Jun 2024

1 commit
Make RelationFlushRelation() work without ResourceOwner during abort
6 Jun 2024Heikki Linnakangasmaster17beta2REL_19_STABLE17beta2REL_18_STABLE17beta2REL_17_STABLE17beta2
LLM analysis, 100% suremodified before commitlight changesfrom patchset #880 of 1 reviewer saw the final version

Matches the relcache and AbortSubTransaction implementation in #88; the committed SQL and expected output capitalize and format SET/RESET statements.

  • Alexander Lakhinsaw an earlier versionLLM analysis, 96% sureReported a reproducible logical-decoding failure in the landed refactor at #86, prompting the new relcache implementation in #88, which he did not subsequently review in this thread.

Feb 2024

1 commit
Fix bug in bulk extending temp relation after failure
2 Feb 2024Heikki Linnakangasmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 1 reviewer saw the final version

No attachment contains this fix: #84 reports the disk-full reproduction, and #85 explains the added ResourceOwnerEnlarge call in ExtendBufferedRelLocal and the documented Forget/Remember restriction.

  • Alexander Lakhinsaw an earlier versionLLM analysis, 96% sureTested the landed refactor and supplied the disk-full reproducer in #84; the corrective implementation and later relcache patch followed that testing.

Jan 2024

1 commit
Add missing PGDLLIMPORT markings
16 Jan 2024Heikki Linnakangasmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 1 reviewer saw the final version

#81 identifies missing PGDLLIMPORT on the buffer resource descriptors; #82 confirms the fix and additional markings in src/include/postmaster/walsummarizer.h, with no attached patch.

  • Anton A. Melnikovsaw an earlier versionLLM analysis, 95% sureIdentified missing export markings in the landed resource-descriptor declarations at #81, with no review of the later patchsets.

Nov 2023

8 commits
Add test_dsa module.
15 Nov 2023Heikki Linnakangasmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #76

Extracts the test_dsa component of #76, with comment formatting and a test tweak changing the explicit dsa_free loop from 100 to 500 allocations.

Clear CurrentResourceOwner earlier in CommitTransaction.
15 Nov 2023Heikki Linnakangasmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #761 of 1 reviewer saw the final version

Takes only the third component of #76, moving CurrentResourceOwner = NULL earlier in CommitTransaction; both changed lines match the attachment exactly.

  • Alexander Lakhinsaw the final versiontrailerReported the commit-time failure in #72 and tested the resulting correction series in #75, after which this component remained unchanged.
Fix dsa.c with different resource owners.
15 Nov 2023Heikki Linnakangasmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #761 of 3 reviewers saw the final version

Extracts the second component of #76: dsa_area stores its ResourceOwner and NULL denotes pinned mappings; implementation differences are only whitespace and comment wrapping.

  • Alexander Lakhinsaw an earlier versiontrailerReported the original failure in #72 and a Valgrind-detected stale-owner bug in #75, prompting the subsequent pinned-mapping logic change.
  • Thomas Munrosaw the final versiontrailerEndorsed explicit DSA ownership and explained why retaining dsa_pin_mapping supports exception-safe construction in #77.
  • Andres Freundsaw an earlier versiontrailerSupported the explicit-owner direction and discussed attach-then-pin behavior in #74, before the pinned-mapping correction in #76.
Fix bug in the new ResourceOwner implementation.
8 Nov 2023Heikki Linnakangasmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 1 reviewer saw the final version

#70 reports the assertion failure and #71 explains the fix: ResourceOwnerReleaseAll must test nhash rather than hash allocation, matching the commit's two condition changes.

  • Alexander Lakhinsaw an earlier versionLLM analysis, 97% sureSupplied the assertion-failure reproducer in #70 before this fix, with subsequent DSA and relcache revisions beyond the version he tested.
Change pgcrypto to use the new ResourceOwner mechanism.
8 Nov 2023Heikki Linnakangasmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #671 of 2 reviewers saw the final version

Extracts the fourth component of #67; all added and removed lines in contrib/pgcrypto/openssl.c match exactly.

  • Peter Eisentrautsaw the final versiontrailerReviewed pgcrypto's release-priority choice and the descriptor API in #59, then endorsed the revised APIs in #65, after which this component was unchanged.
  • Andres Freundsaw an earlier versiontrailerRequested const descriptors and typed inline wrappers in #60, before the callback and descriptor API changes incorporated into the pgcrypto conversion.
Use a faster hash function in resource owners.
8 Nov 2023Heikki Linnakangasmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #674 of 9 reviewers saw the final version

Extracts the third component of #67; hash_resource_elem and the new murmurhash64 in src/include/common/hashfn.h have identical changed lines.

