Previously even if postmaster died and WaitLatch() woke up with that event
while pg_promote() was waiting for the standby promotion to finish,
pg_promote() did nothing special and kept waiting until timeout occurred.
This could cause a busy loop.
This patch make pg_promote() return false immediately when postmaster
dies, to avoid such a busy loop.
Back-patch to v12 where pg_promote() was added.
Author: Fujii Masao
Reviewed-by: Michael Paquier
Discussion: https://postgr.es/m/CAHGQGwEs9ROgSp+QF+YdDU+xP8W=CY1k-_Ov-d_Z3JY+to3eXA@mail.gmail.com
This approach provides a much tighter binding between a data directory
and the associated SysV shared memory block (and SysV or named-POSIX
semaphores, if we're using those). Key collisions are still possible,
but only between data directories stored on different filesystems,
so the situation should be negligible in practice. More importantly,
restarting the postmaster with a different port number no longer
risks failing to identify a relevant shared memory block, even when
postmaster.pid has been removed. A standalone backend is likewise
much more certain to detect conflicting leftover backends.
(In the longer term, we might now think about deprecating the port as
a cluster-wide value, so that one postmaster could support sockets
with varying port numbers. But that's for another day.)
The hazards fixed here apply only on Unix systems; our Windows code
paths already use identifiers derived from the data directory path
name rather than the port.
src/test/recovery/t/017_shm.pl, which intends to test key-collision
cases, has been substantially rewritten since it can no longer use
two postmasters with identical port numbers to trigger the case.
Instead, use Perl's IPC::SharedMem module to create a conflicting
shmem segment directly. The test script will be skipped if that
module is not available. (This means that some older buildfarm
members won't run it, but I don't think that that results in any
meaningful coverage loss.)
Patch by me; thanks to Noah Misch and Peter Eisentraut for discussion
and review.
Discussion: https://postgr.es/m/16908.1557521200@sss.pgh.pa.us
detoast.c/h contain functions required to detoast a datum, partially
or completely, plus a few other utility functions for examining the
size of toasted datums.
toast_internals.c/h contain functions that are used internally to the
TOAST subsystem but which (mostly) do not need to be accessed from
outside.
heaptoast.c/h contains code that is intrinsically specific to the
heap AM, either because it operates on HeapTuples or is based on the
layout of a heap page.
detoast.c and toast_internals.c are placed in
src/backend/access/common rather than src/backend/access/heap. At
present, both files still have dependencies on the heap, but that will
be improved in a future commit.
Patch by me, reviewed and tested by Prabhat Sabu, Thomas Munro,
Andres Freund, and Álvaro Herrera.
Discussion: http://postgr.es/m/CA+TgmoZv-=2iWM4jcw5ZhJeL18HF96+W1yJeYrnGMYdkFFnEpQ@mail.gmail.com
Use the explicit_bzero() function in places where it is important that
security information such as passwords is cleared from memory. There
might be other places where it could be useful; this is just an
initial collection.
For platforms that don't have explicit_bzero(), provide various
fallback implementations. (explicit_bzero() itself isn't standard,
but as Linux/glibc, FreeBSD, and OpenBSD have it, it's the most common
spelling, so it makes sense to make that the invocation point.)
Discussion: https://www.postgresql.org/message-id/flat/42d26bde-5d5b-c90d-87ae-6cab875f73be%402ndquadrant.com
The logic ending progress reporting for a backend entry introduced by
b6fb647 causes callers of pgstat_progress_end_command() to do some extra
work when track_activities is enabled as the process fields are reset in
the backend entry even if no command were started for reporting.
This resets the fields only if a command is registered for progress
reporting, and only if track_activities is enabled.
