that the backend's compiled-in setting matches
the on-disk data.
Also tweak the code in inv_api.c so that fetches from pg_largeobject
explicitly verify that the length of the data field is not more than
LOBLKSIZE. Formerly we just had Asserts() for that, which is no protection
at all in production builds. In some of the call sites an overlength data
value would translate directly to a security-relevant stack clobber, so it
seems worth one extra runtime comparison to be sure.
In the back branches, we can't change the contents of pg_control; but we
can still make the extra checks in inv_api.c, which will offer some amount
of protection against running with the wrong value of LOBLKSIZE.
Andres Freund [Wed, 4 Jun 2014 21:26:30 +0000 (23:26 +0200)]
Fix longstanding bug in HeapTupleSatisfiesVacuum().
HeapTupleSatisfiesVacuum() didn't properly discern between
DELETE_IN_PROGRESS and INSERT_IN_PROGRESS for rows that have been
inserted in the current transaction and deleted in a aborted
subtransaction of the current backend. At the very least that caused
problems for CLUSTER and CREATE INDEX in transactions that had
aborting subtransactions producing rows, leading to warnings like:
WARNING: concurrent delete in progress within table "..."
possibly in an endless, uninterruptible, loop.
Instead of treating *InProgress xmins the same as *IsCurrent ones,
treat them as being distinct like the other visibility routines. As
implemented this separatation can cause a behaviour change for rows
that have been inserted and deleted in another, still running,
transaction. HTSV will now return INSERT_IN_PROGRESS instead of
DELETE_IN_PROGRESS for those. That's both, more in line with the other
visibility routines and arguably more correct. The latter because a
INSERT_IN_PROGRESS will make callers look at/wait for xmin, instead of
xmax.
The only current caller where that's possibly worse than the old
behaviour is heap_prune_chain() which now won't mark the page as
prunable if a row has concurrently been inserted and deleted. That's
harmless enough.
As a cautionary measure also insert a interrupt check before the gotos
in IndexBuildHeapScan() that lead to the uninterruptible loop. There
are other possible causes, like a row that several sessions try to
update and all fail, for repeated loops and the cost of doing so in
the retry case is low.
As this bug goes back all the way to the introduction of
subtransactions in 573a71a5da backpatch to all supported releases.
Tom Lane [Tue, 3 Jun 2014 16:01:37 +0000 (12:01 -0400)]
Make plpython_unicode regression test work in more database encodings.
This test previously used a data value containing U+0080, and would
therefore fail if the database encoding didn't have an equivalent to
that; which only about half of our supported server encodings do.
We could fall back to using some plain-ASCII character, but that seems
like it's losing most of the point of the test. Instead switch to using
U+00A0 (no-break space), which translates into all our supported encodings
except the four in the EUC_xx family.
Per buildfarm testing. Back-patch to 9.1, which is as far back as this
test is expected to succeed everywhere. (9.0 has the test, but without
back-patching some 9.1 code changes we could not expect to get consistent
results across platforms anyway.)
Andres Freund [Tue, 3 Jun 2014 12:02:54 +0000 (14:02 +0200)]
Set the process latch when processing recovery conflict interrupts.
Because RecoveryConflictInterrupt() didn't set the process latch
anything using the latter to wait for events didn't get notified about
recovery conflicts. Most latch users are never the target of recovery
conflicts, which explains the lack of reports about this until
now.
Since 9.3 two possible affected users exist though: The sql callable
pg_sleep() now uses latches to wait and background workers are
expected to use latches in their main loop. Both would currently wait
until the end of WaitLatch's timeout.
Fix by adding a SetLatch() to RecoveryConflictInterrupt(). It'd also
be possible to fix the issue by having each latch user set
set_latch_on_sigusr1. That seems failure prone and though, as most of
these callsites won't often receive recovery conflicts and thus will
likely only be tested against normal query cancels et al. It'd also be
unnecessarily verbose.
Backpatch to 9.1 where latches were introduced. Arguably 9.3 would be
sufficient, because that's where pg_sleep() was converted to waiting
on the latch and background workers got introduced; but there could be
user level code making use of the latch pre 9.3.
Tom Lane [Fri, 30 May 2014 22:18:24 +0000 (18:18 -0400)]
On OS X, link libpython normally, ignoring the "framework" framework.
As of Xcode 5.0, Apple isn't including the Python framework as part of the
SDK-level files, which means that linking to it might fail depending on
whether Xcode thinks you've selected a specific SDK version. According to
their Tech Note 2328, they've basically deprecated the framework method of
linking to libpython and are telling people to link to the shared library
normally. (I'm pretty sure this is in direct contradiction to the advice
they were giving a few years ago, but whatever.) Testing says that this
approach works fine at least as far back as OS X 10.4.11, so let's just
rip out the framework special case entirely. We do still need a special
case to decide that OS X provides a shared library at all, unfortunately
(I wonder why the distutils check doesn't work ...). But this is still
less of a special case than before, so it's fine.
Back-patch to all supported branches, since we'll doubtless be hearing
about this more as more people update to recent Xcode.
Tom Lane [Thu, 29 May 2014 17:51:12 +0000 (13:51 -0400)]
When using the OSSP UUID library, cache its uuid_t state object.
The original coding in contrib/uuid-ossp created and destroyed a uuid_t
object (or, in some cases, even two of them) each time it was called.
