Autoconf is showing its age, fewer and fewer contributors know how to wrangle
it. Recursive make has a lot of hard to resolve dependency issues and slow
incremental rebuilds. Our home-grown MSVC build system is hard to maintain for
developers not using Windows and runs tests serially. While these and other
issues could individually be addressed with incremental improvements, together
they seem best addressed by moving to a more modern build system.
After evaluating different build system choices, we chose to use meson, to a
good degree based on the adoption by other open source projects.
We decided that it's more realistic to commit a relatively early version of
the new build system and mature it in tree.
This commit adds an initial version of a meson based build system. It supports
building postgres on at least AIX, FreeBSD, Linux, macOS, NetBSD, OpenBSD,
Solaris and Windows (however only gcc is supported on aix, solaris). For
Windows/MSVC postgres can now be built with ninja (faster, particularly for
incremental builds) and msbuild (supporting the visual studio GUI, but
building slower).
Several aspects (e.g. Windows rc file generation, PGXS compatibility, LLVM
bitcode generation, documentation adjustments) are done in subsequent commits
requiring further review. Other aspects (e.g. not installing test-only
extensions) are not yet addressed.
When building on Windows with msbuild, builds are slower when using a visual
studio version older than 2019, because those versions do not support
MultiToolTask, required by meson for intra-target parallelism.
The plan is to remove the MSVC specific build system in src/tools/msvc soon
after reaching feature parity. However, we're not planning to remove the
autoconf/make build system in the near future. Likely we're going to keep at
least the parts required for PGXS to keep working around until all supported
versions build with meson.
Some initial help for postgres developers is at
https://wiki.postgresql.org/wiki/Meson
With contributions from Thomas Munro, John Naylor, Stone Tickle and others.
Author: Andres Freund <andres@anarazel.de>
Author: Nazir Bilal Yavuz <byavuz81@gmail.com>
Author: Peter Eisentraut <peter@eisentraut.org>
Reviewed-By: Peter Eisentraut <peter.eisentraut@enterprisedb.com>
Discussion: https://postgr.es/m/20211012083721.hvixq4pnh2pixr3j@alap3.anarazel.de
Make sure that function declarations use names that exactly match the
corresponding names from function definitions in optimizer, parser,
utility, libpq, and "commands" code, as well as in remaining library
code. Do the same for all code related to frontend programs (with the
exception of pg_dump/pg_dumpall related code).
Like other recent commits that cleaned up function parameter names, this
commit was written with help from clang-tidy. Later commits will handle
ecpg and pg_dump/pg_dumpall.
Author: Peter Geoghegan <pg@bowt.ie>
Reviewed-By: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/CAH2-WznJt9CMM9KJTMjJh_zbL5hD9oX44qdJ4aqZtjFi-zA3Tg@mail.gmail.com
As reported by Yura Sokolov, scanning the snapshot->xip array has
noticeable impact on scalability when there are a large number of
concurrent writers. Use the optimized (on x86-64) routine from b6ef16756
to speed up searches through the [sub]xip arrays. One benchmark showed
a 5% increase in transaction throughput with 128 concurrent writers,
and a 50% increase in a pathological case of 1024 writers. While a hash
table would have scaled even better, it was ultimately rejected because
of concerns around code complexity and memory allocation. Credit to Andres
Freund for the idea to optimize linear search using SIMD instructions.
Nathan Bossart
Reviewed by: Andres Freund, John Naylor, Bharath Rupireddy, Masahiko Sawada
Discussion: https://postgr.es/m/20220713170950.GA3116318%40nathanxps13
This function gave the wrong answer when there's more than one
RegisteredSnapshots entry, whether or not any of them is the
CatalogSnapshot. This leads to assertion failure in some scenarios
involving fetching toasted data using a cursor. (As per discussion,
I'm dubious that this is the right contract to be enforcing at all;
but it surely doesn't help to be enforcing it incorrectly.)
Fetching toasted data using a cursor is evidently under-tested,
so add a test case too.
Per report from Erik Rijkers. This is new code, so no need for
back-patch.