Author: Masahiho Sawada
Discussion: https://postgr.es/m/CAD21AoCry_vJ0E-m5oxJXGL3pnos-xYGCzF95rK5Bbi3Uf-rpA@mail.gmail.com
Backpatch-through: 9.6
Since the addition of fsync requests in bc34223 to make base backup data
consistent on disk once pg_basebackup finishes, each tablespace tar file
is individually flushed once completed, with an additional flush of the
parent directory when the base backup finishes. While holding a
connection to the server, a fsync request taking a long time may cause a
failure of the base backup, which is annoying for any integration. A
recent example of breakage can involve tcp_user_timeout, but
wal_sender_timeout can cause similar problems.
While reviewing the code, there was a second issue causing too many
fsync requests to be done for the same WAL data. As recursive fsyncs
are done at the end of the backup for both the plain and tar formats
from the base target directory where everything is written, it is fine
to disable fsyncs when fetching or streaming WAL.
Reported-by: Ryohei Takahashi
Author: Michael Paquier
Reviewed-by: Ryohei Takahashi
Discussion: https://postgr.es/m/OSBPR01MB4550DAE2F8C9502894A45AAB82BE0@OSBPR01MB4550.jpnprd01.prod.outlook.com
Backpatch-through: 10
This function is only used by xlogreader.c itself, so there's no need to
export it. It was introduced by commit 3b02ea4f07 with the apparent
intention that it could be used externally, but I couldn't find any
external code calling it.
I (Álvaro) couldn't resist the urge to sort nearby function prototypes
properly while at it.
Author: Antonin Houska
Discussion: https://postgr.es/m/14984.1554998742@spoje.net
The message was included as a parameter when this function was added in
dcb2bda9b7, but I don't think it has ever served any useful purpose.
Let's stop spreading it pointlessly.
Reviewed by Amit Langote and Peter Eisentraut.
Discussion: https://postgr.es/m/20190806224728.GA17233@alvherre.pgsql
Clarify in the help output and documentation that -n, -t etc. take a
"pattern" rather than a "schema" or "table" etc. This was especially
confusing now that the new pg_dumpall --exclude-database option was
documented with "pattern" and the others not, even though they all
behave the same.
Discussion: https://www.postgresql.org/message-id/flat/b85f3fa1-b350-38d1-1893-4f7911bd7310%402ndquadrant.com
Document that the tablespace sizes are in units of kilobytes. Make
the pg_basebackup source code a bit clearer about this, too.
Reviewed-by: Magnus Hagander <magnus@hagander.net>
The leak happens in str_tolower, str_toupper and str_initcap, which are
used in several places including their equivalent SQL-level functions,
and can only be triggered when using an ICU-provided collation when
converting the input string.
b615920 fixed a similar leak. Backpatch down 10 where ICU collations
have been introduced.
Author: Konstantin Knizhnik
Discussion: https://postgr.es/m/94c0ad0a-cbc2-e4a3-7829-2bdeaf9146db@postgrespro.ru
Backpatch-through: 10
In what seems like a fit of misplaced optimization,
ExtractReplicaIdentity() accessed the relation's replica-identity
index without taking any lock on it. Usually, the surrounding query
already holds some lock so this is safe enough ... but in the case
of a previously-planned delete, there might be no existing lock.
Given a suitable test case, this is exposed in v12 and HEAD by an
assertion added by commit b04aeb0a0.
The whole thing's rather poorly thought out anyway; rather than
looking directly at the index, we should use the index-attributes
bitmap that's held by the parent table's relcache entry, as the
caller functions do. This is more consistent and likely a bit
faster, since it avoids a cache lookup. Hence, change to doing it
that way.
While at it, rather than blithely assuming that the identity
columns are non-null (with catastrophic results if that's wrong),
add assertion checks that they aren't null. Possibly those should
be actual test-and-elog, but I'll leave it like this for now.
In principle, this is a bug that's been there since this code was
introduced (in 9.4). In practice, the risk seems quite low, since
we do have a lock on the index's parent table, so concurrent
changes to the index's catalog entries seem unlikely. Given the
precedent that commit 9c703c169 wasn't back-patched, I won't risk
back-patching this further than v12.