This is not the intended usage: you're supposed to keep the uuid_t object
around so that the library can cache its state across uses. (Other UUID
libraries seem to keep equivalent state behind-the-scenes in static
variables, but OSSP chose differently.) Aside from being quite inefficient,
creating a new uuid_t loses knowledge of the previously generated UUID,
which in theory could result in duplicate V1-style UUIDs being created
on sufficiently fast machines.
On at least some platforms, creating a new uuid_t also draws some entropy
from /dev/urandom, leaving less for the rest of the system. This seems
sufficiently unpleasant to justify back-patching this change.
It turns out that the %name-prefix syntax without "=" does not work
at all in pre-2.4 Bison. We are not prepared to make such a large
jump in minimum required Bison version just to suppress a warning
message in a version hardly any developers are using yet.
When 3.0 gets more popular, we'll figure out a way to deal with this.
In the meantime, BISONFLAGS=-Wno-deprecated is recommendable for
anyone using 3.0 who doesn't want to see the warning.
Tom Lane [Wed, 28 May 2014 19:42:01 +0000 (15:42 -0400)]
Fix bogus %name-prefix option syntax in all our Bison files.
%name-prefix doesn't use an "=" sign according to the Bison docs, but it
silently accepted one anyway, until Bison 3.0. This was originally a
typo of mine in commit 012abebab1bc72043f3f670bf32e91ae4ee04bd2, and we
seem to have slavishly copied the error into all the other grammar files.
Per report from Vik Fearing; analysis by Peter Eisentraut.
Back-patch to all active branches, since somebody might try to build
a back branch with up-to-date tools.
Magnus Hagander [Wed, 28 May 2014 11:03:21 +0000 (13:03 +0200)]
Ensure cleanup in case of early errors in streaming base backups
Move the code that sends the initial status information as well as the
calculation of paths inside the ENSURE_ERROR_CLEANUP block. If this code
failed, we would "leak" a counter of number of concurrent backups, thereby
making the system always believe it was in backup mode. This could happen
if the sending failed (which it probably never did given that the small
amount of data to send would never cause a flush). It is very low risk, but
all operations after do_pg_start_backup should be protected.
Tom Lane [Tue, 27 May 2014 02:23:39 +0000 (22:23 -0400)]
Avoid unportable usage of sscanf(UINT64_FORMAT).
On Mingw, it seems that scanf() doesn't necessarily accept the same format
codes that printf() does, and in particular it may fail to recognize %llu
even though printf() does. Since configure only probes printf() behavior
while setting up the INT64_FORMAT macros, this means it's unsafe to use
those macros with scanf(). We had only one instance of such a coding
pattern, in contrib/pg_stat_statements, so change that code to avoid
the problem.
Per buildfarm warnings. Back-patch to 9.0 where the troublesome code
was introduced.
Use 0-based numbering in comments about backup blocks.
The macros and functions that work with backup blocks in the redo function
use 0-based numbering, so let's use that consistently in the function that
generates the records too. Makes it so much easier to compare the
generation and replay functions.
Backpatch to 9.0, where we switched from 1-based to 0-based numbering.
Initialize tsId and dbId fields in WAL record of COMMIT PREPARED.
Commit dd428c79 added dbId and tsId to the xl_xact_commit struct but missed
that prepared transaction commits reuse that struct. Fix that.
Because those fields were left unitialized, replaying a commit prepared WAL
record in a hot standby node would fail to remove the relcache init file.
That can lead to "could not open file" errors on the standby. Relcache init
file only needs to be removed when a system table/index is rewritten in the
transaction using two phase commit, so that should be rare in practice. In
HEAD, the incorrect dbId/tsId values are also used for filtering in logical
replication code, causing the transaction to always be filtered out.
Analysis and fix by Andres Freund. Backpatch to 9.0 where hot standby was
introduced.
Tom Lane [Thu, 15 May 2014 19:58:05 +0000 (15:58 -0400)]
Fix unportable setvbuf() usage in initdb.
In yesterday's commit 2dc4f011fd61501cce507be78c39a2677690d44b, I tried
to force buffering of stdout/stderr in initdb to be what it is by
default when the program is run interactively on Unix (since that's how
most manual testing is done). This tripped over the fact that Windows
doesn't support _IOLBF mode. We dealt with that a long time ago in
syslogger.c by falling back to unbuffered mode on Windows. Export that
solution in port.h and use it in initdb.
The proc array can contain duplicate XIDs, when a transaction is just being
prepared for two-phase commit. To cope, remove any duplicates in
txid_current_snapshot(). Also ignore duplicates in the input functions, so
that if e.g. you have an old pg_dump file that already contains duplicates,
it will be accepted.
Report and fix by Jan Wieck. Backpatch to all supported versions.
Fix race condition in preparing a transaction for two-phase commit.
To lock a prepared transaction's shared memory entry, we used to mark it
with the XID of the backend. When the XID was no longer active according
to the proc array, the entry was implicitly considered as not locked
anymore. However, when preparing a transaction, the backend's proc array
entry was cleared before transfering the locks (and some other state) to
the prepared transaction's dummy PGPROC entry, so there was a window where
another backend could finish the transaction before it was in fact fully
prepared.
To fix, rewrite the locking mechanism of global transaction entries. Instead
of an XID, just have simple locked-or-not flag in each entry (we store the
locking backend's backend id rather than a simple boolean, but that's just
for debugging purposes). The backend is responsible for explicitly unlocking
the entry, and to make sure that that happens, install a callback to unlock
it on abort or process exit.
