Oct - Nov 2021

4 commits
Add hardening to catch invalid TIDs in indexes.
5 Nov 2021Peter Geogheganmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #951 of 1 reviewer saw the final version

This commits the five-file invalid-index-TID hardening portion of #95, leaving pruneheap.c to a separate commit. Remaining edits add unlikely() annotations, simplify comments, narrow a variable's scope, and introduce an equivalent local offset variable in genam.c.

  • Andres Freundsaw the final versiontrailerAndres requested runtime errors in #86, reviewed the resulting patch and requested unlikely() annotations in #96, and accepted the follow-up in #98.
Add various assertions to heap pruning code.
5 Nov 2021Peter Geogheganmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #951 of 1 reviewer saw the final version

This extracts the pruneheap.c portion of #95, whose assertions originated in #90. The executable assertions are preserved; the differences clarify comments about redirect sources, redirect targets, and LP_DEAD transitions.

  • Andres Freundsaw the final versiontrailerAndres reviewed the pruning portion in #96, requested clearer redirect wording, and accepted Peter's revised explanation in #98.
Don't overlook indexes during parallel VACUUM.
2 Nov 2021Peter Geogheganmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1REL_14_STABLE14.1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #922 of 2 reviewers saw the final version

This commits the first patch of #92, including the vacuumlazy.c fix and parallel VACUUM regression test, while deferring the assertion patch. Differences are comment edits and test cleanup; the commit also credits Masahiko Sawada as co-author for the test's source material.

  • Kamigishi Reisaw the final versionLLM analysis, 99% sureKamigishi tested #84 against the saved database and reported that VACUUM now processed the previously omitted indexes in #85; subsequent changes retained the fix and added documentation and tests.
  • Andres Freundsaw the final versionLLM analysis, 99% sureAndres reviewed #84 and requested regression coverage for the distinction between parallel-worthy and smaller indexes in #86, which Peter added in #92.
Further harden nbtree posting split code.
27 Oct 2021Peter Geogheganmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1REL_14_STABLE14.1REL_13_STABLE13.5
LLM analysis, 96% surewritten from the discussion, no matching patch

The commit adds duplicate-TID checks in _bt_insertonpg() and _bt_binsrch_insert(), matching the failure analysis in #19 and the landing announcement in #22. No attachment contains these changes: #18 instead changes _bt_findinsertloc() and _bt_stepright() insertion-state handling.

Credits across 4 commits

Author
Peter Geoghegan3
Masahiko Sawada1
Reviewer
Andres Freund2
Reported by
Andrew Gierth1
Kamigishi Rei1
Committer
Peter Geoghegan4