Sep 2024

2 commits
Ensure standby promotion point in 043_wal_replay_wait.pl
17 Sep 2024Alexander Korotkovmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #1151 of 1 reviewer saw the final version

Takes the promotion-point test fix from #115, including pg_switch_wal() and wait_for_catchup(). The two wording hunks bundled into that attachment instead landed in 85b98b8d5a.

  • Alexander Lakhinsaw the final versionLLM analysis, 100% sureReported six failures in ten runs of #113, then confirmed 100 successful runs of #115 in #116 before the commit.
Minor cleanup related to pg_wal_replay_wait() procedure
17 Sep 2024Alexander Korotkovmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #1151 of 1 reviewer saw the final version

Combines the cleanup patch in #115 with its second attachment's wording corrections in func.sgml and xlog.c, leaving the promotion synchronization fix separate. Remaining differences are test formatting.

  • Alexander Lakhinsaw the final versiontrailerIdentified wording problems in #112 and #114 and tested the revised series in #116.

Aug 2024

3 commits
Adjust pg_wal_replay_wait() procedure behavior on promoted standby
10 Aug 2024Alexander Korotkovmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #1100 of 1 reviewer saw the final version

Exactly matches the promotion-behavior patch in #110, rechecking replay position after recovery ends and adding its documentation and regression test. The other two patches in that submission are outside this commit.

  • Kevin Hale Boyessaw an earlier versionLLM analysis, 98% sureQuestioned recovery-check ordering in #102, prompting the revised promotion logic explained in #104.
pg_wal_replay_wait(): Fix typo in the doc
4 Aug 2024Alexander Korotkovmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

Corrects 'changes made of primary' to 'changes made on primary' in func.sgml, exactly the typo reported in #102 and acknowledged in #103; no attached fix preceded it.

  • Kevin Hale Boyessaw the final versionLLM analysis, 100% sureSupplied the exact documentation correction in #102, and that correction remains applicable through the final patchset.
Implement pg_wal_replay_wait() stored procedure
2 Aug 2024Alexander Korotkovmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1010 of 9 reviewers saw the final version

Matches #101 apart from rebased context in src/backend/storage/ipc/ipci.c, assigning the procedure OID in src/include/catalog/pg_proc.dat, and adding the catalog-version bump; implementation, documentation, and tests otherwise match. The patch header and commit credit Kartyshov Ivan and sender Alexander Korotkov as co-authors.

  • Michael Paquiersaw an earlier versiontrailerExamined snapshot implications and per-record replay wakeups in #3, before the procedure and waiter-management redesigns.
  • Peter Eisentrautsaw an earlier versiontrailerReviewed client and pooler usability in #4 and proposed interval timeouts in #44, before substantial API and implementation changes.
  • Dilip Kumarsaw an earlier versiontrailerRequested clearer separation of the alternatives and missing syntax documentation in #23, before the waiter-management and procedure redesigns.
  • Amit Kapilasaw an earlier versiontrailerReviewed syntax, implicit-transaction usability, and the procedure proposal in #26, #34, and #39, before subsequent implementation and snapshot changes.
  • Alexander Lakhinsaw an earlier versiontrailerOff-list corrections were incorporated in #29, #57, and #63, before the pairing-heap, locking, and snapshot revisions.
  • Bharath Rupireddysaw an earlier versiontrailerReviewed validation, replay overhead, timeout handling, and error reporting in #33 and #66, before substantial waiter-management and snapshot changes.
  • Euler Taveirasaw an earlier versiontrailerProposed the adopted procedure name and bigint-millisecond timeout in #59 and #62, before the locking and snapshot redesigns.
  • Heikki Linnakangassaw an earlier versiontrailerReviewed the initially committed implementation and its follow-ups in #90 and #92, exposing waiter-cleanup races and multi-command restrictions that drove the revert and relanding changes.
  • Kyotaro Horiguchisaw an earlier versiontrailerReported a logical-walsender assertion failure and proposed latch, timeout, documentation, and test improvements in #96, before the final snapshot-check revision in #101.

Apr 2024

7 commits
Revert: Implement pg_wal_replay_wait() stored procedure
11 Apr 2024Alexander Korotkovmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1reverts 6 commits here
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 1 reviewer saw the final version

Reverts the initial feature and five follow-up commits after Heikki's race-condition and readiness review in #90. Korotkov confirms the revert in #91; no revert attachment appears.

  • Heikki Linnakangassaw an earlier versionLLM analysis, 100% sureDemonstrated waiter-cleanup races and requested reversion in #90, with substantial feature changes following before the thread's final patchset.
Clarify what is protected by WaitLSNLock
6 Apr 2024Alexander Korotkovmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #860 of 1 reviewer saw the final version

Exactly matches the second patch in #86, moving procInfos initialization inside WaitLSNLock and documenting minWaitedLSN and inHeap protection.