Discussion: https://postgr.es/m/dc9dd229-ed30-6c62-4c41-d733ffff776b@xs4all.nl
Most of these are cases where we could call memcpy() or other libc
functions with a NULL pointer and a zero count, which is forbidden
by POSIX even though every production version of libc allows it.
We've fixed such things before in a piecemeal way, but apparently
never made an effort to try to get them all. I don't claim that
this patch does so either, but it gets every failure I observe in
check-world, using clang 12.0.1 on current RHEL8.
numeric.c has a different issue that the sanitizer doesn't like:
"ln(-1.0)" will compute log10(0) and then try to assign the
resulting -Inf to an integer variable. We don't actually use the
result in such a case, so there's no live bug.
Back-patch to all supported branches, with the idea that we might
start running a buildfarm member that tests this case. This includes
back-patching c1132aae3 (Check the size in COPY_POINTER_FIELD),
which previously silenced some of these issues in copyfuncs.c.
Discussion: https://postgr.es/m/CALNJ-vT9r0DSsAOw9OXVJFxLENoVS_68kJ5x0p44atoYH+H4dg@mail.gmail.com
Commit <FIXME> fixed the bug that RemoveTempRelationsCallback() did not
push/register a snapshot. That only went unnoticed because often a valid
catalog snapshot exists and is returned by GetOldestSnapshot(). But due to
invalidation processing that is not reliable.
Thus assert in init_toast_snapshot() that there is a registered or active
snapshot, using the new HaveRegisteredOrActiveSnapshot().
Author: Andres Freund
Discussion: https://postgr.es/m/20220219180002.6tubjq7iw7m52bgd@alap3.anarazel.de
Commit 84f5c2908 forgot to consider the possibility that
EnsurePortalSnapshotExists could run inside a subtransaction with
lifespan shorter than the Portal's. In that case, the new active
snapshot would be popped at the end of the subtransaction, leaving
a dangling pointer in the Portal, with mayhem ensuing.
To fix, make sure the ActiveSnapshot stack entry is marked with
the same subtransaction nesting level as the associated Portal.
It's certainly safe to do so since we won't be here at all unless
the stack is empty; hence we can't create an out-of-order stack.
Let's also apply this logic in the case where PortalRunUtility
sets portalSnapshot, just to be sure that path can't cause similar
problems. It's slightly less clear that that path can't create
an out-of-order stack, so add an assertion guarding it.
Report and patch by Bertrand Drouvot (with kibitzing by me).
Back-patch to v11, like the previous commit.
Discussion: https://postgr.es/m/ff82b8c5-77f4-3fe7-6028-fcf3303e82dd@amazon.com
Also "make reformat-dat-files".
The only change worthy of note is that pgindent messed up the formatting
of launcher.c's struct LogicalRepWorkerId, which led me to notice that
that struct wasn't used at all anymore, so I just took it out.
Given a permanent relation rewritten in the current transaction, the
old_snapshot_threshold mechanism assumed the relation had never been
subject to early pruning. Hence, a query could fail to report "snapshot
too old" when the rewrite followed an early truncation. ALTER TABLE SET
TABLESPACE is probably the only rewrite mechanism capable of exposing
this bug. REINDEX sets indcheckxmin, avoiding the problem. CLUSTER has
zeroed page LSNs since before old_snapshot_threshold existed, so
old_snapshot_threshold has never cooperated with it. ALTER TABLE
... SET DATA TYPE makes the table look empty to every past snapshot,
which is strictly worse. Back-patch to v13, where commit
c6b92041d3 broke this.
Kyotaro Horiguchi and Noah Misch
Discussion: https://postgr.es/m/20210113.160705.2225256954956139776.horikyota.ntt@gmail.com
Invent a new flag bit HASH_STRINGS to specify C-string hashing, which
was formerly the default; and add assertions insisting that exactly
one of the bits HASH_STRINGS, HASH_BLOBS, and HASH_FUNCTION be set.
This is in hopes of preventing recurrences of the type of oversight
fixed in commit a1b8aa1e4 (i.e., mistakenly omitting HASH_BLOBS).
