Jun 2025

1 commit
Test that vacuum removes tuples older than OldestXmin
24 Jun 2025Melanie Plagemanmaster18beta2REL_19_STABLE18beta2REL_18_STABLE18beta2REL_17_STABLE17.6
LLM analysis, 100% surelanded as postedfrom patchset #901 of 2 reviewers saw the final version

The commit diff exactly matches #90, including the Meson registration and all 278 lines of src/test/recovery/t/048_vacuum_horizon_floor.pl. Melanie confirms pushing the test to master and PostgreSQL 17 in #92.

  • John Naylorsaw the final versiontrailerJohn reported cross-platform test results in #85 and #87, proposed the adopted row-count and comment simplifications in #89, and approved the committed version in #91.
  • Peter Geoghegansaw an earlier versiontrailerPeter endorsed the original reproducer in #2, tested failure without the fix and success with it in #36, and suggested diagnostics in #39, before the restored test gained substantial setup and scan-control changes.

Jul - Aug 2024

6 commits
Lower minimum maintenance_work_mem to 64kB
10 Aug 2024John Naylormaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1REL_17_STABLE17rc1Also in: Lowering the minimum value for maintenance_work_mem (not this thread)
LLM analysis, 99% surepart of a posted patchsetmoderate changesfrom patchset #701 of 2 reviewers saw the final version

Retains the 64kB maintenance_work_mem change from #70, drops its radixtree.h changes, and adds matching autovacuum_work_mem handling, sample configuration updates, and a shared-memory warning. John explains in #78 that slab contexts already allocate lazily enough, making the radix-tree changes unnecessary.

  • Andres Freundsaw the final versionLLM analysis, 98% sureProposed lowering vacuum's memory minimum in #54, an adopted design explicitly credited in bbf668d66f.
  • Melanie Plagemansaw an earlier versionLLM analysis, 100% sureReported successful testing and a runtime reduction with #70 in #72, before the radix-tree changes were dropped and autovacuum handling was added.
Add accidentally omitted test to meson build file
29 Jul 2024Melanie PlagemanREL_16_STABLE16.4
LLM analysis, 99% surewritten from the discussion, no matching patch

Adds the omitted test entry to PostgreSQL 16's src/test/recovery/meson.build, exactly the branch-specific oversight Melanie identifies in #71. No attached follow-up patch implements that correction.

Use DELETE instead of UPDATE to speed up vacuum test
29 Jul 2024Melanie PlagemanREL_16_STABLE16.4REL_15_STABLE15.8REL_14_STABLE14.13
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 1 reviewer saw the final version

The single executable change replaces UPDATE with DELETE in the existing recovery test. Masahiko Sawada proposes it in #64, and Melanie reports accepting and benchmarking it in #71; no corresponding follow-up attachment appears.

  • Masahiko Sawadasaw an earlier versionLLM analysis, 97% sureAnalyzed the landed test and proposed the adopted DELETE optimization in #64, with further confirmation in #75; subsequent restoration revisions substantially changed the test setup.
Revert "Test that vacuum removes tuples older than OldestXmin"
22 Jul 2024Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1REL_17_STABLE17beta3reverts 1 commit here
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 1 reviewer saw the final version

Reverts the test on master and PostgreSQL 17 following Tom Lane's buildfarm failure report in #41 and Melanie's diagnosis and proposed revert in #58. Melanie confirms the revert in #60; no revert patch was attached.

  • Tom Lanesaw an earlier versionLLM analysis, 98% sureReproduced the landed test's timeout and reported a 908-second run in #41, prompting the revert before the later test restoration and stability changes.
Test that vacuum removes tuples older than OldestXmin
19 Jul 2024Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1REL_17_STABLE17beta3REL_16_STABLE16.4REL_15_STABLE15.8REL_14_STABLE14.13Also in: BUG #17257: (auto)vacuum hangs within lazy_scan_prune() (part of a posted patchset)
LLM analysis, 98% surepart of a posted patchsetlight changesfrom patchset #372 of 2 reviewers saw the final version

Lands the test member of the two-patch series in #37, with platform-dependent row counts, VERBOSE logging, and comment/formatting changes in src/test/recovery/t/043_vacuum_horizon_floor.pl. The later restoration grouped into this candidate matches #90 exactly, but the canonical commit derives from #37.

  • Peter Geoghegansaw the final versiontrailerVerified failure without the fix and success with it in #36, requested explanatory comments, and suggested VERBOSE diagnostics in #39; only light changes followed before the canonical landing.
  • John Naylorsaw the final versiontrailerHis credit belongs to the grouped restoration: he tested platform behavior in #87, proposed simplifications in #89, and approved the final version in #91.
Ensure vacuum removes all visibly dead tuples older than OldestXmin
19 Jul 2024Melanie Plagemanmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1REL_17_STABLE17beta3Also in: BUG #17257: (auto)vacuum hangs within lazy_scan_prune() (not this thread)
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #375 of 5 reviewers saw the final version

The diff exactly matches the production-fix member of #37 in src/backend/access/heap/pruneheap.c and src/backend/access/heap/vacuumlazy.c, excluding the series' test patch. Attribution disagrees with thread #71035: its claimed source #689666 dates to July 8, 2024, whereas this thread supplies the later, exactly matching patchset on July 17, 2024.

  • Peter Geoghegansaw the final versiontrailerEndorsed the approach in #2 and tested the fix successfully in #36, after which its implementation remained unchanged.
  • Robert Haassaw the final versiontrailerChallenged the two-horizon design in #8, explored alternatives, and accepted Melanie's approach in #33, with only explanatory changes to the fix afterward.
  • Andres Freundsaw the final versiontrailerAnalyzed the alternatives in #23 and explained in #30 and #32 why preventing backward movement or modifying global visibility objects was unattractive.
  • Heikki Linnakangassaw the final versiontrailerSupported the proposed fix and requested permanent regression coverage in #4, before comment-only changes to the production fix.
  • Noah Mischsaw the final versiontrailerReviewed the horizon consistency problem in #20 and clarified the longer-term unification concern in #22, while the committed fix retained its original logic.

Credits across 7 commits

Author
Melanie Plageman3
John NaylorLLM1
Reviewer
Peter Geoghegan3
Andres Freund1
Heikki Linnakangas1
John Naylor1
Noah Misch1
Robert Haas1
Reported by
Tom Lane1
Committer
Melanie Plageman6
John Naylor1

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