Apr 2022

3 commits
Fix second race condition in 002_archiving.pl with archive_cleanup_command
18 Apr 2022Michael Paquiermaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #123 of 3 reviewers saw the final version

The commit extracts the checkpoint-ordering hunk from #12, lightly revises its comment, and leaves the history-file changes to a separate commit. Its Author credit identifies Tom Lane, whose original implementation appears in #7.

  • Andres Freundsaw the final versionLLM analysis, 85% sureContinuing the investigation of #7, Andres reported in #9 that an additional WAL switch made the tests pass; the checkpoint-ordering fix subsequently changed only in comments.
  • Thomas Munrosaw the final versionLLM analysis, 99% sureThomas tested #12 under Valgrind and explicitly reported the archive_cleanup_command checkpoint test passing in #13.
  • Michael Paquiersaw the final versioncommittercommitted the patch
Fix race in TAP test 002_archiving.pl when restoring history file
18 Apr 2022Michael Paquiermaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1REL_14_STABLE14.3REL_13_STABLE13.7+2 more
LLM analysis, 100% surepart of a posted patchsetmoderate changesfrom patchset #120 of 2 reviewers saw the final version

The commit takes the history-file wait, corrected standby description, and restoration-log assertion from #12, excluding its checkpoint-ordering fix. As proposed in #14, it replaces the explicit usleep loop with poll_query_until() querying pg_stat_file().

  • Andres Freundsaw an earlier versionLLM analysis, 84% sureWhile investigating the test with #7, Andres identified its misleading standby description and questioned its history-file behavior in #9, prompting the corrections and additional synchronization in #12.
  • Thomas Munrosaw an earlier versionLLM analysis, 99% sureThomas tested #12, found the missing usleep import, and reported a later promotion timeout in #13, after which Michael replaced the polling implementation.
Don't retry restore_command while reading ahead.
16 Apr 2022Thomas Munromaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surelanded as postedfrom patchset #10

The src/backend/access/transam/xlogprefetcher.c changes match #10: use <= for the read-ahead barrier and set no_readahead_until to the decoded queue tail when nonblocking reading runs out of data. Only comment wrapping differs in the code diff.

Credits across 3 commits

Author
Michael PaquierLLM1
Thomas MunroLLM1
Tom Lane1
Reviewer
Michael Paquiercommitter1
Reported by
Andres Freund1
Committer
Michael Paquier2
Thomas Munro1

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