Also, when HASH_STRINGS is specified, insist that the keysize be
more than 8 bytes. This is a heuristic, but it should catch
accidental use of HASH_STRINGS for integer or pointer keys.
(Nearly all existing use-cases set the keysize to NAMEDATALEN or
more, so there's little reason to think this restriction should
be problematic.)
Tweak hash_create() to insist that the HASH_ELEM flag be set, and
remove the defaults it had for keysize and entrysize. Since those
defaults were undocumented and basically useless, no callers
omitted HASH_ELEM anyway.
Also, remove memset's zeroing the HASHCTL parameter struct from
those callers that had one. This has never been really necessary,
and while it wasn't a bad coding convention it was confusing that
some callers did it and some did not. We might as well save a few
cycles by standardizing on "not".
Also improve the documentation for hash_create().
In passing, improve reinit.c's usage of a hash table by storing
the key as a binary Oid rather than a string; and, since that's
a temporary hash table, allocate it in CurrentMemoryContext for
neatness.
Discussion: https://postgr.es/m/590625.1607878171@sss.pgh.pa.us
The previous coding was confused about whether head_timestamp was
intended to represent the timestamp for the newest bucket in the
mapping or the oldest timestamp for the oldest bucket in the mapping.
Decide that it's intended to be the oldest one, and repair
accordingly.
To do that, we need to do two things. First, when advancing to a
new bucket, don't categorically set head_timestamp to the new
timestamp. Do this only if we're blowing out the map completely
because a lot of time has passed since we last maintained it. If
we're replacing entries one by one, advance head_timestamp by
1 minute for each; if we're filling in unused entries, don't
advance head_timestamp at all.
Second, fix the computation of how many buckets we need to advance.
The previous formula would be correct if head_timestamp were the
timestamp for the new bucket, but we're now making all the code
agree that it's the timestamp for the oldest bucket, so adjust the
formula accordingly.
This is certainly a bug fix, but I don't feel good about
back-patching it without the introspection tools added by commit
aecf5ee2bb, and perhaps also some
actual tests. Since back-patching the introspection tools might
not attract sufficient support and since there are no automated
tests of these fixes yet, I'm just committing this to master for
now.
Patch by me, reviewed by Thomas Munro, Dilip Kumar, Hamid Akhtar.
Discussion: http://postgr.es/m/CA+TgmoY=aqf0zjTD+3dUWYkgMiNDegDLFjo+6ze=Wtpik+3XqA@mail.gmail.com
This makes it possible for code outside snapmgr.c to examine the
contents of this data structure. This commit does not add any code
which actually does so; a subsequent commit will make that change.
Patch by me, reviewed by Thomas Munro, Dilip Kumar, Hamid Akhtar.
Discussion: http://postgr.es/m/CA+TgmoY=aqf0zjTD+3dUWYkgMiNDegDLFjo+6ze=Wtpik+3XqA@mail.gmail.com
Previous commits made it faster/more scalable to compute snapshots. But not
building a snapshot is still faster. Now that GetSnapshotData() does not
maintain RecentGlobal* anymore, that is actually not too hard:
This commit introduces xactCompletionCount, which tracks the number of
top-level transactions with xids (i.e. which may have modified the database)
that completed in some form since the start of the server.
We can avoid rebuilding the snapshot's contents whenever the current
xactCompletionCount is the same as it was when the snapshot was
originally built. Currently this check happens while holding
ProcArrayLock. While it's likely possible to perform the check without
acquiring ProcArrayLock, it seems better to do that separately /
later, some careful analysis is required. Even with the lock this is a
significant win on its own.
On a smaller two socket machine this gains another ~1.03x, on a larger
machine the effect is roughly double (earlier patch version tested
though). If we were able to safely avoid the lock there'd be another
significant gain on top of that.
Author: Andres Freund <andres@anarazel.de>
Reviewed-By: Robert Haas <robertmhaas@gmail.com>
Reviewed-By: Thomas Munro <thomas.munro@gmail.com>
Reviewed-By: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
xact.h included utils/datetime.h, which cannot be used in the frontend
(it includes fmgr.h, which needs Datum). But xact.h only needs the
definition of TimestampTz from it, which is available directly in
datatypes/timestamp.h. Change xact.h to include that instead of
utils/datetime.h, so that it can be used in client programs.
