Oct 2020

2 commits
Try to unbreak 021_row_visibility.pl on mingw.
6 Oct 2020Andres Freundmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

Adds Config and MSYS CRLF normalization to 021_row_visibility.pl exactly along the inline proposal in #114, which Andrew tested successfully in #118; no attachment contains this fix.

  • Andres Freundsaw the final versioncommittercommitted the patch
Fix and test snapshot behavior on standby.
1 Oct 2020Andres Freundmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surelanded as postedfrom patchset #111

The complete backend fix and recovery test in #111 have the same stable patch ID as the commit. Its Author trailers credit both sender Andres Freund and Ian Barwick, whose initial test and fix appear in #104 and #105.

Aug 2020

10 commits
snapshot scalability: cache snapshots using a xact completion counter.
18 Aug 2020Andres Freundmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #691 of 3 reviewers saw the final version

Lands the snapshot-cache component of #69, including xactCompletionCount and GetSnapshotDataReuse(); the changed code matches, with a comment typo corrected.

  • Robert Haassaw the final versiontrailerReviewed the counter definition, helper naming and snapshot-reuse explanation in #35, after which this component received light cleanup.
  • Thomas Munrosaw an earlier versiontrailerQuestioned unlocked counter reads, memory ordering and 32-bit atomicity in #7, before caching was changed to hold ProcArrayLock.
  • David Rowleysaw an earlier versiontrailerReported performance testing of the development series in #20, before the subsequently posted locked-cache revision.
Fix use of wrong index in ComputeXidHorizons().
16 Aug 2020Andres Freundmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surewritten from the discussion, no matching patch

Exactly implements Andres's inline correction in #78, replacing other_xids[pgprocno] with other_xids[index] in ComputeXidHorizons(); #75 supplies the reproducer, not the committed fix.

Make vacuum a bit more verbose to debug BF failure.
16 Aug 2020Andres Freundmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surewritten from the discussion, no matching patch

Adds temporary diagnostics in heapam.c and vacuumlazy.c following the buildfarm failure and instrumentation plan in #70 and #72; no attached patch contains these changes.

snapshot scalability: Introduce dense array of in-progress xids.
14 Aug 2020Andres Freundmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #690 of 3 reviewers saw the final version

Lands the dense in-progress XID array component of #69, preserving its PGPROC mirroring and dual-lock protocol. Remaining differences are comments, documentation and rebase adjustments.

  • Robert Haassaw an earlier versiontrailerReviewed array density, locking, mirrored fields and allocation arithmetic in #29 and #39, before GetSnapshotData()'s workspace-based loop was simplified.
  • Thomas Munrosaw an earlier versiontrailerQuestioned release of both locks on error paths in #7, before subsequent snapshot-loop and locking cleanup.
  • David Rowleysaw an earlier versiontrailerReviewed the original dense-XID patch's declarations and comments in #6 and benchmarked the series in #20, before later loop revisions.
snapshot scalability: Move PGXACT->vacuumFlags to ProcGlobal->vacuumFlags.
14 Aug 2020Andres Freundmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surepart of a posted patchsetmoderate changesfrom patchset #690 of 3 reviewers saw the final version

Lands the vacuumFlags component of #69, moving mirrored flags into PGPROC and ProcGlobal. Unlike the attachment, the commit retains ComputeXidHorizons()'s early continue when neither xmin nor xid is valid.

  • Robert Haassaw an earlier versiontrailerReviewed vacuumFlags naming and unlocked access through PGPROC in #35, before the committed horizon-guard adjustment.
  • Thomas Munrono review foundtrailerThe thread shows his reviews of related caching and XID machinery in #7 and #64, but no identifiable review of the vacuumFlags component.
  • David Rowleysaw an earlier versiontrailerBenchmarked the development series in #20, with no later review covering the committed horizon-guard adjustment.
snapshot scalability: Move subxact info to ProcGlobal, remove PGXACT.
14 Aug 2020Andres Freundmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surepart of a posted patchsetmoderate changesfrom patchset #690 of 3 reviewers saw the final version

Lands the subtransaction-state migration and PGXACT removal from #69. In twophase.c, the committed GXactLoadSubxactData() stores nsubxacts instead of incorrectly assigning PGPROC_MAX_CACHED_SUBXIDS.