Tom Lane [Thu, 15 May 2014 01:14:02 +0000 (21:14 -0400)]
In initdb, ensure stdout/stderr buffering behavior is what we expect.
Since this program may print to either stdout or stderr, the relative
ordering of its messages depends on the buffering behavior of those files.
Force stdout to be line-buffered and stderr to be unbuffered, ensuring
that the behavior will match standard Unix interactive behavior, even
when stdout and stderr are rerouted to a file.
Per complaint from Tomas Vondra. The particular case he pointed out is
new in HEAD, but issues of the same sort could arise in any branch with
other error messages, so back-patch to all branches.
I'm unsure whether we might not want to do this in other client programs
as well. For the moment, just fix initdb.
Initialize padding bytes in btree_gist varbit support.
The code expands a varbit gist leaf key to a node key by copying the bit
data twice in a varlen datum, as both the lower and upper key. The lower key
was expanded to INTALIGN size, but the padding bytes were not initialized.
That's a problem because when the lower/upper keys are compared, the padding
bytes are used compared too, when the values are otherwise equal. That could
lead to incorrect query results.
REINDEX is advised for any btree_gist indexes on bit or bit varying data
type, to fix any garbage padding bytes on disk.
Per Valgrind, reported by Andres Freund. Backpatch to all supported
versions.
Tom Lane [Fri, 9 May 2014 01:45:02 +0000 (21:45 -0400)]
Document permissions needed for pg_database_size and pg_tablespace_size.
Back in 8.3, we installed permissions checks in these functions (see
commits 8bc225e7990a and cc26599b7206). But we forgot to document that
anywhere in the user-facing docs; it did get mentioned in the 8.3 release
notes, but nobody's looking at that any more. Per gripe from Suya Huang.
Noah Misch [Thu, 8 May 2014 23:29:02 +0000 (19:29 -0400)]
Un-break ecpg test suite under --disable-integer-datetimes.
Commit 4318daecc959886d001a6e79c6ea853e8b1dfb4b broke it. The change in
sub-second precision at extreme dates is normal. The inconsistent
truncation vs. rounding is essentially a bug, albeit a longstanding one.
Back-patch to 8.4, like the causative commit.
Protect against torn pages when deleting GIN list pages.
To-be-deleted list pages contain no useful information, as they are being
deleted, but we must still protect the writes from being torn by a crash
after a partial write. To do that, re-initialize the pages on WAL replay.
Jeff Janes caught this with a test program to test partial writes.
Backpatch to all supported versions.
Tom Lane [Thu, 8 May 2014 01:38:44 +0000 (21:38 -0400)]
Avoid buffer bloat in libpq when server is consistently faster than client.
If the server sends a long stream of data, and the server + network are
consistently fast enough to force the recv() loop in pqReadData() to
iterate until libpq's input buffer is full, then upon processing the last
incomplete message in each bufferload we'd usually double the buffer size,
due to supposing that we didn't have enough room in the buffer to finish
collecting that message. After filling the newly-enlarged buffer, the
cycle repeats, eventually resulting in an out-of-memory situation (which
would be reported misleadingly as "lost synchronization with server").
Of course, we should not enlarge the buffer unless we still need room
after discarding already-processed messages.
This bug dates back quite a long time: pqParseInput3 has had the behavior
since perhaps 2003, getCopyDataMessage at least since commit 70066eb1a1ad
in 2008. Probably the reason it's not been isolated before is that in
common environments the recv() loop would always be faster than the server
(if on the same machine) or faster than the network (if not); or at least
it wouldn't be slower consistently enough to let the buffer ramp up to a
problematic size. The reported cases involve Windows, which perhaps has
different timing behavior than other platforms.
Per bug #7914 from Shin-ichi Morita, though this is different from his
proposed solution. Back-patch to all supported branches.
Tom Lane [Wed, 7 May 2014 18:25:22 +0000 (14:25 -0400)]
Fix failure to set ActiveSnapshot while rewinding a cursor.
ActiveSnapshot needs to be set when we call ExecutorRewind because some
plan node types may execute user-defined functions during their ReScan
calls (nodeLimit.c does so, at least). The wisdom of that is somewhat
debatable, perhaps, but for now the simplest fix is to make sure the
required context is valid. Failure to do this typically led to a
null-pointer-dereference core dump, though it's possible that in more
complex cases a function could be executed with the wrong snapshot
leading to very subtle misbehavior.
Per report from Leif Jensen. It's been broken for a long time, so
back-patch to all active branches.
It was designed to test the longest possible interval output length,
so removing four zeros from the number of hours, as this patch does,
is not ideal. But the test still has some utility for its original
purpose, and there aren't a lot of other good options.
Noah Misch suggested a different approach where we test that the
output either matches what we expect from integer timestamps or what
we expect from floating-point timestamps. That seemed to obscure an
otherwise simple test, however.
Michael Meskes [Tue, 6 May 2014 11:04:30 +0000 (13:04 +0200)]
Fix handling of array of char pointers in ecpglib.
When array of char * was used as target for a FETCH statement returning more
than one row, it tried to store all the result in the first element. Instead it
should dump array of char pointers with right offset, use the address instead
of the value of the C variable while reading the array and treat such variable
as char **, instead of char * for pointer arithmetic.