Now that xmin isn't needed for GetSnapshotData() anymore, it leads to
unnecessary cacheline ping-pong to have it in PGXACT, as it is updated
considerably more frequently than the other PGXACT members.
After the changes in dc7420c2c9, this is a very straight-forward change.
For highly concurrent, snapshot acquisition heavy, workloads this change alone
can significantly increase scalability. E.g. plain pgbench on a smaller 2
socket machine gains 1.07x for read-only pgbench, 1.22x for read-only pgbench
when submitting queries in batches of 100, and 2.85x for batches of 100
'SELECT';. The latter numbers are obviously not to be expected in the
real-world, but micro-benchmark the snapshot computation
scalability (previously spending ~80% of the time in GetSnapshotData()).
Author: Andres Freund <andres@anarazel.de>
Reviewed-By: Robert Haas <robertmhaas@gmail.com>
Reviewed-By: Thomas Munro <thomas.munro@gmail.com>
Reviewed-By: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
To make GetSnapshotData() more scalable, it cannot not look at at each proc's
xmin: While snapshot contents do not need to change whenever a read-only
transaction commits or a snapshot is released, a proc's xmin is modified in
those cases. The frequency of xmin modifications leads to, particularly on
higher core count systems, many cache misses inside GetSnapshotData(), despite
the data underlying a snapshot not changing. That is the most
significant source of GetSnapshotData() scaling poorly on larger systems.
Without accessing xmins, GetSnapshotData() cannot calculate accurate horizons /
thresholds as it has so far. But we don't really have to: The horizons don't
actually change that much between GetSnapshotData() calls. Nor are the horizons
actually used every time a snapshot is built.
The trick this commit introduces is to delay computation of accurate horizons
until there use and using horizon boundaries to determine whether accurate
horizons need to be computed.
The use of RecentGlobal[Data]Xmin to decide whether a row version could be
removed has been replaces with new GlobalVisTest* functions. These use two
thresholds to determine whether a row can be pruned:
1) definitely_needed, indicating that rows deleted by XIDs >= definitely_needed
are definitely still visible.
2) maybe_needed, indicating that rows deleted by XIDs < maybe_needed can
definitely be removed
GetSnapshotData() updates definitely_needed to be the xmin of the computed
snapshot.
When testing whether a row can be removed (with GlobalVisTestIsRemovableXid())
and the tested XID falls in between the two (i.e. XID >= maybe_needed && XID <
definitely_needed) the boundaries can be recomputed to be more accurate. As it
is not cheap to compute accurate boundaries, we limit the number of times that
happens in short succession. As the boundaries used by
GlobalVisTestIsRemovableXid() are never reset (with maybe_needed updated by
GetSnapshotData()), it is likely that further test can benefit from an earlier
computation of accurate horizons.
To avoid regressing performance when old_snapshot_threshold is set (as that
requires an accurate horizon to be computed), heap_page_prune_opt() doesn't
unconditionally call TransactionIdLimitedForOldSnapshots() anymore. Both the
computation of the limited horizon, and the triggering of errors (with
SetOldSnapshotThresholdTimestamp()) is now only done when necessary to remove
tuples.
This commit just removes the accesses to PGXACT->xmin from
GetSnapshotData(), but other members of PGXACT residing in the same
cache line are accessed. Therefore this in itself does not result in a
significant improvement. Subsequent commits will take advantage of the
fact that GetSnapshotData() now does not need to access xmins anymore.
Note: This contains a workaround in heap_page_prune_opt() to keep the
snapshot_too_old tests working. While that workaround is ugly, the tests
currently are not meaningful, and it seems best to address them separately.
