Nov 2022

1 commit
Improve heuristics for compressing the KnownAssignedXids array.
29 Nov 2022Tom Lanemaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1REL_15_STABLE15.2REL_14_STABLE14.7REL_13_STABLE13.10+2 more
LLM analysis, 100% suremodified before commitmoderate changesfrom patchset #621 of 5 reviewers saw the final version

The commit retains the reason-based compression, transaction counter, idle maintenance, and compact-array check from #62, whose header credits Simon Riggs jointly with the sender. Tom Lane moved lock acquisition inside KnownAssignedXidsCompress after its checks, added another xlogrecovery.c idle call, moved the declaration to storage/procarray.h, and selected frequency 128 following #63; his edits and push are confirmed in #64 and #65.

  • Andres Freundsaw an earlier versionLLM analysis, 99% sureChallenged replay overhead in #6 and identified transaction-ID modulus sampling failures in #45, before the event-driven compression and locking revisions.
  • Andrey Borodinsaw an earlier versionLLM analysis, 99% sureReviewed and helped advance the offset implementation in #14, #21, #25, and #27, then raised workload-dependent compression costs in #32 before the approach was substantially revised.
  • Alexander Korotkovsaw an earlier versionLLM analysis, 100% sureExplained in #17 that the offset array could use ordinary 32-bit values instead of atomic wrappers, a change adopted in #18 before that implementation was superseded.
  • Kyotaro Horiguchisaw an earlier versionLLM analysis, 100% sureProposed storing next-entry indexes instead of distances in #28, which #36 adopted before the switch to compression heuristics.
  • Tom Lanesaw the final versioncommittercommitted the patch

Credits across 1 commit

Author
Simon RiggsLLM1
Reviewer
Tom Lanecommitter1
Committer
Tom Lane1

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