Patch by Ashutosh Bapat <ashutosh.bapat@enterprisedb.com>
Tom Lane [Mon, 5 May 2014 18:43:49 +0000 (14:43 -0400)]
Fix possible cache invalidation failure in ReceiveSharedInvalidMessages.
Commit fad153ec45299bd4d4f29dec8d9e04e2f1c08148 modified sinval.c to reduce
the number of calls into sinvaladt.c (which require taking a shared lock)
by keeping a local buffer of collected-but-not-yet-processed messages.
However, if processing of the last message in a batch resulted in a
recursive call to ReceiveSharedInvalidMessages, we could overwrite that
message with a new one while the outer invalidation function was still
working on it. This would be likely to lead to invalidation of the wrong
cache entry, allowing subsequent processing to use stale cache data.
The fix is just to make a local copy of each message while we're processing
it.
Spotted by Andres Freund. Back-patch to 8.4 where the bug was introduced.
Tom Lane [Fri, 2 May 2014 19:30:38 +0000 (15:30 -0400)]
Fix "quiet inline" configure test for newer clang compilers.
This test used to just define an unused static inline function and check
whether that causes a warning. But newer clang versions warn about
unused static inline functions when defined inside a .c file, but not
when defined in an included header, which is the case we care about.
Change the test to cope.
Tom Lane [Thu, 1 May 2014 19:19:17 +0000 (15:19 -0400)]
Fix failure to detoast fields in composite elements of structured types.
If we have an array of records stored on disk, the individual record fields
cannot contain out-of-line TOAST pointers: the tuptoaster.c mechanisms are
only prepared to deal with TOAST pointers appearing in top-level fields of
a stored row. The same applies for ranges over composite types, nested
composites, etc. However, the existing code only took care of expanding
sub-field TOAST pointers for the case of nested composites, not for other
structured types containing composites. For example, given a command such
as
UPDATE tab SET arraycol = ARRAY[(ROW(x,42)::mycompositetype] ...
where x is a direct reference to a field of an on-disk tuple, if that field
is long enough to be toasted out-of-line then the TOAST pointer would be
inserted as-is into the array column. If the source record for x is later
deleted, the array field value would become a dangling pointer, leading
to errors along the line of "missing chunk number 0 for toast value ..."
when the value is referenced. A reproducible test case for this was
provided by Jan Pecek, but it seems likely that some of the "missing chunk
number" reports we've heard in the past were caused by similar issues.
Code-wise, the problem is that PG_DETOAST_DATUM() is not adequate to
produce a self-contained Datum value if the Datum is of composite type.
Seen in this light, the problem is not just confined to arrays and ranges,
but could also affect some other places where detoasting is done in that
way, for example form_index_tuple().
I tried teaching the array code to apply toast_flatten_tuple_attribute()
along with PG_DETOAST_DATUM() when the array element type is composite,
but this was messy and imposed extra cache lookup costs whether or not any
TOAST pointers were present, indeed sometimes when the array element type
isn't even composite (since sometimes it takes a typcache lookup to find
that out). The idea of extending that approach to all the places that
currently use PG_DETOAST_DATUM() wasn't attractive at all.
This patch instead solves the problem by decreeing that composite Datum
values must not contain any out-of-line TOAST pointers in the first place;
that is, we expand out-of-line fields at the point of constructing a
composite Datum, not at the point where we're about to insert it into a
larger tuple. This rule is applied only to true composite Datums, not
to tuples that are being passed around the system as tuples, so it's not
as invasive as it might sound at first. With this approach, the amount
of code that has to be touched for a full solution is greatly reduced,
and added cache lookup costs are avoided except when there actually is
a TOAST pointer that needs to be inlined.
The main drawback of this approach is that we might sometimes dereference
a TOAST pointer that will never actually be used by the query, imposing a
rather large cost that wasn't there before. On the other side of the coin,
if the field value is used multiple times then we'll come out ahead by
avoiding repeat detoastings. Experimentation suggests that common SQL
coding patterns are unaffected either way, though. Applications that are
very negatively affected could be advised to modify their code to not fetch
columns they won't be using.
In future, we might consider reverting this solution in favor of detoasting
only at the point where data is about to be stored to disk, using some
method that can drill down into multiple levels of nested structured types.
That will require defining new APIs for structured types, though, so it
doesn't seem feasible as a back-patchable fix.
Note that this patch changes HeapTupleGetDatum() from a macro to a function
call; this means that any third-party code using that macro will not get
protection against creating TOAST-pointer-containing Datums until it's
recompiled. The same applies to any uses of PG_RETURN_HEAPTUPLEHEADER().
It seems likely that this is not a big problem in practice: most of the
tuple-returning functions in core and contrib produce outputs that could
not possibly be toasted anyway, and the same probably holds for third-party
extensions.
This bug has existed since TOAST was invented, so back-patch to all
supported branches.
Tom Lane [Wed, 30 Apr 2014 17:46:22 +0000 (13:46 -0400)]
Check for interrupts and stack overflow during rule/view dumps.
Since ruleutils.c recurses, it could be driven to stack overflow by
deeply nested constructs. Very large queries might also take long
enough to deparse that a check for interrupts seems like a good idea.
Stick appropriate tests into a couple of key places.
Noted by Greg Stark. Back-patch to all supported branches.
Some popen() calls were missing SYSTEMQUOTEs, which caused initdb and
pg_upgrade to fail on Windows, if the installation path contained both
spaces and @ signs.
Patch by Nikhil Deshpande. Backpatch to all supported versions.