Per report from Hadi Moshayedi.
Discussion: https://postgr.es/m/CAK=1=Wrek44Ese1V7LjKiQS-Nd-5LgLi_5_CskGbpggKEf3tKQ@mail.gmail.com
After an unexpected connection loss and successful reconnection,
psql neglected to resynchronize its internal state about the server,
such as server version. Ordinarily we'd be reconnecting to the same
server and so this isn't really necessary, but there are scenarios
where we do need to update --- one example is where we have a list
of possible connection targets and they're not all alike.
Define "resynchronize" as including connection_warnings(), so that
this case acts the same as \connect. This seems useful; for example,
if the server version did change, the user might wish to know that.
An attuned user might also notice that the new connection isn't
SSL-encrypted, for example, though this approach isn't especially
in-your-face about such changes. Although this part is a behavioral
change, it only affects interactive sessions, so it should not break
any applications.
Also, in do_connect, make sure that we desynchronize correctly when
abandoning an old connection in non-interactive mode.
These problems evidently are the result of people patching only one
of the two places where psql deals with connection changes, so insert
some cross-referencing comments in hopes of forestalling future bugs
of the same ilk.
Lastly, in Windows builds, issue codepage mismatch warnings only at
startup, not during reconnections. psql's codepage can't change
during a reconnect, so complaining about it again seems like useless
noise.
Peter Billen and Tom Lane. Back-patch to all supported branches.
Discussion: https://postgr.es/m/CAMTXbE8e6U=EBQfNSe01Ej17CBStGiudMAGSOPaw-ALxM-5jXg@mail.gmail.com
Similarly to the signed versions added in 4d6ad31, this adds a set of
inline functions for overflow checks with unsigned integers, including
uint16, uint32 and uint64. This relies on compiler built-in overflow
checks by default if available. The behavior of unsigned integers is
well-defined so the fallback implementations checks are simple for
additions and subtractions. Multiplications avoid division-based checks
which are expensive if possible, still this can happen for uint64 if
128-bit integers are not available.
While on it, the code in common/int.h is reorganized to avoid too many
duplicated comments. The new macros will be used in a follow-up patch.
All thanks to Andres Freund for the input provided.
Author: Fabien Coelho, Michael Paquier
Discussion: https://postgr.es/m/20190830073423.GB2354@paquier.xyz
The comment describing the string format was a lie. Make it agree with
reality, add/improve some other comments, fix coding style for loops with
empty bodies. Also add an Assert that we counted parameters correctly,
because the spread-out logic for that looks pretty fragile.
No actual bugs fixed here, so no need to back-patch.
Discussion: https://postgr.es/m/848B1649C8A6274AA527C4472CA11EDD5FC70CBE@G01JPEXMBYT02
The comment did not match what the code actually did for integers with
the 43rd bit set. You get an integer like that, if you have a posting
list with two adjacent TIDs that are more than 2^31 blocks apart.
According to the comment, we would store that in 6 bytes, with no
continuation bit on the 6th byte, but in reality, the code encodes it
using 7 bytes, with a continuation bit on the 6th byte as normal.
The decoding routine also handled these 7-byte integers correctly, except
for an overflow check that assumed that one integer needs at most 6 bytes.
Fix the overflow check, and fix the comment to match what the code
actually does. Also fix the comment that claimed that there are 17 unused
bits in the 64-bit representation of an item pointer. In reality, there
are 64-32-11=21.
Fitting any item pointer into max 6 bytes was an important property when
this was written, because in the old pre-9.4 format, item pointers were
stored as plain arrays, with 6 bytes for every item pointer. The maximum
of 6 bytes per integer in the new format guaranteed that we could convert
any page from the old format to the new format after upgrade, so that the
new format was never larger than the old format. But we hardly need to
worry about that anymore, and running into that problem during upgrade,
where an item pointer is expanded from 6 to 7 bytes such that the data
doesn't fit on a page anymore, is implausible in practice anyway.
