Sep 2022

1 commit
Fix cache invalidation bug in recovery_prefetch.
3 Sep 2022Thomas Munromaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1REL_15_STABLE15beta4
LLM analysis, 100% suremodified before commitmoderate changesfrom patchset #1290 of 3 reviewers saw the final version

Combines all three patches from #129 across xlogreader.c, xlogrecovery.c and xlogreader.h. Besides comment edits, it restores unconditional XLogReaderInvalReadState at ReadPageInternal's error exit, which the attachment removed.

  • Noah Mischsaw an earlier versionLLM analysis, 96% sureProvided the deterministic promotion-PANIC reproducer in #116 and confirmed recovery_prefetch-dependent results in #118, before the corrective attachments.
  • Tomas Vondrasaw an earlier versionLLM analysis, 98% sureTested Noah's reproducer and isolated prefetching in #117, then questioned cache performance in #128; the committed error-path logic was adjusted afterward.
  • Kyotaro Horiguchisaw an earlier versionLLM analysis, 99% sureAnalyzed the XLREAD_WOULDBLOCK failure path and proposed an alternative fix in #121 and #122, informing the later retry redesign.

Mar - Apr 2022

5 commits
Revert "Logical decoding of sequences"
7 Apr 2022Tomas Vondramaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1reverts 8 commits here
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 2 reviewers saw the final version

Reverts the sequence infrastructure, plugin implementations and follow-up fixes after the snapshot and ordering problems discussed in #98, #107 and #111. No attached revert exists; #113 announces it and attaches a replacement design.

  • Amit Kapilasaw an earlier versionLLM analysis, 98% sureDemonstrated missing sequence updates in #98 and explained remaining redesign risks in #111, providing the substantive basis for reverting the feature.
  • Masahiko Sawadasaw an earlier versionLLM analysis, 91% sureConfirmed the subscription-refresh problem in #99 and identified replay after subscriber crashes in #108, adding to the unresolved correctness problems.
Minor improvements in sequence decoding code and docs
25 Mar 2022Tomas Vondramaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 99% surepart of a posted patchsetmoderate changesfrom patchset #921 of 1 reviewer saw the final version

Takes the ObjectsInPublicationToOids cleanup from #92 but omits its tab-completion consolidation, as stated in #94. It also incorporates co-author Amit's #90 privilege check, relkind lookup cleanup and documentation corrections.

  • Tomas Vondrasaw the final versioncommittercommitted the patch
Handle sequences in preprocess_pubobj_list
25 Mar 2022Tomas Vondramaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 1 reviewer saw the final version

Implements Tomas's response in #88 to Amit's multi-sequence publication crash in #86. The gram.y continuation handling and regression test were not posted as an attachment.

  • Amit Kapilasaw an earlier versionLLM analysis, 100% sureReported the reproducible CREATE PUBLICATION crash with two sequences in #86, directly prompting the parser fix.
Update tab-completion for CREATE PUBLICATION with sequences
25 Mar 2022Tomas Vondramaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surelanded as postedfrom patchset #871 of 1 reviewer saw the final version

The src/bin/psql/tab-complete.c changes match Masahiko's attachment in #87, including CREATE PUBLICATION sequence names and schema completion.

  • Tomas Vondrasaw the final versioncommittercommitted the patch
Add decoding of sequences to built-in replication
24 Mar 2022Tomas Vondramaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1Also in: Include sequence relation support in logical replication (not this thread)
LLM analysis, 100% suremodified before commitlight changesfrom patchset #821 of 5 reviewers saw the final version

Matches the built-in replication patch in #82, with documentation corrections, clearer errors/comments and the final protocol message tag 'Q'. The substantive SetSequence transactional/in-place split is already present in the attachment.