  • Alvaro Herrerasaw an earlier versionLLM analysis, 96% sureQuestioned protection of the committed pairing-heap state in #83, leading Korotkov to move assignments under the lock and document the invariant in #86.
Use an LWLock instead of a spinlock in waitlsn.c
6 Apr 2024Alexander Korotkovmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #861 of 2 reviewers saw the final version

Exactly matches the first patch in #86, originally supplied by Herrera in #83, replacing spinlock operations and registering WaitLSNLock. The second patch was committed separately.

  • Thomas Munrosaw an earlier versionLLM analysis, 85% sureFlagged unbounded backend-array work and SetLatch() under a spinlock in #56, before the pairing-heap and LWLock revisions.
  • Alexander Korotkovsaw the final versioncommittercommitted the patch
Call WaitLSNCleanup() in AbortTransaction()
6 Apr 2024Alexander Korotkovmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

Adds WaitLSNCleanup() to AbortTransaction() in xact.c following Herrera's error-cleanup objection in #87, as confirmed in #88. The commit explicitly credits Herrera as author; no matching attachment preceded it.

  • Alexander Korotkovsaw the final versioncommittercommitted the patch
Use the pairing heap instead of a flat array for LSN replay waiters
3 Apr 2024Alexander Korotkovmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 1 reviewer saw the final version

Implements Korotkov's pairing-heap proposal in #82 after Herrera questioned flat-array scaling in #79. It adds pairingheap_initialize() and rewrites waiter management; the implementation was committed before any matching attachment.

  • Alvaro Herrerasaw an earlier versionLLM analysis, 100% sureChallenged the flat array's insertion cost in #79, prompting the structural replacement announced in #82.
Minor improvements for waitlsn.c
3 Apr 2024Alexander Korotkovmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1Also in: pgsql: Implement pg_wal_replay_wait() stored procedure (not this thread)
LLM analysis, 100% surewritten from the discussion, no matching patch1 of 3 reviewers saw the final version

Combines Herrera's inline include cleanup and shorter spinlock scope from #79 with endtime initialization proposed in #80. The commit also credits Masahiko Sawada's report outside the supplied thread.

  • Alvaro Herrerasaw an earlier versionLLM analysis, 98% sureReviewed the committed code and proposed include removal and moving cur initialization before locking in #79, before the later waiter-management reworks.
  • Daniel Gustafssonsaw an earlier versionLLM analysis, 98% sureReported the NetBSD compiler failure and proposed initializing endtime in #80, before subsequent changes to the waiting implementation.
  • Masahiko Sawadasaw the final versionLLM analysis, 88% sureThe Reported-by credit and second Discussion trailer in e37662f221 identify an additional report, although he has no message in the supplied thread.
Implement pg_wal_replay_wait() stored procedure
2 Apr 2024Alexander Korotkovmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% suremodified before commitmoderate changesfrom patchset #690 of 9 reviewers saw the final version

The canonical April commit matches #69 except for InvalidateCatalogSnapshot(), a catalog-version bump, and a corrected timeout-test regex. Its grouped August relanding derives from #101, explaining the additional later reviewers in the supplied metadata.

  • Michael Paquiersaw an earlier versiontrailerExamined the alternatives and per-record replay wakeups in #3, before the procedure and waiter-management reworks.
  • Peter Eisentrautsaw an earlier versiontrailerReviewed interface and pooler usability in #4 and proposed interval timeout arguments in #44, before further API and implementation changes.
  • Dilip Kumarsaw an earlier versiontrailerRequested clearer separation of the alternatives and missing syntax documentation in #23, before the sorted-array and procedure redesigns.
  • Amit Kapilasaw an earlier versiontrailerReviewed syntax and implicit-transaction usability in #26 and #34, then assessed the procedure proposal in #39 before further implementation changes.
  • Alexander Lakhinsaw an earlier versiontrailerOff-list corrections are explicitly incorporated in #29, #57, and #63, before promotion handling and subsequent snapshot changes.
  • Bharath Rupireddysaw an earlier versiontrailerReviewed validation, recovery overhead, timeout behavior, and errors in #33 and #66, followed by local logic changes in #69 and at commit.
  • Euler Taveirasaw an earlier versiontrailerProposed the final procedure name and bigint-millisecond timeout in #59 and #62, before promotion and validation changes.
  • Heikki Linnakangassaw an earlier versiontrailerReviewed the landed implementation's descendant in #90, identifying waiter races and execution restrictions that led to the revert and substantial relanding changes.
  • Kyotaro Horiguchisaw an earlier versiontrailerReviewed the relanding series in #96, finding a logical-walsender assertion failure and proposing latch, timeout, and documentation changes before the later snapshot revision.

Credits across 12 commits

Author
Alexander Korotkov3 LLM6
Alvaro Herrera2
Kartyshov Ivan2
Reviewer
Alexander Lakhin3
Alexander Korotkovcommitter2
Amit Kapila2
Bharath Rupireddy2
Dilip Kumar2
Euler Taveira2
Michael Paquier2
Peter Eisentraut2
Heikki Linnakangas1
Kyotaro Horiguchi1
Reported by
Alexander Lakhin2
Alvaro Herrera2
Kevin Hale Boyes2
Daniel Gustafsson1
Masahiko Sawada1
Committer
Alexander Korotkov12

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