Backpatch to all supported versions.
This also includes a little test module to test these large distances
between item pointers, without requiring a 16 TB table. It is not
backpatched, I'm including it more for the benefit of future development
of new posting list formats.
Discussion: https://www.postgresql.org/message-id/33bfc20a-5c86-f50c-f5a5-58e9925d05ff%40iki.fi
Reviewed-by: Masahiko Sawada, Alexander Korotkov
RelationAllowsEarlyPruning() performed a catalog scan, but is used
in two contexts where that was a bad idea:
1. In heap_page_prune_opt(), which runs very frequently in some large
scans. This caused major performance problems in a field report
that was easy to reproduce.
2. In TestForOldSnapshot(), which runs while we hold a buffer content
lock. It's not clear if this was guaranteed to be free of buffer
deadlock risk.
The check was introduced in commit 2cc41acd8 and defended against a
real problem: 9.6's hash indexes have no page LSN and so we can't
allow early pruning (ie the snapshot-too-old feature). We can remove
the check from all later releases though: hash indexes are now logged,
and there is no way to create UNLOGGED indexes on regular logged
tables.
If a future release allows such a combination, it might need to put
a similar check in place, but it'll need some more thought.
Back-patch to 10.
Author: Thomas Munro
Reviewed-by: Tom Lane, who spotted the second problem
Discussion: https://postgr.es/m/CA%2BhUKGKT8oTkp5jw_U4p0S-7UG9zsvtw_M47Y285bER6a2gD%2Bg%40mail.gmail.com
Discussion: https://postgr.es/m/CAA4eK1%2BWy%2BN4eE5zPm765h68LrkWc3Biu_8rzzi%2BOYX4j%2BiHRw%40mail.gmail.com
check_float4_val() checks after underflow and overflow of values
converted from float8 to float4, but there has never been any regression
tests for that. This brings the coverage of float.h to 100%.
Author: Movead Li
Discussion: https://postgr.es/m/20190822174636998766188@highgo.ca
In this case, the transfer uses a libpq connection, which is subject to
the timeout parameters set at system level, and this can make the rewind
operation suddenly canceled which is not good for automation. One
workaround to such issues would be to use PGOPTIONS to enforce the
wanted timeout parameters, but that's annoying, and for example pg_dump,
which can run potentially long-running queries disables all types of
timeouts.
lock_timeout and statement_timeout are the ones which can cause problems
now. Note that pg_rewind does not use transactions, so disabling
idle_in_transaction_session_timeout is optional, but it feels safer to
do so for the future.
This is back-patched down to 9.5. idle_in_transaction_session_timeout
is only present since 9.6.
Author: Alexander Kukushkin
Discussion: https://postgr.es/m/CAFh8B=krcVXksxiwVQh1SoY+ziJ-JC=6FcuoBL3yce_40Es5_g@mail.gmail.com
Backpatch-through: 9.5
Commit a4327296d taught pg_regress proper to do this, but
missed the opportunity to do likewise in the isolationtester
and ecpg variants of pg_regress. Seems like this might be
helpful for tracking down issues exposed by those tests.
Turns out that returning "unrecognized signal" is confusing.
Make it explicit that the platform lacks any support for signal names.
(At least of the machines in the buildfarm, only HPUX lacks it.)
Back-patch to v12 where we invented this function.
Discussion: https://postgr.es/m/3067.1566870481@sss.pgh.pa.us
Commit efada2b8e9, which made the nbtree page deletion algorithm more
robust, removed the concept of a half-dead internal page. Remove a
comment about half dead parent pages that was overlooked.
An empty file name or subdirectory name leads join_path_components() to
just produce the parent directory name, which leads to weird failures or
recursive inclusions. Let's throw a specific error for that. It takes
only slightly more code to detect all-blank names, so do so.