Fix two bugs in WAL-logging of GIN pending-list pages.
In writeListPage, never take a full-page image of the page, because we
have all the information required to re-initialize in the WAL record
anyway. Before this fix, a full-page image was always generated, unless
full_page_writes=off, because when the page is initialized its LSN is
always 0. In stable-branches, keep the code to restore the backup blocks
if they exist, in case that the WAL is generated with an older minor
version, but in master Assert that there are no full-page images.
In the redo routine, add missing "off++". Otherwise the tuples are added
to the page in reverse order. That happens to be harmless because we
always scan and remove all the tuples together, but it was clearly wrong.
Also, it was masked by the first bug unless full_page_writes=off, because
the page was always restored from a full-page image.
Tom Lane [Thu, 24 Apr 2014 17:29:48 +0000 (13:29 -0400)]
Reset pg_stat_activity.xact_start during PREPARE TRANSACTION.
Once we've completed a PREPARE, our session is not running a transaction,
so its entry in pg_stat_activity should show xact_start as null, rather
than leaving the value as the start time of the now-prepared transaction.
I think possibly this oversight was triggered by faulty extrapolation
from the adjacent comment that says PrepareTransaction should not call
AtEOXact_PgStat, so tweak the wording of that comment.
Noted by Andres Freund while considering bug #10123 from Maxim Boguk,
although this error doesn't seem to explain that report.
Tom Lane [Thu, 24 Apr 2014 01:21:15 +0000 (21:21 -0400)]
Fix incorrect pg_proc.proallargtypes entries for two built-in functions.
pg_sequence_parameters() and pg_identify_object() have had incorrect
proallargtypes entries since 9.1 and 9.3 respectively. This was mostly
masked by the correct information in proargtypes, but a few operations
such as pg_get_function_arguments() (and thus psql's \df display) would
show the wrong data types for these functions' input parameters.
In HEAD, fix the wrong info, bump catversion, and add an opr_sanity
regression test to catch future mistakes of this sort.
In the back branches, just fix the wrong info so that installations
initdb'd with future minor releases will have the right data. We
can't force an initdb, and it doesn't seem like a good idea to add
a regression test that will fail on existing installations.
Andrew Dunstan [Wed, 16 Apr 2014 17:35:46 +0000 (13:35 -0400)]
Attempt to get plpython regression tests working again for MSVC builds.
This has probably been broken for quite a long time. Buildfarm member
currawong's current results suggest that it's been broken since 9.1, so
backpatch this to that branch.
This only supports Python 2 - I will handle Python 3 separately, but
this is a fairly simple fix.
Tom Lane [Wed, 16 Apr 2014 17:21:03 +0000 (13:21 -0400)]
Use AF_UNSPEC not PF_UNSPEC in getaddrinfo calls.
According to the Single Unix Spec and assorted man pages, you're supposed
to use the constants named AF_xxx when setting ai_family for a getaddrinfo
call. In a few places we were using PF_xxx instead. Use of PF_xxx
appears to be an ancient BSD convention that was not adopted by later
standardization. On BSD and most later Unixen, it doesn't matter much
because those constants have equivalent values anyway; but nonetheless
this code is not per spec.
In the same vein, replace PF_INET by AF_INET in one socket() call, which
wasn't even consistent with the other socket() call in the same function
let alone the remainder of our code.
Per investigation of a Cygwin trouble report from Marco Atzeri. It's
probably a long shot that this will fix his issue, but it's wrong in
any case.
Magnus Hagander [Wed, 16 Apr 2014 15:18:02 +0000 (17:18 +0200)]
Fix timeout in LDAP lookup of libpq connection parameters
Bind attempts to an LDAP server should time out after two seconds,
allowing additional lines in the service control file to be parsed
(which provide a fall back to a secondary LDAP server or default options).
The existing code failed to enforce that timeout during TCP connect,
resulting in a hang far longer than two seconds if the LDAP server
does not respond.
Bruce Momjian [Wed, 16 Apr 2014 14:45:48 +0000 (10:45 -0400)]
check socket creation errors against PGINVALID_SOCKET
Previously, in some places, socket creation errors were checked for
negative values, which is not true for Windows because sockets are
unsigned. This masked socket creation errors on Windows.
Backpatch through 9.0. 8.4 doesn't have the infrastructure to fix this.
Use correctly-sized buffer when zero-filling a WAL file.
I mixed up BLCKSZ and XLOG_BLCKSZ when I changed the way the buffer is
allocated a couple of weeks ago. With the default settings, they are both
8k, but they can be changed at compile-time.
Don't reset the rightlink of a page when replaying a page update record.
This was a leftover from pre-hot standby days, when it was not possible to
have scans concurrent with WAL replay. Resetting the right-link was not
necessary back then either, but it was done for the sake of tidiness. But
with hot standby, it's wrong, because a concurrent scan might still need it.
Backpatch all versions with hot standby, 9.0 and above.
Tom Lane [Sat, 5 Apr 2014 22:16:17 +0000 (18:16 -0400)]
Block signals earlier during postmaster startup.