  • Robert Haassaw an earlier versiontrailerReviewed subtransaction-state naming, mirrored values and patch boundaries in #35, before the final two-phase cached-count correction.
  • Thomas Munrono review foundtrailerHis identifiable reviews concern caching, XID arithmetic and shared locking in #7 and #64, without a specific review of the subtransaction-state component.
  • David Rowleysaw an earlier versiontrailerReported whole-series performance testing in #20, but no subsequent review of the final subtransaction representation and two-phase count correction is shown.
snapshot scalability: Move PGXACT->xmin back to PGPROC.
13 Aug 2020Andres Freundmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #692 of 3 reviewers saw the final version

Lands the xmin relocation component of #69; the substantive field and access changes match, with an additional MyPgXact-to-MyProc correction in access/transam/README.

  • Robert Haassaw the final versiontrailerExplicitly reviewed the xmin move in #29, after which this component changed only lightly.
  • Thomas Munrono review foundtrailerHis detailed comments in #7 address caching, horizon arithmetic and array locks, without an identifiable review of this xmin relocation.
  • David Rowleysaw the final versiontrailerIncluded the xmin relocation in his stated review of the first nine patches in #6, and its subsequent changes were light.
snapshot scalability: Don't compute global horizons while building snapshots.
12 Aug 2020Andres Freundmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surepart of a posted patchsetmoderate changesfrom patchset #690 of 3 reviewers saw the final version

Lands the deferred global-horizon component of #69. Beyond comment and signature cleanup, the commit rewrites HeapTupleSatisfiesNonVacuumable() to use HeapTupleSatisfiesVacuumHorizon() and GlobalVisTestIsRemovableXid(), a missing conversion in the attachment.

  • Robert Haassaw an earlier versiontrailerReviewed horizon structures, pruning logic and wraparound assertions in #27, #32 and #33, before later horizon-consistency and non-vacuumable-tuple fixes.
  • Thomas Munrosaw an earlier versiontrailerReviewed horizon arithmetic in #7 and personally flagged the vacuum failure acknowledged in #58, before the resulting correctness fixes and final tuple-check conversion.
  • David Rowleysaw an earlier versiontrailerReviewed the original invisibility abstraction in #6 and benchmarked it in #20, before substantial old-snapshot and horizon-consistency changes.
Track latest completed xid as a FullTransactionId.
12 Aug 2020Andres Freundmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 99% surepart of a posted patchsetlight changesfrom patchset #581 of 3 reviewers saw the final version

Lands the extracted FullTransactionId prerequisite from #58. The commit drops Full from variable names, renames FullXidViaRelative() to FullXidRelativeTo(), and makes minor expression and comment adjustments.

  • Thomas Munrosaw the final versiontrailerReviewed the original locking and conversion hazards in #7 and the final range assertion and branch-free conversion in #64.
  • Robert Haassaw an earlier versiontrailerReviewed the embedded XID-tracking changes and range assertion in #27 and #32, before recovery-specific maintenance was separated and refined.
  • David Rowleysaw an earlier versiontrailerReviewed the initial horizon and FullTransactionId machinery in #6, before latest-completed-XID tracking was widened and extracted.
Rename VariableCacheData.nextFullXid to nextXid.
11 Aug 2020Andres Freundmaster14beta1REL_19_STABLE14beta1REL_18_STABLE14beta1REL_17_STABLE14beta1REL_16_STABLE14beta1REL_15_STABLE14beta1REL_14_STABLE14beta1
LLM analysis, 100% surewritten from the discussion, no matching patch

Implements Andres's proposal in #58 to remove redundant Full naming, renaming nextFullXid and its references throughout transaction, checkpoint and utility code; no attached patch implements that rename.

Apr 2020

1 commit
snapshot scalability: Move delayChkpt from PGXACT to PGPROC.
8 Apr 2020Andres Freundmaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #242 of 3 reviewers saw the final version

Lands only the delayChkpt relocation component of #24; that individual attachment and the commit have identical stable patch IDs.

  • Robert Haassaw the final versiontrailerExplicitly assessed the delayChkpt component while reviewing the series in #27, and its code landed unchanged.
  • Thomas Munrono review foundtrailerNo identifiable delayChkpt review appears in the thread; his substantive review in #7 concerns other components.
  • David Rowleysaw the final versiontrailerQuestioned delayChkpt's placement and padding in PGPROC in #6, and the relocation subsequently changed only lightly.

Credits across 13 commits

Author
Andres Freund11
Andrew DunstanLLM1
Ian Lawrence Barwick1
Reviewer
David Rowley8
Robert Haas8
Thomas Munro8
Andres Freundcommitter1
Reported by
Ian Lawrence Barwick1
Committer
Andres Freund13

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