Also, detect direct recursion, ie a file calling itself. As coded
this will also detect recursion via "include_dir '.'", which is
perhaps more likely than explicitly including the file itself.
Detecting indirect recursion would require API changes for guc-file.l
functions, which seems not worth it since extensions might call them.
The nesting depth limit will catch such cases eventually, just not
with such an on-point error message.
In passing, adjust the example usages in postgresql.conf.sample
to perhaps eliminate the problem at the source: there's no reason
for the examples to suggest that an empty value is valid.
Per a trouble report from Brent Bates. Back-patch to 9.5; the
issue is old, but the code in 9.4 is enough different that the
patch doesn't apply easily, and it doesn't seem worth the trouble
to fix there.
Ian Barwick and Tom Lane
Discussion: https://postgr.es/m/8c8bcbca-3bd9-dc6e-8986-04a5abdef142@2ndquadrant.com
FD_SETSIZE needs to be declared before winsock2.h, or it is possible to
run into buffer overflow issues when using --jobs. This is similar to
pgbench's solution done in a23c641.
This has been introduced by 71d84ef, and older versions have been using
the default value of FD_SETSIZE, defined at 64.
Per buildfarm member jacana, but this impacts all Windows animals
running the TAP tests. I have reproduced the failure locally to check
the patch.
Author: Michael Paquier
Reviewed-by: Andrew Dunstan
Discussion: https://postgr.es/m/20190826054000.GE7005@paquier.xyz
Backpatch-through: 9.5
If a test case tried to set an invalid value of synchronous_standby_names,
the test script didn't detect that, which seems like a bad idea.
Noticed while testing a proposed patch that broke some of these
test cases.
A report from Alvaro Herrera shows that if we're in PM_STARTUP
state, and we spawn a dead_end child to reject some incoming
connection request, and that child dies with an unexpected exit
code, the postmaster does not respond well. We correctly send
SIGQUIT to the startup process, but then:
* if the startup process exits with nonzero exit code, as expected,
we thought that that indicated a crash and aborted startup.
* if the startup process exits with zero exit code, which is possible
due to the inherent race condition, we'd advance to PM_RUN state
which is fine --- but the code forgot that AbortStartTime would be
nonzero in this situation. We'd either die on the Asserts saying
that it was zero, or perhaps misbehave later on. (A quick look
suggests that the only misbehavior might be busy-waiting due to
DetermineSleepTime doing the wrong thing.)
To fix the first point, adjust the state-machine logic to recognize
that a nonzero exit code is expected after sending SIGQUIT, and have
it transition to a state where we can restart the startup process.
To fix the second point, change the Asserts to clear the variable
rather than just claiming it should be clear already.
Perhaps we could improve this further by not treating a crash of
a dead_end child as a reason for panic'ing the database. However,
since those child processes are connected to shared memory, that
seems a bit risky. There are few good reasons for a dead_end child
to report failure anyway (the cause of this in Alvaro's report is
quite unclear). On balance, therefore, a minimal fix seems best.
This is an oversight in commit 45811be94. While that was back-patched,
I'm hesitant to back-patch this change. The lack of reasons for a
dead_end child to fail suggests that the case should be very rare in
the field, which squares with the lack of reports; so it seems like
this might not be worth the risk of introducing new issues. In any
case we can let it bake awhile in HEAD before considering a back-patch.
Discussion: https://postgr.es/m/20190615160950.GA31378@alvherre.pgsql
Previously 'uname -r' on Msys2 reported a kernele release starting with
2. The latest version starts with 3. In commit 1638623f we specifically
looked for one starting with 2. This is now changed to look for any
digit between 2 and 9.
backpatch to release 10.
On msys2, 'uname -s' reports a string starting MSYS instead on MINGW
as happens on msys1. Treat these both the same way. This reverts
608a710195 in favor of a more general solution.
Backpatch to all live branches.