Formerly, we set up the postmaster's signal handling only when we were
about to start launching subprocesses. This is a bad idea though, as
it means that for example a SIGINT arriving before that will kill the
postmaster instantly, perhaps leaving lockfiles, socket files, shared
memory, etc laying about. We'd rather that such a signal caused orderly
postmaster termination including releasing of those resources. A simple
fix is to move the PostmasterMain stanza that initializes signal handling
to an earlier point, before we've created any such resources. Then, an
early-arriving signal will be blocked until we're ready to deal with it
in the usual way. (The only part that really needs to be moved up is
blocking of signals, but it seems best to keep the signal handler
installation calls together with that; for one thing this ensures the
kernel won't drop any signals we wished to get. The handlers won't get
invoked in any case until we unblock signals in ServerLoop.)
Per a report from MauMau. He proposed changing the way "pg_ctl stop"
works to deal with this, but that'd just be masking one symptom not
fixing the core issue.
It's been like this since forever, so back-patch to all supported branches.
Tom Lane [Sat, 5 Apr 2014 16:41:34 +0000 (12:41 -0400)]
Fix processing of PGC_BACKEND GUC parameters on Windows.
EXEC_BACKEND builds (i.e., Windows) failed to absorb values of PGC_BACKEND
parameters if they'd been changed post-startup via the config file. This
for example prevented log_connections from working if it were turned on
post-startup. The mechanism for handling this case has always been a bit
of a kluge, and it wasn't revisited when we implemented EXEC_BACKEND.
While in a normal forking environment new backends will inherit the
postmaster's value of such settings, EXEC_BACKEND backends have to read
the settings from the CONFIG_EXEC_PARAMS file, and they were mistakenly
rejecting them. So this case has always been broken in the Windows port;
so back-patch to all supported branches.
Tom Lane [Sat, 5 Apr 2014 03:09:45 +0000 (23:09 -0400)]
Fix tablespace creation WAL replay to work on Windows.
The code segment that removes the old symlink (if present) wasn't clued
into the fact that on Windows, symlinks are junction points which have
to be removed with rmdir().
Backpatch to 9.0, where the failing code was introduced.
MauMau, reviewed by Muhammad Asif Naeem and Amit Kapila
Tom Lane [Thu, 3 Apr 2014 18:18:41 +0000 (14:18 -0400)]
Fix documentation about joining pg_locks to other views.
The advice to join to pg_prepared_xacts via the transaction column was not
updated when the transaction column was replaced by virtualtransaction.
Since it's not quite obvious how to do that join, give an explicit example.
For consistency also give an example for the adjacent case of joining to
pg_stat_activity. And link-ify the view references too, just because we
can. Per bug #9840 from Alexey Bashtanov.
Tom Lane [Thu, 3 Apr 2014 15:05:55 +0000 (11:05 -0400)]
Fix documentation about size of interval type.
It's been 16 bytes, not 12, for ages. This was fixed in passing in HEAD
(commit 146604ec), but as a factual error it should have been back-patched.
Per gripe from Tatsuhito Kasahara.
Avoid palloc in critical section in GiST WAL-logging.
Memory allocation can fail if you run out of memory, and inside a critical
section that will lead to a PANIC. Use conservatively-sized arrays in stack
instead.
There was previously no explicit limit on the number of pages a GiST split
can produce, it was only limited by the number of LWLocks that can be held
simultaneously (100 at the moment). This patch adds an explicit limit of 75
pages. That should be plenty, a typical split shouldn't produce more than
2-3 page halves.
The bug has been there forever, but only backpatch down to 9.1. The code
was changed significantly in 9.1, and it doesn't seem worth the risk or
trouble to adapt this for 9.0 and 8.4.
Tom Lane [Wed, 2 Apr 2014 21:11:34 +0000 (17:11 -0400)]
Fix assorted issues in client host name lookup.
The code for matching clients to pg_hba.conf lines that specify host names
(instead of IP address ranges) failed to complain if reverse DNS lookup
failed; instead it silently didn't match, so that you might end up getting
a surprising "no pg_hba.conf entry for ..." error, as seen in bug #9518
from Mike Blackwell. Since we don't want to make this a fatal error in
situations where pg_hba.conf contains a mixture of host names and IP
addresses (clients matching one of the numeric entries should not have to
have rDNS data), remember the lookup failure and mention it as DETAIL if
we get to "no pg_hba.conf entry". Apply the same approach to forward-DNS
lookup failures, too, rather than treating them as immediate hard errors.
Along the way, fix a couple of bugs that prevented us from detecting an
rDNS lookup error reliably, and make sure that we make only one rDNS lookup
attempt; formerly, if the lookup attempt failed, the code would try again
for each host name entry in pg_hba.conf. Since more or less the whole
point of this design is to ensure there's only one lookup attempt not one
per entry, the latter point represents a performance bug that seems
sufficient justification for back-patching.
Also, adjust src/port/getaddrinfo.c so that it plays as well as it can
with this code. Which is not all that well, since it does not have actual
support for rDNS lookup, but at least it should return the expected (and
required by spec) error codes so that the main code correctly perceives the
lack of functionality as a lookup failure. It's unlikely that PG is still
being used in production on any machines that require our getaddrinfo.c,
so I'm not excited about working harder than this.
Tom Lane [Wed, 2 Apr 2014 01:30:18 +0000 (21:30 -0400)]
Fix bugs in manipulation of PgBackendStatus.st_clienthostname.
Initialization of this field was not being done according to the
st_changecount protocol (it has to be done within the changecount increment
range, not outside). And the test to see if the value should be reported
as null was wrong. Noted while perusing uses of Port.remote_hostname.
This was wrong from the introduction of this code (commit 4a25bc145),
so back-patch to 9.1.
