Jul 2024

2 commits
Count individual SQL commands in pg_restore's --transaction-size mode.
29 Jul 2024Tom Lanemaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1REL_17_STABLE17beta3
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #990 of 2 reviewers saw the final version

This takes exactly the third patch from #99, adding semicolon-based accounting in pg_backup_archiver.c. Applying that patch to the commit's parent produces the committed file byte-for-byte; the partial verdict reflects the series split.

  • Justin Pryzbysaw an earlier versionLLM analysis, 94% sureHe reported the transaction-batching memory regression in #89 and supplied a reproducible schema in #92, before the attached fixes existed.
  • Alexander Korotkovsaw an earlier versionLLM analysis, 99% sureDuring discussion of #94, he proposed the adopted semicolon-counting approach in #97; his explicit final-patch approval in #100 postdates the commit.
Reduce number of commands dumpTableSchema emits for binary upgrade.
29 Jul 2024Tom Lanemaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1REL_17_STABLE17beta3
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #990 of 2 reviewers saw the final version

This combines the first two patches from #99: ONLY cleanup and consolidation of catalog updates in dumpTableSchema. Applying both to the commit's parent reproduces pg_dump.c and its test exactly, leaving the transaction-accounting patch for a separate commit.

  • Justin Pryzbysaw an earlier versionLLM analysis, 95% sureHis regression report and wide-partition reproducer in #89 and #92 identified the failure this fix addresses, without a subsequent pre-commit review of the attached fixes.
  • Alexander Korotkovsaw an earlier versionLLM analysis, 96% sureHe independently confirmed the diagnosis in #95 and accepted the update-consolidation direction in #97, before it was expanded to dropped columns and inherited constraints.

Apr 2024

2 commits
Invent --transaction-size option for pg_restore.
1 Apr 2024Tom Lanemaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #772 of 8 reviewers saw the final version

This combines the fourth and sixth patches from #77, implementing --transaction-size and scaling pg_upgrade's batch size by parallel jobs. Applying those two patches to the commit's parent reproduces every affected file exactly.

  • Nathan Bossartsaw the final versionLLM analysis, 100% sureHe reviewed the earlier series in #71, reviewed the final code in #86, and accepted leaving transaction size non-configurable in #88.
  • Laurenz Albesaw the final versionLLM analysis, 100% sureHe tested the final series with fifteen million large objects and reported runtime, memory, disk usage, and a repeat run in #78 and #80.
  • Robins Tharakansaw an earlier versionLLM analysis, 96% sureHe tested all four patches and reported extensive dump/restore benchmarks in #65 and #67, before the parallel-job transaction-size adjustment was added.
  • Zhihong Yusaw an earlier versionLLM analysis, 98% sureHe identified redundant transaction-state tracking in the original batching implementation in #31, which was removed in #33 before the later general rewrite.
  • Bruce Momjiansaw an earlier versionLLM analysis, 90% sureHis proposal in #34 to forward arbitrary restore options was implemented in #47, although that control interface was subsequently replaced.
  • Jacob Championsaw an earlier versionLLM analysis, 98% sureHe reviewed the pg_upgrade integration and identified the risks of arbitrary dump/restore options in #50 and #52, prompting their eventual removal.
  • Andres Freundsaw an earlier versionLLM analysis, 84% sureHe supported replacing the existing user-controlled strategy with automatic behavior in #57, a direction adopted in #59 and retained by the final implementation.
  • Magnus Hagandersaw an earlier versionLLM analysis, 88% sureHe diagnosed excessive restore transactions and proposed grouping restore commands in #6, redirecting the original limit workaround toward transaction batching.
Rearrange pg_dump's handling of large objects for better efficiency.
1 Apr 2024Tom Lanemaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #772 of 4 reviewers saw the final version

This combines the first, second, third, and fifth patches from #77: buffer preliminaries, grouped large-object metadata/data, SECTION_DATA placement, and shared ACL storage. Applying that combination to the commit's parent reproduces every affected file exactly.

  • Nathan Bossartsaw the final versionLLM analysis, 100% sureHe reviewed the grouping design in #71, suggested named constants adopted in #76, and completed another code review in #86.
  • Laurenz Albesaw the final versionLLM analysis, 100% sureHis final-series testing in #78 demonstrated substantial memory and dump-size reductions without noticeable memory leaks.
  • Robins Tharakansaw an earlier versionLLM analysis, 99% sureHe benchmarked empty and populated large objects and parallel scaling in #65 and #67, before shared ACL storage was added.
  • Michael Bancksaw an earlier versionLLM analysis, 83% sureHe reported promising pg_dump benchmarks in #81, but did not identify the tested patch version or retain its measurements.

Credits across 4 commits

Author
Tom LaneLLM4
Committer
Tom Lane4

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