When trying to use a high number of jobs, vacuumdb (and more recently
reindexdb) has only checked for a maximum number of jobs used, causing
confusing failures when running out of file descriptors when the jobs
open connections to Postgres. This commit changes the error handling so
as we do not check anymore for a maximum number of allowed jobs when
parsing the option value with FD_SETSIZE, but check instead if a file
descriptor is within the supported range when opening the connections
for the jobs so as this is detected at the earliest time possible.
Also, improve the error message to give a hint about the number of jobs
recommended, using a wording given by the reviewers of the patch.
Reported-by: Andres Freund
Author: Michael Paquier
Reviewed-by: Andres Freund, Álvaro Herrera, Tom Lane
Discussion: https://postgr.es/m/20190818001858.ho3ev4z57fqhs7a5@alap3.anarazel.de
Backpatch-through: 9.5
POSIX permits getopt() to advance optind beyond argc when the last
argv entry is an option that requires an argument and hasn't got one.
It seems that no major platforms actually do that, but musl does,
so that something like "psql -f" would crash with that libc.
Add a check that optind is in range before trying to look at the
possibly-bogus option.
Report and fix by Quentin Rameau. Back-patch to all supported
branches.
Discussion: https://postgr.es/m/20190825100617.GA6087@fifth.space
We were setting extra_float_digits = 0 to avoid platform-dependent
output in this test, but that's still able to expose platform-specific
roundoff behavior in some new test cases added by commit a3d284485,
as reported by Peter Eisentraut. Reduce it to -1 to hide that.
(Over in geometry.sql, we're using -3, which is an ancient decision
dating to 337f73b1b. I wonder whether that's overkill now. But
there's probably little value in trying to change it.)
Back-patch to v12 where a3d284485 came in; there's no evidence that
we have any platform-dependent issues here before that.
Discussion: https://postgr.es/m/15551268-e224-aa46-084a-124b64095ee3@2ndquadrant.com
Bleeding-edge LLVM has stopped supplying replacements for various
C++14 library features, for people on older C++ versions. Since we're
not ready to require C++14 yet, just use plain old new instead of
make_unique. As revealed by buildfarm animal seawasp.
Back-patch to 11.
Reviewed-by: Andres Freund
Discussion: https://postgr.es/m/CA%2BhUKGJWG7unNqmkxg7nC5o3o-0p2XP6co4r%3D9epqYMm8UY4Mw%40mail.gmail.com
The Postgres approach to coupling locks during an ascent of the tree is
slightly different to the approach taken by Lehman and Yao. Add a new
paragraph to the "Differences to the Lehman & Yao algorithm" section of
the nbtree README that explains the similarities and differences.
This is useful for developers to find out if an isolation spec is
over-engineered or if it needs more work by warning at the end of a
test run if a step is not used, generating a failure with extra diffs.
While on it, clean up all the specs which include steps not used in any
permutations to simplify them.
Author: Michael Paquier
Reviewed-by: Asim Praveen, Melanie Plageman
Discussion: https://postgr.es/m/20190819080820.GG18166@paquier.xyz
The original purpose of the dry-run mode is to be able to print all the
possible permutations from a spec file, but it has become less useful
since isolation tests has improved regarding deadlock detection as one
step not wanted by the author could block indefinitely now (originally
the step blocked would have been detected rather quickly). Per
discussion, let's remove it.
Author: Michael Paquier
Reviewed-by: Asim Praveen, Melanie Plageman
Discussion: https://postgr.es/m/20190819080820.GG18166@paquier.xyz
Adjust the struct comment that describes how page splits use their
descent stack to cascade up the tree from the leaf level.
In passing, fix up some unrelated nbtree comments that had typos or were
obsolete.
Peter Eisentraut
Fujii Masao
Tom Lane
Robert Haas
Michael Paquier
Alvaro Herrera
Etsuro Fujita
Heikki Linnakangas
Thomas Munro
Peter Geoghegan
Andrew Dunstan