Noah Misch [Sat, 29 Mar 2014 04:52:56 +0000 (00:52 -0400)]
Secure Unix-domain sockets of "make check" temporary clusters.
Any OS user able to access the socket can connect as the bootstrap
superuser and in turn execute arbitrary code as the OS user running the
test. Protect against that by placing the socket in the temporary data
directory, which has mode 0700 thanks to initdb. Back-patch to 8.4 (all
supported versions). The hazard remains wherever the temporary cluster
accepts TCP connections, notably on Windows.
Attempts to run "make check" from a directory with a long name will now
fail. An alternative not sharing that problem was to place the socket
in a subdirectory of /tmp, but that is only secure if /tmp is sticky.
The PG_REGRESS_SOCK_DIR environment variable is available as a
workaround when testing from long directory paths.
As a convenient side effect, this lets testing proceed smoothly in
builds that override DEFAULT_PGSOCKET_DIR. Popular non-default values
like /var/run/postgresql are often unwritable to the build user.
Tom Lane [Wed, 26 Mar 2014 20:41:41 +0000 (16:41 -0400)]
Fix refcounting bug in PLy_modify_tuple().
We must increment the refcount on "plntup" as soon as we have the
reference, not sometime later. Otherwise, if an error is thrown in
between, the Py_XDECREF(plntup) call in the PG_CATCH block removes a
refcount we didn't add, allowing the object to be freed even though
it's still part of the plpython function's parsetree.
This appears to be the cause of crashes seen on buildfarm member
prairiedog. It's a bit surprising that we've not seen it fail repeatably
before, considering that the regression tests have been exercising the
faulty code path since 2009.
The real-world impact is probably minimal, since it's unlikely anyone would
be provoking the "TD["new"] is not a dictionary" error in production, and
that's the only case that is actually wrong. Still, it's a bug affecting
the regression tests, so patch all supported branches.
In passing, remove dead variable "plstr", and demote "platt" to a local
variable inside the PG_TRY block, since we don't need to clean it up
in the PG_CATCH path.
Fujii Masao [Mon, 17 Mar 2014 11:42:35 +0000 (20:42 +0900)]
Fix bug in clean shutdown of walsender that pg_receiving is connecting to.
On clean shutdown, walsender waits for all WAL to be replicated to a standby,
and exits. It determined whether that replication had been completed by
checking whether its sent location had been equal to a standby's flush
location. Unfortunately this condition never becomes true when the standby
such as pg_receivexlog which always returns an invalid flush location is
connecting to walsender, and then walsender waits forever.
This commit changes walsender so that it just checks a standby's write
location if a flush location is invalid.
Back-patch to 9.1 where enough infrastructure for this exists.
Tom Lane [Mon, 17 Mar 2014 01:43:49 +0000 (21:43 -0400)]
Fix advertised dispsize for libpq's sslmode connection parameter.
"8" was correct back when "disable" was the longest allowed value, but
since "verify-full" was added, it should be "12". Given the lack of
complaints, I wouldn't be surprised if nobody is actually using these
values ... but still, if they're in the API, they should be right.
Noticed while pursuing a different problem. It's been wrong for quite
a long time, so back-patch to all supported branches.
Tom Lane [Fri, 14 Mar 2014 00:59:51 +0000 (20:59 -0400)]
Prevent interrupts while reporting non-ERROR elog messages.
This should eliminate the risk of recursive entry to syslog(3), which
appears to be the cause of the hang reported in bug #9551 from James
Morton.
Arguably, the real problem here is auth.c's willingness to turn on
ImmediateInterruptOK while executing fairly wide swaths of backend code.
We may well need to work at narrowing the code ranges in which the
authentication_timeout interrupt is enabled. For the moment, though,
this is a cheap and reasonably noninvasive fix for a field-reported
failure; the other approach would be complex and not necessarily
bug-free itself.
Tom Lane [Thu, 13 Mar 2014 16:03:03 +0000 (12:03 -0400)]
Avoid transaction-commit race condition while receiving a NOTIFY message.
Use TransactionIdIsInProgress, then TransactionIdDidCommit, to distinguish
whether a NOTIFY message's originating transaction is in progress,
committed, or aborted. The previous coding could accept a message from a
transaction that was still in-progress according to the PGPROC array;
if the client were fast enough at starting a new transaction, it might fail
to see table rows added/updated by the message-sending transaction. Which
of course would usually be the point of receiving the message. We noted
this type of race condition long ago in tqual.c, but async.c overlooked it.
The race condition probably cannot occur unless there are multiple NOTIFY
senders in action, since an individual backend doesn't send NOTIFY signals
until well after it's done committing. But if two senders commit in close
succession, it's certainly possible that we could see the second sender's
message within the race condition window while responding to the signal
from the first one.
Per bug #9557 from Marko Tiikkaja. This patch is slightly more invasive
than what he proposed, since it removes the now-redundant
TransactionIdDidAbort call.
Back-patch to 9.0, where the current NOTIFY implementation was introduced.
In WAL replay, restore GIN metapage unconditionally to avoid torn page.
We don't take a full-page image of the GIN metapage; instead, the WAL record
contains all the information required to reconstruct it from scratch. But
to avoid torn page hazards, we must re-initialize it from the WAL record
every time, even if it already has a greater LSN, similar to how normal full
page images are restored.
This was highly unlikely to cause any problems in practice, because the GIN
metapage is small. We rely on an update smaller than a 512 byte disk sector
to be atomic elsewhere, at least in pg_control. But better safe than sorry,
and this would be easy to overlook if more fields are added to the metapage
so that it's no longer small.
