Sep 2026

1 commit
Wait for transactions of an initial decoding snapshot to commit
24 Sep 2026Álvaro HerreramasterunreleasedREL_19_STABLE19rc1REL_18_STABLE18.7REL_17_STABLE17.12REL_16_STABLE16.16REL_15_STABLE15.20REL_14_STABLE14.25Also in: Race conditions in logical decoding (part of a posted patchset)

May 2026

4 commits
Allow old WAL recycling during REPACK CONCURRENTLY
29 May 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #3713 of 3 reviewers saw the final version

Extracts the repack_worker.c portion of #371, retaining LogicalIncreaseRestartDecodingForSlot and rewriting its explanatory comment; the TAP test is omitted. #373 confirms splitting the implementation into two commits.

  • Antonin Houskasaw the final versiontrailerApproved the worker-slot advancement in #302, after which this component received only rebasing, explanatory changes and tests.
  • Amit Kapilasaw the final versiontrailerRequested temporary-slot safety explanations in #369 and said the revised code looked good in #372.
  • Álvaro Herrerasaw the final versioncommittercommitted the patch
Advance restart_lsn more eagerly in LogicalConfirmReceivedLocation
29 May 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #3712 of 2 reviewers saw the final version

Extracts the logical.c portion of #371, recomputing the required WAL LSN whenever restart_lsn advances, with only comment edits. The worker change and TAP test are separate.

  • Amit Kapilasaw the final versiontrailerReviewed the combined WAL-recycling patch in #369 and approved its revised code in #372.
  • Álvaro Herrerasaw the final versioncommittercommitted the patch
Remove unnecessary signal handler change
29 May 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 100% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

#323 identifies the obsolete background-worker signal-handler override; the commit removes that call and its comment from repack_worker.c without a matching attachment.

  • Alexander Lakhinsaw the final versionLLM analysis, 100% sureIdentified the redundant SIGTERM handler assignment and obsolete explanation in #323, which the commit removes.
Revert "Allow logical replication snapshots to be database-specific"
24 May 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1reverts 1 commit here
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 5 reviewers saw the final version

#363 proposes reverting database-specific snapshots because architectural problems cannot be resolved before beta, and #365 confirms the revert. The attached diagnostic fixes are not the source of this reversal.

  • Amit Kapilasaw an earlier versionLLM analysis, 98% sureExplained the snapshot architecture problems in #362 and agreed to the revert in #364, before the thread's remaining follow-up patches.
  • Andres Freundsaw an earlier versionLLM analysis, 97% sureExplained in #276 how indirect shared-catalog access could corrupt caches and why deeper infrastructure was needed.
  • Hayato Kuroda (Fujitsu)saw an earlier versionLLM analysis, 97% sureQuestioned the assertion-only protection and unsafe reuse of serialized database-specific snapshots in #271.
  • Mihail Nikalayeusaw an earlier versionLLM analysis, 98% sureReported reproducible duplicate-key and lost-data failures attributable to database-specific decoding in #342.
  • Masahiko Sawadasaw an earlier versionLLM analysis, 96% sureIdentified cross-database snapshot restoration and standby-decoding problems in #351 and #352.

Apr 2026

12 commits
Move REPACK (CONCURRENTLY) test out of stock regression tests
23 Apr 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 99% suremodified before commitlight changesfrom patchset #3361 of 1 reviewer saw the final version

Implements the test relocation from #336 into contrib/test_decoding, with revised explanatory test comments; #337 confirms the push.

  • Tom Lanesaw the final versionLLM analysis, 96% sureIdentified the wal_level=minimal regression and required moving these tests to another suite in #330, the recommendation implemented by #336.
REPACK: do not require REPLICATION or LOGIN
20 Apr 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 99% suremodified before commitlight changesfrom patchset #3271 of 2 reviewers saw the final version

Combines both privilege patches from #327, with test adjustments and movement of cluster in parallel_schedule. Antonin supplied the original REPLICATION-check removal in #325 and is credited as author.

  • Chao Lisaw an earlier versiontrailerReported independently finding the same issue and approved the REPLICATION fix in #326, before LOGIN bypass and broader tests were added.
  • Álvaro Herrerasaw the final versioncommittercommitted the patch
Add missing initialization
16 Apr 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #3102 of 2 reviewers saw the final version

Combines both attachments in #310: initializing shared->initialized and replacing the shared DSM pointer with worker-local storage. Differences are placement, naming and commentary for the local variable.