  • Peter Eisentrautsaw an earlier versiontrailerReviewed grammar, duplication, catalogs, protocol documentation and tests through #75, before #82 changed non-transactional sequence application.
  • Amit Kapilasaw the final versiontrailerRequested in-place sequence updates in #78 and subsequently reviewed the committed version for synchronization, privileges and parser failures in #86 and #90.
  • Hannu Krosingsaw an earlier versiontrailerProposed commit-time sequence-state logging in #13, before the substantial later publication and subscriber revisions.
  • Andres Freundsaw an earlier versiontrailerCritiqued the experimental commit-time approach and unlogged sequence consumption in #19, well before the final built-in replication implementation.
  • Petr Jelineksaw an earlier versiontrailerSupported returning to direct decoding in #18 and discussed in-place updates and subscriber conflict semantics in #79 and #81, before the SetSequence revision.

Mar 2022

2 commits
Stabilize test_decoding touching with sequences
8 Mar 2022Tomas Vondramaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 1 reviewer saw the final version

Adds include-sequences=0 to additional test_decoding queries after Amit's unexpected-output report in #56. Tomas reproduced checkpoint sensitivity in #62 and announced the unattached fix in #63.

  • Amit Kapilasaw an earlier versionLLM analysis, 99% sureReported the unexpected decoded sequence row and failing expected output in #56, leading to disabling sequence output in unrelated tests.
Call ReorderBufferProcessXid from sequence_decode
7 Mar 2022Tomas Vondramaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surewritten from the discussion, no matching patch0 of 2 reviewers saw the final version

Adds exactly the ReorderBufferProcessXid call identified as missing from sequence_decode in #57. Tomas accepts the diagnosis in #59 and announces the fix in #61, without an attached patch.

  • Amit Kapilasaw an earlier versionLLM analysis, 100% sureInspected sequence_decode and identified its missing ReorderBufferProcessXid call in #57, which the commit adds.
  • Peter Smithsaw an earlier versionLLM analysis, 85% sureReported another occurrence of the reorder-buffer assertion in #58 before the missing-call fix was committed.

Feb 2022

2 commits
Add decoding of sequences to test_decoding
11 Feb 2022Tomas Vondramaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1Also in: Include sequence relation support in logical replication (not this thread)
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #411 of 3 reviewers saw the final version

Commits only the test_decoding component of #41, leaving built-in replication pending. Normalized component comparison finds only two removed redundant comments in test_decoding.c.

  • Peter Eisentrautsaw the final versiontrailerReviewed the tests and supplied corrections, then declared the initial components ready apart from cosmetic issues in #39, which were incorporated before landing.
  • Hannu Krosingsaw an earlier versiontrailerReviewed sequence semantics in #13, before the output option, callback arguments and tests were revised.
  • Andres Freundsaw an earlier versiontrailerReviewed the alternative sequence semantics in #19, before that experiment was dropped and the plugin interface and defaults were finalized.
Logical decoding of sequences
10 Feb 2022Tomas Vondramaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1Also in: Include sequence relation support in logical replication (not this thread)
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #401 of 3 reviewers saw the final version

Commits the infrastructure component of #40, excluding its test_decoding and built-in replication patches. Differences within that component are documentation/comment polishing and removal of a duplicate semicolon in DecodeSeqTuple.

  • Peter Eisentrautsaw the final versiontrailerSupplied infrastructure fixes in #21, followed up in #23 and judged the infrastructure ready in #39, with only light changes before commit.
  • Hannu Krosingsaw an earlier versiontrailerProposed an alternative logging design in #13, before changes to tuple access, callback arguments and logical-only XID assignment.
  • Andres Freundsaw an earlier versiontrailerIdentified architectural and unlogged-consumption problems with the experimental design in #19, before the return to direct decoding and subsequent implementation revisions.

Credits across 10 commits

Author
Cary Huang3
Tomas Vondra3
Amit Kapila1
Masahiko Sawada1
Thomas MunroLLM1
vignesh C1
Reviewer
Andres Freund3
Hannu Krosing3
Peter Eisentraut3
Tomas Vondracommitter2
Amit Kapila1
Petr Jelinek1
Reported by
Amit Kapila3
Noah Misch1
Committer
Tomas Vondra9
Thomas Munro1

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