Apr 2022

5 commits
Have VACUUM warn on relfrozenxid "in the future".
5 Apr 2022Peter Geogheganmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 99% surepart of a posted patchsetlight changesfrom patchset #1311 of 1 reviewer saw the final version

Lands the complete warning component of the four-patch series in #131; its src/backend/commands/vacuum.c changes match the attachment, with editorial changes to the commit message.

  • Andres Freundsaw the final versiontrailerAndres tested the diagnostic patch and advocated WARNING in #129, then reviewed the final warning component and said it looked good in #135.
vacuumlazy.c: Further consolidate resource allocation.
4 Apr 2022Peter Geogheganmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 99% surepart of a posted patchsetlight changesfrom patchset #131

Extracts the resource-allocation component of #131, moving dead_items allocation and cleanup to heap_vacuum_rel() in vacuumlazy.c; it retains the final progress-report calls that the attachment proposed deleting.

Generalize how VACUUM skips all-frozen pages.
3 Apr 2022Peter Geogheganmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 99% surepart of a posted patchsetlight changesfrom patchset #1311 of 1 reviewer saw the final version

Lands the range-skipping component of #131, including lazy_scan_skip() and skippedallvis in vacuumlazy.c; implementation differences are whitespace and diff alignment, while the commit explanation is revised.

  • Robert Haassaw the final versiontrailerRobert examined the skipping refactor in #100 and clarified its concurrency argument in #102, after which this component received only naming and explanatory edits.
Set relfrozenxid to oldest extant XID seen by VACUUM.
3 Apr 2022Peter Geogheganmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 99% surepart of a posted patchsetlight changesfrom patchset #1312 of 2 reviewers saw the final version

Lands the oldest-extant-XID component and isolation tests from #131, preserving the heapam.c tracking logic; the documentation tip is split into 05023a237c, with additional comment, naming, and declaration cleanup.

  • Andres Freundsaw the final versiontrailerAndres requested isolation coverage in #59, reviewed MultiXact handling in #83, and reproduced and diagnosed the late assertion failure in #114 and #119.
  • Robert Haassaw the final versiontrailerRobert reviewed the horizon-tracking contracts and explanations in #58, #100, and #105, with only light cleanup following his latest review.
Doc: Add relfrozenxid Tip to XID wraparound section.
2 Apr 2022Peter Geogheganmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 99% surepart of a posted patchsetlight changesfrom patchset #1311 of 1 reviewer saw the final version

Extracts and rewords the maintenance.sgml tip already included in the first patch of #131, documenting relfrozenxid/relminmxid reporting through VACUUM VERBOSE and autovacuum logs.

  • Robert Haassaw the final versionLLM analysis, 85% sureRobert explicitly reviewed the first patch's explanatory text and maintenance.sgml changes in #105, before its existing logging tip was separated and lightly reworded.

Feb 2022

3 commits
Consolidate VACUUM xid cutoff logic.
12 Feb 2022Peter Geogheganmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 99% surepart of a posted patchsetlight changesfrom patchset #39

Lands the cutoff-consolidation component of #39, making vacuum_set_xid_limits() return the aggressive-VACUUM decision and simplifying its callers; differences are explanatory comments rather than cutoff logic.

Add VACUUM instrumentation for scanned pages, relfrozenxid.
12 Feb 2022Peter Geogheganmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 99% surepart of a posted patchsetlight changesfrom patchset #390 of 1 reviewer saw the final version

Lands the complete instrumentation component of #39; changed lines in vacuumlazy.c, analyze.c, vacuum.c, and vacuum.h match after accounting for file ordering.

  • Andres Freundsaw an earlier versiontrailerAndres challenged removal of pin-related diagnostics and distinguished cleanup-lock failures from horizon-blocked tuples in #2 and #4, before substantial counter and reporting changes were implemented.
Simplify lazy_scan_heap's handling of scanned pages.
11 Feb 2022Peter Geogheganmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 99% surepart of a posted patchsetlight changesfrom patchset #390 of 2 reviewers saw the final version

Lands the scan-processing component of #39, including lazy_scan_noprune(), unified scanned_pages accounting, and the isolation-test adjustment; final edits primarily polish comments and reorganize equivalent recordfreespace branches.

  • Andres Freundsaw an earlier versiontrailerAndres reviewed locking, tuple accounting, helper contracts, and missing tests in #2 and #4, before the no-pruning FSM path and other local logic were revised.
  • Masahiko Sawadasaw an earlier versiontrailerMasahiko reviewed the first component in #10, proposing LVRelState storage for skipwithvm and questioning last-page scanning, before a later normal-tuple hastup fix changed truncation-state handling.

Credits across 8 commits

Author
Peter Geoghegan8
Reviewer
Andres Freund4
Robert Haas2
Masahiko Sawada1
Committer
Peter Geoghegan8