  • Alexander Lakhinsaw the final versionLLM analysis, 90% sureReported the committed worker's invalid-boolean sanitizer failure in #306, leading directly to the small initialization correction in #310.
  • Álvaro Herrerasaw the final versioncommittercommitted the patch
Simplify declaration of memcpy target
8 Apr 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 100% surewritten from the discussion, no matching patch2 of 2 reviewers saw the final version

#284 reports the 32-bit tuple-header storage warning, and #286 proposes the committed uint64 chunk_header replacement in restore_tuple; no attachment supplies this fix.

  • Tomas Vondrasaw the final versionLLM analysis, 100% sureReported the 32-bit bounds warning that prompted the tuple-header storage correction in #284.
  • Tom Lanesaw the final versionLLM analysis, 100% sureProposed the exact uint64 declaration adopted by the commit in #286.
Reserve replication slots specifically for REPACK
7 Apr 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 99% surepart of a posted patchsetlight changesfrom patchset #2521 of 1 reviewer saw the final version

Takes the reserved-replication-slots component of #252, with comment edits and removal of an incorrect worker-limit hint; the snapshot and deadlock components are separate.

  • Srinath Reddy Sadipirallasaw the final versiontrailerFound the Windows initialization failure credited in #225 and reviewed its slot-accounting correction in #231, followed only by light changes to this component.
doc: Add an example of REPACK (CONCURRENTLY)
7 Apr 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 100% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

#242 requests a concurrent REPACK example and #243 accepts it; the commit adds that example to doc/src/sgml/ref/repack.sgml without an intervening attachment.

  • vignesh Csaw the final versionLLM analysis, 100% sureRequested the concurrent-mode documentation example while reviewing #239 in #242, and the committer accepted that request.
Allow logical replication snapshots to be database-specific
7 Apr 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #2521 of 1 reviewer saw the final version

Takes the database-specific snapshot component of #252, whose changed code matches apart from whitespace; Antonin's preceding replacement is #241.

  • Álvaro Herrerasaw the final versiontrailerReviewed horizon and cleanup behavior in #234 and #236, integrated the final revision and reviewed the committed version as committer.
Avoid different-size pointer-to-integer cast
7 Apr 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 100% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

#266 reports the pointer-width build failure, and #267 confirms the one-line replacement of (Datum) NULL with (Datum) 0 in repack.c.

  • John Naylorsaw the final versionLLM analysis, 100% sureReported the buildfarm pointer-to-integer-size failure in #266, leading directly to the committed cast correction.
Fix valgrind failure
7 Apr 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 100% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

#259 supplies the inline zero-initialization fix for serialized snapshot padding after #257; the commit uses an initializer and also moves pfree after BufFileClose.

  • Tom Lanesaw the final versionLLM analysis, 99% sureReported the Valgrind failure in #257, prompting the padding initialization diagnosed and supplied in #259.
Add CONCURRENTLY option to REPACK
6 Apr 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1Also in: why there is not VACUUM FULL CONCURRENTLY? (not this thread)
LLM analysis, 100% surepart of a posted patchsetmoderate changesfrom patchset #2392 of 7 reviewers saw the final version

Takes the concurrent feature from #239, with local restore_tuple memory-context, external-value validation and worker-helper changes described in #252; the rename component landed separately. This April 2026 submission supersedes the June 2025 source claimed by thread #49123, whose attribution I therefore disagree with.

  • Matthias van de Meentsaw an earlier versiontrailerReviewed replication privileges and resource exhaustion in #151 and #160, before the table-AM and tuple-slot rework.
  • Srinath Reddy Sadipirallasaw an earlier versiontrailerStress-tested successive versions and found decoding, transaction-validation and cleanup defects in #123, #165, #181 and #215, before later local logic changes.
  • Amit Kapilasaw the final versiontrailerReviewed snapshot and shared-catalog assumptions earlier and examined replica-identity and apply-worker choices for the final submitted version in #249.
  • Jim Jonesno review foundtrailerHis report in #175 concerns the existing command's temporary-table selection and was moved to a separate bugfix discussion, with no identifiable review of this concurrent implementation.
  • Robert Treatsaw an earlier versiontrailerEvaluated REPLICATION privileges and the dedicated-slot design in #153, before substantial implementation changes.
  • Noriyoshi Shinodano review foundtrailerHis contribution in #139 documents the already-committed basic progress view, with no identifiable review of the concurrent implementation.
  • vignesh Csaw the final versiontrailerReviewed the final submission in #242, raising replica-identity, testing, documentation, include and naming issues.
Rename cluster.c to repack.c (and corresponding .h)
6 Apr 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #239

Takes the cluster.c/h-to-repack.c/h rename component of #239; its changed lines match modulo whitespace, while the concurrent feature is a separate commit.