Author: Andres Freund <andres@anarazel.de>
Reviewed-By: Robert Haas <robertmhaas@gmail.com>
Reviewed-By: Thomas Munro <thomas.munro@gmail.com>
Reviewed-By: David Rowley <dgrowleyml@gmail.com>
Discussion: https://postgr.es/m/20200301083601.ews6hz5dduc3w2se@alap3.anarazel.de
When maintaining or merging patches, one of the most common sources
for conflicts are the list of objects in makefiles. Especially when
the split across lines has been changed on both sides, which is
somewhat common due to attempting to stay below 80 columns, those
conflicts are unnecessarily laborious to resolve.
By splitting, and alphabetically sorting, OBJS style lines into one
object per line, conflicts should be less frequent, and easier to
resolve when they still occur.
Author: Andres Freund
Discussion: https://postgr.es/m/20191029200901.vww4idgcxv74cwes@alap3.anarazel.de
Otherwise, after a deleted page gets even older, it becomes unrecyclable
again. B-tree has the same problem, and has had since time immemorial,
but let's at least fix this in GiST, where this is new.
Backpatch to v12, where GiST page deletion was introduced.
Reviewed-by: Andrey Borodin
Discussion: https://www.postgresql.org/message-id/835A15A5-F1B4-4446-A711-BF48357EB602%40yandex-team.ru
When building an initial slot snapshot, snapshots are marked with
historic MVCC snapshots as type with the marker field being set in
SnapBuildBuildSnapshot() but not overriden in SnapBuildInitialSnapshot().
Existing callers of SnapBuildBuildSnapshot() do not care about the type
of snapshot used, but extensions calling it actually may, as reported.
While on it, mark correctly the snapshot type when importing one. This
is cosmetic as the field is enforced to 0.
Author: Antonin Houska
Reviewed-by: Álvaro Herrera, Michael Paquier
Discussion: https://postgr.es/m/23215.1527665193@localhost
Backpatch-through: 9.4
Given these routines are heap specific, and that there will be more
generic visibility support in via table AM, it makes sense to move the
prototypes to heapam.h (routines like HeapTupleSatisfiesVacuum will
not be exposed in a generic fashion, because they are too storage
specific).
Similarly, the code in tqual.c is specific to heap, so moving it into
access/heap/ makes sense.
Author: Andres Freund
Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
The code in tqual.c is largely heap specific. Due to the upcoming
pluggable storage work, it therefore makes sense to move it into
access/heap/ (as the file's header notes, the tqual name isn't very
good).
But the various statically allocated snapshot and snapshot
initialization functions are now (see previous commit) generic and do
not depend on functions declared in tqual.h anymore. Therefore move.
Also move XidInMVCCSnapshot as that's useful for future AMs, and
already used outside of tqual.c.
Author: Andres Freund
Discussion: https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.de
This is in preparation for allowing the same snapshot be used for
different table AMs. With the current callback based approach we would
need one callback for each supported AM, which clearly would not be
extensible. Thus add a new Snapshot->snapshot_type field, and move
the dispatch into HeapTupleSatisfiesVisibility() (which is now a
function). Later work will then dispatch calls to
HeapTupleSatisfiesVisibility() and other AMs visibility functions
depending on the type of the table. The central SnapshotType enum
also seems like a good location to centralize documentation about the
intended behaviour of various types of snapshots.
As tqual.h isn't included by bufmgr.h any more (as HeapTupleSatisfies*
isn't referenced by TestForOldSnapshot() anymore) a few files now need
to include it directly.
Author: Andres Freund, loosely based on earlier work by Haribabu Kommi
Discussion:
https://postgr.es/m/20180703070645.wchpu5muyto5n647@alap3.anarazel.dehttps://postgr.es/m/20160812231527.GA690404@alvherre.pgsql
A collection of typos I happened to spot while reading code, as well as
grepping for common mistakes.
Backpatch to all supported versions, as applicable, to avoid conflicts
when backporting other commits in the future.
We call this thing a "transaction block" everywhere except in a few
functions, where it is mysteriously called a "transaction chain". In
the SQL standard, a transaction chain is something different. So rename
these functions to match the common terminology.
Reviewed-by: Alvaro Herrera <alvherre@alvh.no-ip.org>
Previously it was possible that a tuple was not pruned during vacuum,
even though its update xmax (i.e. the updating xid in a multixact with
both key share lockers and an updater) was below the cutoff horizon.
