This impacts the statistics retrieved in transactions for the following
views when updating the value of stats_fetch_consistency, leading to
behaviors contrary to what is documented since 605994651b6a as an update
of this parameter should discard all statistics snapshot data:
- pg_stat_archiver
- pg_stat_bgwriter
- pg_stat_checkpointer
- pg_stat_io
- pg_stat_slru
- pg_stat_wal
For example, updating stats_fetch_consistency from "snapshot" to "cache"
in a transaction did not re-fetch any fresh data, using data cached from
the time when "snapshot" was in use.
Author: Shinya Kato
Discussion: https://postgr.es/m/d77fc5190d4dbe1738d77231488e768b@oss.nttdata.com
Backpatch-through: 15
Previously stats in the startup process would only get reported during
shutdown of the startup process. It has been that way for a long time, but
became a lot more noticeable with the new pg_stat_io view, which separates out
IO done by different backend types...
While replaying after every XLOG_RUNNING_XACTS isn't the prettiest approach,
it has the advantage of being quite easy. Given that we're well past feature
freeze...
It's not a problem that we don't report stats more frequently with
wal_level=minimal, in that case stats can't be read before the stats process
has shut down.
Besides the above, this commit also changes pgstat_report_stat() to acquire
the timestamp with GetCurrentTimestamp() instead of
GetCurrentTransactionStopTimestamp().
Thanks to Melih Mutlu, Kyotaro Horiguchi for prototypes of other approaches to
solving this issue.
Reported-by: Fujii Masao <masao.fujii@oss.nttdata.com>
Discussion: https://postgr.es/m/5315aedc-fbca-1556-c5de-dc2e00b23a14@oss.nttdata.com
Run pgindent, pgperltidy, and reformat-dat-files.
This set of diffs is a bit larger than typical. We've updated to
pg_bsd_indent 2.1.2, which properly indents variable declarations that
have multi-line initialization expressions (the continuation lines are
now indented one tab stop). We've also updated to perltidy version
20230309 and changed some of its settings, which reduces its desire to
add whitespace to lines to make assignments etc. line up. Going
forward, that should make for fewer random-seeming changes to existing
code.
Discussion: https://postgr.es/m/20230428092545.qfb3y5wcu4cm75ur@alvherre.pgsql
An update of the GUC stats_fetch_consistency in a transaction would be
able to trigger an assertion when doing cache->snapshot. In this case,
when retrieving a pgstat entry after the switch, a new snapshot would be
rebuilt, confusing pgstat_build_snapshot() because a snapshot is already
cached with an unexpected mode ("cache").
In order to fix this problem, this commit adds a flag to force a
snapshot clear each time this GUC is changed. Some tests are added to
check, while on it.
Some optimizations in avoiding the snapshot clear should be possible
depending on what is cached and the current GUC value, I guess, but this
solution is simple, and ensures that the state of the cache is updated
each time a new pgstat entry is fetched, hence being consistent with the
level wanted by the client that has set the GUC.
Note that cache->none and snapshot->none would not cause issues, as
fetching a pgstat entry would be retrieved from shared memory on the
second attempt, however a snapshot would still be cached. Similarly,
none->snapshot and none->cache would build a new snapshot on the second
fetch attempt. Finally, snapshot->cache would cache a new snapshot on
the second attempt.
Reported-by: Alexander Lakhin
Author: Kyotaro Horiguchi
Discussion: https://postgr.es/m/17804-2a118cd046f2d0e5@postgresql.org
backpatch-through: 15
This structure included only PgStat_FunctionCounts, and removing it
facilitates some upcoming refactoring for pgstatfuncs.c to use more
macros rather that mostly-duplicated functions.
Author: Bertrand Drouvot
Reviewed-by: Nathan Bossart
Discussion: https://postgr.es/m/11d531fe-52fc-c6ea-7e8e-62f1b6ec626e@gmail.com
This commit adds the infrastructure for more detailed IO statistics. The calls
to actually count IOs, a system view to access the new statistics,
documentation and tests will be added in subsequent commits, to make review
easier.
While we already had some IO statistics, e.g. in pg_stat_bgwriter and
pg_stat_database, they did not provide sufficient detail to understand what
the main sources of IO are, or whether configuration changes could avoid
IO. E.g., pg_stat_bgwriter.buffers_backend does contain the number of buffers
written out by a backend, but as that includes extending relations (always
done by backends) and writes triggered by the use of buffer access strategies,
it cannot easily be used to tune background writer or checkpointer. Similarly,
pg_stat_database.blks_read cannot easily be used to tune shared_buffers /
compute a cache hit ratio, as the use of buffer access strategies will often
prevent a large fraction of the read blocks to end up in shared_buffers.
The new IO statistics count IO operations (evict, extend, fsync, read, reuse,
and write), and are aggregated for each combination of backend type (backend,
autovacuum worker, bgwriter, etc), target object of the IO (relations, temp
relations) and context of the IO (normal, vacuum, bulkread, bulkwrite).
What is tracked in this series of patches, is sufficient to perform the
aforementioned analyses. Further details, e.g. tracking the number of buffer
hits, would make that even easier, but was left out for now, to keep the scope
of the already large patchset manageable.
Bumps PGSTAT_FILE_FORMAT_ID.