Allow index_create to suppress index_build progress reporting
5 Apr 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 99% surepart of a posted patchsetmoderate changesfrom patchset #2340 of 1 reviewer saw the final version

Extracts progress suppression from #234, descended from Antonin's #150, and changes index_create_copy to pass index_create flag bits rather than another boolean.

  • Mihail Nikalayeusaw an earlier versiontrailerIdentified remaining AM-specific progress writes and TOAST flag propagation in #170, before the committed flag-interface redesign.

Mar 2026

2 commits
Document the 'command' column of pg_stat_progress_repack
12 Mar 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1392 of 2 reviewers saw the final version

Includes the missing command-column documentation from #139 and adds explanatory updates for pg_stat_progress_cluster compatibility in monitoring.sgml, as announced in #143.

  • Antonin Houskasaw the final versionLLM analysis, 100% sureExplicitly approved the documentation patch in #141, followed only by additional documentation edits.
  • Álvaro Herrerasaw the final versioncommittercommitted the patch
Introduce the REPACK command
10 Mar 2026Álvaro Herreramaster19beta1REL_19_STABLE19beta1Also in: why there is not VACUUM FULL CONCURRENTLY? (not this thread)
LLM analysis, 100% surepart of a posted patchsetmoderate changesfrom patchset #1361 of 7 reviewers saw the final version

Commits the basic-command component of #136, with local changes in src/backend/commands/cluster.c to skipped-relation lock release, ANALYZE transactions and partitioned-table errors, confirmed by #137. This March 2026 submission supersedes the June 2025 source claimed by thread #49123, whose attribution I therefore disagree with.

  • Mihail Nikalayeusaw an earlier versiontrailerReviewed basic command behavior, progress reporting and cleanup in #37, #59, #83 and #113, before the committer's local logic changes.
  • Álvaro Herrerasaw the final versiontrailerReworked and reviewed the command throughout the thread and announced committing the final version with additional changes in #137.
  • Robert Treatsaw an earlier versiontrailerReviewed command tags, clustered-index persistence, ANALYZE and user-facing behavior in #2, #5 and #38, before later implementation changes.
  • Euler Taveirano review foundtrailerHis in-thread feedback in #18 concerns the proposed command-line utilities and their deprecation, rather than the SQL implementation that landed.
  • Matheus Alcantarano review foundtrailerThe commit credits him, but no review activity by him appears in this thread.
  • Junwang Zhaono review foundtrailerThe commit credits him, but no review activity by him appears in this thread.
  • jian hesaw an earlier versiontrailerReviewed grammar, ANALYZE options, defaults, completion and tests in #41 and #48, before later command revisions.

Sep 2025

1 commit
Split vacuumdb to create vacuuming.c/h
26 Sep 2025Álvaro Herreramaster19beta1REL_19_STABLE19beta1
LLM analysis, 99% surepart of a posted patchsetmoderate changesfrom patchset #330 of 1 reviewer saw the final version

Takes the vacuuming.c/h extraction from #33, changing objfilter to bits32 inside vacuumingOptions and replacing analyze booleans with RunMode; the accompanying REPACK patch remains uncommitted.

  • Antonin Houskano review foundtrailerAlthough credited by the commit, his visible earlier feedback concerns the utility's symlink implementation, and no review of the vacuuming.c/h extraction is identifiable here.

Credits across 20 commits

Author
Antonin Houska7
Alvaro Herrera2 LLM5
Zhijie Hou (Fujitsu)2
Rui Zhao1
Shinoda, Noriyoshi (PN Japan FSIP)1
Reviewer
Alvaro Herreracommitter8
Amit Kapila3
Antonin Houska2
Mihail Nikalayeu2
Robert Treat2
Srinath Reddy Sadipiralla2
Chao Li1
Euler Taveira1
jian he1
Jim Jones1
Junwang Zhao1
Matheus Alcantara1
Matthias van de Meent1
Shinoda, Noriyoshi (PN Japan FSIP)1
vignesh C1
Reported by
Alexander Lakhin2
Amit Kapila1
Justin Pryzby1
Mihail Nikalayeu1
Tomas Vondra1
Co author
Alvaro Herrera2
Mihail Nikalayeu1
Srinath Reddy Sadipiralla1
Committer
Alvaro Herrera20

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