Mar - Apr 2020

7 commits
Further improve stability fix for partition_aggregate test.
5 Apr 2020Tom Lanemaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1
LLM analysis, 99% surewritten from the discussion, no matching patch

This follow-up changes the upper-level pagg_tab_ml partition boundaries and expected plans after another buildfarm failure; those changes are absent from #156. It extends the unequal-partition solution discussed in #155 and landed in 7cb0a423f9.

Improve stability fix for partition_aggregate test.
2 Apr 2020Tom Lanemaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1
LLM analysis, 100% surelanded as postedfrom patchset #156

Applying #156 to the commit's parent produces byte-identical partition_aggregate.sql and partition_aggregate.out files. Tom confirms pushing the patch in #158.

Attempt to stabilize partitionwise_aggregate test
2 Apr 2020David Rowleymaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1
LLM analysis, 100% surewritten from the discussion, no matching patch

Disables autovacuum on partition_aggregate test tables and reverts that test's earlier VACUUM ANALYZE change after the failures discussed in #152 and #153. No attachment contains this fix; #155 explicitly critiques the already-committed workaround.

Fix race condition in statext_store().
31 Mar 2020Tom Lanemaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1+1 more
LLM analysis, 96% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

Moves table_open(StatisticExtDataRelationId, RowExclusiveLock) before the tuple lookup in statext_store(), implementing Dean's precise fix proposed in #148. The commit credits Dean, while Tom undertakes the implementation in #149 and reports it pushed in #150.

  • Tom Lanesaw the final versioncommittercommitted the patch
Attempt to fix unstable regression tests, take 2
30 Mar 2020David Rowleymaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1
LLM analysis, 100% surewritten from the discussion, no matching patch

Changes additional ANALYZE statements to VACUUM (ANALYZE) for ndistinct, functional_dependencies, and mcv_lists_bool in stats_ext. This is the unattached follow-up investigation described in #145 and #446456, not the telemetry patch #139.

Attempt to fix unstable regression tests
29 Mar 2020David Rowleymaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1
LLM analysis, 99% surewritten from the discussion, no matching patch

Replaces ANALYZE with VACUUM (ANALYZE) for pagg_tab_ml and mcv_lists without adding telemetry. David explicitly says in #140 that he chose this fix instead of the diagnostic attachment #139.

Trigger autovacuum based on number of INSERTs
28 Mar 2020David Rowleymaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1311 of 6 reviewers saw the final version

The code matches David's posted revision #131, with expanded maintenance documentation, a pgstat.c explanatory comment, and a catalog-version bump. The commit credits Laurenz and Darafei as authors; David forwarded and finalized Laurenz's implementation.

  • Alvaro Herrerasaw an earlier versiontrailerAlvaro critiqued the original trigger in #2 and questioned the scale-factor limit in #125, before the explicit disabling logic was added.
  • Masahiko Sawadasaw an earlier versiontrailerMasahiko reviewed vacuum operations, identified the missing multixact freeze-age adjustment in #49, and discussed GIN cleanup in #120, before the final default and disabling changes.
  • Chris Traverssaw an earlier versiontrailerChris explicitly read the original patch in #22 and examined index cleanup and an alternative freeze worker in #25, before the implementation was substantially rewritten.
  • Andres Freundsaw an earlier versiontrailerAndres challenged freezing and index-I/O costs in #81 and #112, then requested explicit disabling in #127, which was implemented in the later unreviewed attachment.
  • Justin Pryzbysaw an earlier versiontrailerJustin tested the patch in #70, found freezing defects in #87, and reviewed scale limits in #124, before the final disabling logic and threshold change.
  • David Rowleysaw the final versioncommittercommitted the patch

Credits across 7 commits

Author
Darafei "Komяpa" Praliaskouski1
Dean RasheedLLM1
Laurenz Albe1
Tom LaneLLM1
Reviewer
Alvaro Herrera1
Andres Freund1
Chris Travers1
David Rowleycommitter1
Justin Pryzby1
Masahiko Sawada1
Tom Lanecommitter1
Committer
David Rowley4
Tom Lane3

N LLM means N of the credits come from LLM analysis, not from commit trailers.