Author: Melanie Plageman <melanieplageman@gmail.com>
Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Justin Pryzby <pryzby@telsasoft.com>
Reviewed-by: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Discussion: https://postgr.es/m/20200124195226.lth52iydq2n2uilq@alap3.anarazel.de
Previously PgStat_StatReplSlotEntry contained the slotname, which was mainly
used when writing out the stats during shutdown, to identify the slot in the
serialized data (at runtime the index in ReplicationSlotCtl->replication_slots
is used, but that can change during a restart). Unfortunately the slotname was
overwritten when the slot's stats were reset.
That turned out to only cause "real" problems if the slot was active during
the reset, triggering an assertion failure at the next
pgstat_report_replslot(). In other paths the stats were re-initialized during
pgstat_acquire_replslot().
Fix this by removing slotname from PgStat_StatReplSlotEntry. Instead we can
get the slot's name from the slot itself. Besides fixing a bug, this also is
architecturally cleaner (a name is not really statistics). This is safe
because stats, for a slot removed while shut down, will not be restored at
startup.
In 15 the slotname is not removed, but renamed, to avoid changing the stats
format. In master, bump PGSTAT_FILE_FORMAT_ID.
This commit does not contain a test for the fix. I think this can only be
tested by a tap test starting pg_recvlogical in the background and checking
pg_recvlogical's output. That type of test is notoriously hard to be reliable,
so committing it shortly before the release is wrapped seems like a bad idea.
Reported-by: Jaime Casanova <jcasanov@systemguards.com.ec>
Author: Andres Freund <andres@anarazel.de>
Reviewed-by: Masahiko Sawada <sawada.mshk@gmail.com>
Reviewed-by: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Discussion: https://postgr.es/m/YxfagaTXUNa9ggLb@ahch-to
Backpatch: 15-, where the bug was introduced in 5891c7a8ed8f
So far they were created below CacheMemoryContext. However, that's not
guaranteed to exist in all situations, leading to memory contexts created as
top-level contexts. There isn't actually a good reason anymore to create them
below CacheMemoryContext, so just creating them below TopMemoryContext seems
the best approach.
Reported-by: Reid Thompson <reid.thompson@crunchydata.com>
Reviewed-by: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Reviewed-by: Andres Freund <andres@anarazel.de>
Author: "Drouvot, Bertrand" <bdrouvot@amazon.com>
Discussion: https://postgr.es/m/b948b729-42fe-f88c-2f4a-0e65d84c049b@amazon.com
Backpatch: 15-
Somewhere during the development of the patch acquiring a lock during read
access to variable-numbered stats got lost. The missing lock acquisition won't
cause corruption, but can lead to reading torn values when accessing
stats. Add the missing lock acquisitions.
Reported-by: Greg Stark <stark@mit.edu>
Reviewed-by: "Drouvot, Bertrand" <bdrouvot@amazon.com>
Reviewed-by: Andres Freund <andres@anarazel.de>
Author: Kyotaro Horiguchi <horikyota.ntt@gmail.com>
Discussion: https://postgr.es/m/CAM-w4HMYkM_DkYhWtUGV+qE_rrBxKOzOF0+5faozxO3vXrc9wA@mail.gmail.com
Backpatch: 15-
Previously the timer was enabled whenever there were any pending stats after
executing a statement, just to then be disabled again when not idle
anymore. That lead to an increase in GetCurrentTimestamp() calls from within
timeout.c compared to 14.
To avoid that increase, leave the timer enabled until stats are reported,
rather than until idle. The timer is only disabled once the pending stats have
been reported.
For me this fixes the increase in GetCurrentTimestamp() calls, there now are
fewer calls in 15 than in 14, in the previously slowed down workload.
While at it, also update assertion in pgstat_report_stat() to be more precise.
Author: Andres Freund <andres@anarazel.de>
Discussion: https://postgr.es/m/20220616233130.rparivafipt6doj3@alap3.anarazel.de
Backpatch: 15-
Somewhat embarrassing oversight in 98f897339b0. Does not have a functional
impact, but is unnecessarily confusing.
Reported-By: Michael Paquier <michael@paquier.xyz>
Discussion: https://postgr.es/m/Yo2351qVYqd/bJws@paquier.xyz
Previously we didn't, which lead to an assertion failure when resetting
partially loaded statistics. This was encountered on the buildfarm, for
as-of-yet unknown reasons.
Ttighten up a validity check when reading the stats file, verifying 'E'
signals the end of the file (rather than just stopping reading). That's then
used in a test appending to the stats file that crashed before the fix in
pgstat_drop_all_entries().
Reported by buildfarm animals mylodon and kestrel, via Tom Lane.
Discussion: https://postgr.es/m/1656446.1650043715@sss.pgh.pa.us
When not loading stats at startup (i.e. pgstat_discard_stats() getting
called), reset timestamps of fixed numbered stats would be left at
0. Oversight in 5891c7a8ed8.
Instead use pgstat_reset_after_failure() and add tests verifying that
fixed-numbered reset timestamps are set appropriately.
Reported-By: "David G. Johnston" <david.g.johnston@gmail.com>
Discussion: https://postgr.es/m/CAKFQuwamFuaQHKdhcMt4Gbw5+Hca2UE741B8gOOXoA=TtAd2Yw@mail.gmail.com
In the stats collector days it was hard to write tests for the stats system,
because fundamentally delivery of stats messages over UDP was not
synchronous (nor guaranteed). Now we easily can force pending stats updates to
be flushed synchronously.
This moves stats.sql into a parallel group, there isn't a reason for it to run
in isolation anymore. And it may shake out some bugs.
Bumps catversion.
Author: Andres Freund <andres@anarazel.de>
Discussion: https://postgr.es/m/20220303021600.hs34ghqcw6zcokdh@alap3.anarazel.de