As the freezing code assumed, rightly so, that that's not supposed to
happen, xmax would be preserved (as a member of a new multixact or
xmax directly). That causes two problems: For one the tuple is below
the xmin horizon, which can cause problems if the clog is truncated or
once there's an xid wraparound. The bigger problem is that that will
break HOT chains, which in turn can lead two to breakages: First,
failing index lookups, which in turn can e.g lead to constraints being
violated. Second, future hot prunes / vacuums can end up making
invisible tuples visible again. There's other harmful scenarios.
Fix the problem by recognizing that tuples can be DEAD instead of
RECENTLY_DEAD, even if the multixactid has alive members, if the
update_xid is below the xmin horizon. That's safe because newer
versions of the tuple will contain the locking xids.
A followup commit will harden the code somewhat against future similar
bugs and already corrupted data.
Author: Andres Freund, with changes by Alvaro Herrera
Reported-By: Daniel Wood
Analyzed-By: Andres Freund, Alvaro Herrera, Robert Haas, Peter
Geoghegan, Daniel Wood, Yi Wen Wong, Michael Paquier
Reviewed-By: Alvaro Herrera, Robert Haas, Michael Paquier
Discussion:
https://postgr.es/m/E5711E62-8FDF-4DCA-A888-C200BF6B5742@amazon.comhttps://postgr.es/m/20171102112019.33wb7g5wp4zpjelu@alap3.anarazel.de
Backpatch: 9.3-
This patch fixes a couple of low-probability bugs that could lead to
reporting an irrelevant errno value (and hence possibly a wrong SQLSTATE)
concerning directory-open or file-open failures. It also fixes places
where we took shortcuts in reporting such errors, either by using elog
instead of ereport or by using ereport but forgetting to specify an
errcode. And it eliminates a lot of just plain redundant error-handling
code.
In service of all this, export fd.c's formerly-static function
ReadDirExtended, so that external callers can make use of the coding
pattern
dir = AllocateDir(path);
while ((de = ReadDirExtended(dir, path, LOG)) != NULL)
if they'd like to treat directory-open failures as mere LOG conditions
rather than errors. Also fix FreeDir to be a no-op if we reach it
with dir == NULL, as such a coding pattern would cause.
Then, remove code at many call sites that was throwing an error or log
message for AllocateDir failure, as ReadDir or ReadDirExtended can handle
that job just fine. Aside from being a net code savings, this gets rid of
a lot of not-quite-up-to-snuff reports, as mentioned above. (In some
places these changes result in replacing a custom error message such as
"could not open tablespace directory" with more generic wording "could not
open directory", but it was agreed that the custom wording buys little as
long as we report the directory name.) In some other call sites where we
can't just remove code, change the error reports to be fully
project-style-compliant.
Also reorder code in restoreTwoPhaseData that was acquiring a lock
between AllocateDir and ReadDir; in the unlikely but surely not
impossible case that LWLockAcquire changes errno, AllocateDir failures
would be misreported. There is no great value in opening the directory
before acquiring TwoPhaseStateLock, so just do it in the other order.
Also fix CheckXLogRemoved to guarantee that it preserves errno,
as quite a number of call sites are implicitly assuming. (Again,
it's unlikely but I think not impossible that errno could change
during a SpinLockAcquire. If so, this function was broken for its
own purposes as well as breaking callers.)
And change a few places that were using not-per-project-style messages,
such as "could not read directory" when "could not open directory" is
more correct.
Back-patch the exporting of ReadDirExtended, in case we have occasion
to back-patch some fix that makes use of it; it's not needed right now
but surely making it global is pretty harmless. Also back-patch the
restoreTwoPhaseData and CheckXLogRemoved fixes. The rest of this is
essentially cosmetic and need not get back-patched.
Michael Paquier, with a bit of additional work by me
Discussion: https://postgr.es/m/CAB7nPqRpOCxjiirHmebEFhXVTK7V5Jvw4bz82p7Oimtsm3TyZA@mail.gmail.com
The lower case spellings are C and C++ standard and are used in most
parts of the PostgreSQL sources. The upper case spellings are only used
in some files/modules. So standardize on the standard spellings.