  • Aleksander Alekseevsaw the final versiontrailerTested the series, specifically reviewed hash casts in #46, and endorsed #57 in #58, with no subsequent substantive hash change.
  • Michael Paquiersaw the final versiontrailerExamined the optimized benchmark results and questioned the array-to-hash performance transition in #28, while the hash algorithm subsequently remained the same.
  • Julien Rouhaudsaw an earlier versiontrailerSupported the cheaper-hash proposal while reviewing the initial refactor in #5, before an attached MurmurHash implementation existed.
  • Kyotaro Horiguchino review foundtrailerReviewed release-loop behavior in #8, but the thread shows no examination of the later hash-function component.
  • Hayato Kurodasaw the final versiontrailerReproduced the optimized series' benchmark results in #27, and the committed hash algorithm differs only in surrounding declarations and comments.
  • Álvaro Herrerasaw an earlier versiontrailerChallenged the slowdown in #19, helping motivate the subsequently added faster-hash component.
  • Zhihong Yuno review foundtrailerQuestioned release-loop termination in #36 and #38, with no identifiable review of the hash-function component.
  • Peter Eisentrautno review foundtrailerReviewed resource descriptors and cleanup APIs in #59 and #65, without identifiable comments or testing specific to the hash optimization.
  • Andres Freundsaw the final versiontrailerBenchmarked the series in #48 and questioned truncating a 64-bit hash in #60, after which the hash implementation remained unchanged.
Make ResourceOwners more easily extensible.
8 Nov 2023Heikki Linnakangasmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #670 of 9 reviewers saw the final version

Extracts the main refactor from #67, retaining its resource storage, release ordering, APIs, and tests; committer edits remove disabled RESOWNER_TRACE instrumentation and polish comments and formatting.

  • Aleksander Alekseevsaw an earlier versiontrailerSupplied cleanup changes, ran regression and sanitizer checks, found missing test integration, and endorsed the series in #58, before later API and sorting changes.
  • Michael Paquiersaw an earlier versiontrailerReviewed array sizing, JIT ownership, copy-paste errors, and benchmark behavior in #2, #12, #23, and #28, before the release-order redesign.
  • Julien Rouhaudsaw an earlier versiontrailerReviewed the initial refactor and identified an obsolete array-capacity comment in #5, before substantial cleanup-contract changes.
  • Kyotaro Horiguchisaw an earlier versiontrailerExamined array and hash release-loop correctness in #8, leading to revisions long before sorted priority-based cleanup.
  • Hayato Kurodasaw an earlier versiontrailerReviewed wrapper consistency and reservation safety in #14 and #17 and reproduced benchmarks in #27, before the later cleanup redesign.
  • Álvaro Herrerasaw an earlier versiontrailerRequested clearer callback names and wrappers and challenged the performance regression in #19, before substantial later revisions.
  • Zhihong Yusaw an earlier versiontrailerExamined hash release-loop conditions and termination in #36 and #38, before those loops were replaced by sorted release.
  • Peter Eisentrautsaw an earlier versiontrailerShaped descriptor naming, warning callbacks, priorities, and typed wrappers in #59 and reviewed the updated APIs in #65, before the final LLVM and empty-hash logic changes.
  • Andres Freundsaw an earlier versiontrailerDrove release-order, error-state, and structure-layout changes in #48, #60, and #64, before the final submission also changed empty-hash sorting behavior.
Move a few ResourceOwnerEnlarge() calls for safety and clarity.
8 Nov 2023Heikki Linnakangasmaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #672 of 9 reviewers saw the final version

Extracts the first component of #67; reservation-call movements in bufmgr.c, localbuf.c, and catcache.c match the attachment exactly.

  • Aleksander Alekseevsaw the final versiontrailerTested successive complete series and endorsed #57 in #58, after the PinBuffer reservation responsibility had reached its committed form.
  • Michael Paquiersaw an earlier versiontrailerIdentified the resource-owner comment typo in #23, before the later reversal of which side of PinBuffer performs reservation.
  • Julien Rouhaudno review foundtrailerHis substantive review in #5 predates this preparatory component, and later participation only requests a rebase.
  • Kyotaro Horiguchino review foundtrailerHis review in #8 concerns resource release loops, with no identifiable review of the later reservation-call movements.
  • Hayato Kurodasaw an earlier versiontrailerAudited all ResourceOwnerEnlarge call sites after the preparatory patch in #17, before PinBuffer reservation responsibility was changed again.
  • Álvaro Herrerano review foundtrailerHis substantive comments in #19 address callbacks and refactor performance rather than this reservation-call component.
  • Zhihong Yuno review foundtrailerHis comments in #36 and #38 concern release-loop behavior rather than reservation placement.
  • Peter Eisentrautno review foundtrailerHis reviews in #59 and #65 concern the new resource APIs, with no identifiable review of this preparatory component.
  • Andres Freundsaw the final versiontrailerQuestioned the overhead of additional ResourceOwnerEnlargeBuffers calls in #60, and those reservation movements remained unchanged through commit.

Credits across 11 commits

Author
Heikki LinnakangasLLM8
Reviewer
Andres Freund5
Peter Eisentraut4
Aleksander Alekseev3
Alvaro Herrera3
Hayato Kuroda (Fujitsu)3
Julien Rouhaud3
Kyotaro Horiguchi3
Michael Paquier3
Zhihong Yu3
Alexander Lakhin2
Thomas Munro1
Reported by
Alexander Lakhin4
Anton A. Melnikov1
Committer
Heikki Linnakangas11

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