Reported by Noah Misch. Backpatch to all supported versions.
Fix dangling smgr_owner pointer when a fake relcache entry is freed.
A fake relcache entry can "own" a SmgrRelation object, like a regular
relcache entry. But when it was free'd, the owner field in SmgrRelation
was not cleared, so it was left pointing to free'd memory.
Amazingly this apparently hasn't caused crashes in practice, or we would've
heard about it earlier. Andres found this with Valgrind.
Report and fix by Andres Freund, with minor modifications by me. Backpatch
to all supported versions.
Tom Lane [Fri, 7 Mar 2014 00:31:16 +0000 (19:31 -0500)]
Avoid getting more than AccessShareLock when deparsing a query.
In make_ruledef and get_query_def, we have long used AcquireRewriteLocks
to ensure that the querytree we are about to deparse is up-to-date and
the schemas of the underlying relations aren't changing. Howwever, that
function thinks the query is about to be executed, so it acquires locks
that are stronger than necessary for the purpose of deparsing. Thus for
example, if pg_dump asks to deparse a rule that includes "INSERT INTO t",
we'd acquire RowExclusiveLock on t. That results in interference with
concurrent transactions that might for example ask for ShareLock on t.
Since pg_dump is documented as being purely read-only, this is unexpected.
(Worse, it used to actually be read-only; this behavior dates back only
to 8.1, cf commit ba4200246.)
Fix this by adding a parameter to AcquireRewriteLocks to tell it whether
we want the "real" execution locks or only AccessShareLock.
Report, diagnosis, and patch by Dean Rasheed. Back-patch to all supported
branches.
Fix lastReplayedEndRecPtr calculation when starting from shutdown checkpoint.
When entering crash recovery followed by archive recovery, and the latest
checkpoint is a shutdown checkpoint, and there are no more WAL records to
replay before transitioning from crash to archive recovery, we would not
immediately allow read-only connections in hot standby mode even if we
could. That's because when starting from a shutdown checkpoint, we set
lastReplayedEndRecPtr incorrectly to the record before the checkpoint
record, instead of the checkpoint record itself. We don't run the redo
routine of the shutdown checkpoint record, but starting recovery from it
goes through the same motions, so it should be considered as replayed.
Reported by Kyotaro HORIGUCHI. All versions with hot standby are affected,
so backpatch to 9.0.
Tom Lane [Sat, 1 Mar 2014 20:21:07 +0000 (15:21 -0500)]
Allow regex operations to be terminated early by query cancel requests.
The regex code didn't have any provision for query cancel; which is
unsurprising given its non-Postgres origin, but still problematic since
some operations can take a long time. Introduce a callback function to
check for a pending query cancel or session termination request, and
call it in a couple of strategic spots where we can make the regex code
exit with an error indicator.
If we ever actually split out the regex code as a standalone library,
some additional work will be needed to let the cancel callback function
be specified externally to the library. But that's straightforward
(certainly so by comparison to putting the locale-dependent character
classification logic on a similar arms-length basis), and there seems
no need to do it right now.
A bigger issue is that there may be more places than these two where
we need to check for cancels. We can always add more checks later,
now that the infrastructure is in place.
Since there are known examples of not-terribly-long regexes that can
lock up a backend for a long time, back-patch to all supported branches.
I have hopes of fixing the known performance problems later, but adding
query cancel ability seems like a good idea even if they were all fixed.
Tom Lane [Tue, 25 Feb 2014 21:04:16 +0000 (16:04 -0500)]
Use SnapshotDirty rather than an active snapshot to probe index endpoints.
If there are lots of uncommitted tuples at the end of the index range,
get_actual_variable_range() ends up fetching each one and doing an MVCC
visibility check on it, until it finally hits a visible tuple. This is
bad enough in isolation, considering that we don't need an exact answer
only an approximate one. But because the tuples are not yet committed,
each visibility check does a TransactionIdIsInProgress() test, which
involves scanning the ProcArray. When multiple sessions do this
concurrently, the ensuing contention results in horrid performance loss.
20X overall throughput loss on not-too-complicated queries is easy to
demonstrate in the back branches (though someone's made it noticeably
less bad in HEAD).
We can dodge the problem fairly effectively by using SnapshotDirty rather
than a normal MVCC snapshot. This will cause the index probe to take
uncommitted tuples as good, so that we incur only one tuple fetch and test
even if there are many such tuples. The extent to which this degrades the
estimate is debatable: it's possible the result is actually a more accurate
prediction than before, if the endmost tuple has become committed by the
time we actually execute the query being planned. In any case, it's not
very likely that it makes the estimate a lot worse.
SnapshotDirty will still reject tuples that are known committed dead, so
we won't give bogus answers if an invalid outlier has been deleted but not
yet vacuumed from the index. (Because btrees know how to mark such tuples
dead in the index, we shouldn't have a big performance problem in the case
that there are many of them at the end of the range.) This consideration
motivates not using SnapshotAny, which was also considered as a fix.
Note: the back branches were using SnapshotNow instead of an MVCC snapshot,
but the problem and solution are the same.
Per performance complaints from Bartlomiej Romanski, Josh Berkus, and
others. Back-patch to 9.0, where the issue was introduced (by commit 40608e7f949fb7e4025c0ddd5be01939adc79eec).