The APIs for ICU, Perl, and Windows define their own TRUE and FALSE, so
those are left as is when using those APIs.
In code comments, we use the lower-case spelling for the C concepts and
keep the upper-case spelling for the SQL concepts.
Reviewed-by: Michael Paquier <michael.paquier@gmail.com>
Up to now async.c has used TransactionIdIsInProgress() to detect whether
a notify message's source transaction is still running. However, that
function has a quick-exit path that reports that XIDs before RecentXmin
are no longer running. If a listening backend is doing nothing but
listening, and not running any queries, there is nothing that will advance
its value of RecentXmin. Once 2 billion transactions elapse, the
RecentXmin check causes active transactions to be reported as not running.
If they aren't committed yet according to CLOG, async.c decides they
aborted and discards their messages. The timing for that is a bit tight
but it can happen when multiple backends are sending notifies concurrently.
The net symptom therefore is that a sufficiently-long-surviving
listen-only backend starts to miss some fraction of NOTIFY traffic,
but only under heavy load.
The only function that updates RecentXmin is GetSnapshotData().
A brute-force fix would therefore be to take a snapshot before
processing incoming notify messages. But that would add cycles,
as well as contention for the ProcArrayLock. We can be smarter:
having taken the snapshot, let's use that to check for running
XIDs, and not call TransactionIdIsInProgress() at all. In this
way we reduce the number of ProcArrayLock acquisitions from one
per message to one per notify interrupt; that's the same under
light load but should be a benefit under heavy load. Light testing
says that this change is a wash performance-wise for normal loads.
I looked around for other callers of TransactionIdIsInProgress()
that might be at similar risk, and didn't find any; all of them
are inside transactions that presumably have already taken a
snapshot.
Problem report and diagnosis by Marko Tiikkaja, patch by me.
Back-patch to all supported branches, since it's been like this
since 9.0.
Discussion: https://postgr.es/m/20170926182935.14128.65278@wrigleys.postgresql.org
In commit fccebe421, we hacked get_actual_variable_range() to scan the
index with SnapshotDirty, so that if there are many uncommitted tuples
at the end of the index range, it wouldn't laboriously scan through all
of them looking for a live value to return. However, that didn't fix it
for the case of many recently-dead tuples at the end of the index;
SnapshotDirty recognizes those as committed dead and so we're back to
the same problem.
To improve the situation, invent a "SnapshotNonVacuumable" snapshot type
and use that instead. The reason this helps is that, if the snapshot
rejects a given index entry, we know that the indexscan will mark that
index entry as killed. This means the next get_actual_variable_range()
scan will proceed past that entry without visiting the heap, making the
scan a lot faster. We may end up accepting a recently-dead tuple as
being the estimated extremal value, but that doesn't seem much worse than
the compromise we made before to accept not-yet-committed extremal values.
The cost of the scan is still proportional to the number of dead index
entries at the end of the range, so in the interval after a mass delete
but before VACUUM's cleaned up the mess, it's still possible for
get_actual_variable_range() to take a noticeable amount of time, if you've
got enough such dead entries. But the constant factor is much much better
than before, since all we need to do with each index entry is test its
"killed" bit.
We chose to back-patch commit fccebe421 at the time, but I'm hesitant to
do so here, because this form of the problem seems to affect many fewer
people. Also, even when it happens, it's less bad than the case fixed
by commit fccebe421 because we don't get the contention effects from
expensive TransactionIdIsInProgress tests.
Dmitriy Sarafannikov, reviewed by Andrey Borodin
Discussion: https://postgr.es/m/05C72CF7-B5F6-4DB9-8A09-5AC897653113@yandex.ru
Peter Eisentraut
Bruce Momjian
Andrew Dunstan
Amit Kapila
Andres Freund
Peter Geoghegan
John Naylor
Tom Lane
Daniel Gustafsson
Michael Paquier
Noah Misch
Robert Haas
Fujii Masao
Heikki Linnakangas
Peter Eisentraut
Simon Riggs