Apr 2023

9 commits
Reduce the log level in 035_standby_logical_decoding.pl.
25 Apr 2023Amit Kapilamaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1Also in: Add two missing tests in 035_standby_logical_decoding.pl (part of a posted patchset)
035_standby_logical_decoding: Add missing waits for replication
11 Apr 2023Andres Freundmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 100% surelanded as postedfrom patchset #3572 of 2 reviewers saw the final version

The committed diff exactly matches #357, adding replay waits before slot creation on both the standby and cascading standby.

  • Noah Mischsaw the final versionLLM analysis, 96% sureReported the landed TAP test hanging on hoverfly in #354, which led to the two replay waits without changing the test's design.
  • Andres Freundsaw the final versioncommittercommitted the patch
For cascading replication, wake physical and logical walsenders separately
8 Apr 2023Andres Freundmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 92% surepart of a posted patchsetlight changesfrom patchset #3372 of 5 reviewers saw the final version

Extracts the selective physical/logical walsender wakeup component of #337; comparison with #334 shows the same logic with comment and formatting edits. The latest attachments are unavailable because of extraction failure.

  • Jeff Davissaw an earlier versiontrailerReviewed condition-variable correctness and later selective wakeups, supplying archive-wakeup corrections and explanatory wording in #240 and #293, before subsequent replay-condition changes.
  • Robert Haassaw an earlier versiontrailerRequested a user-visible explanation and approved the clarified approach in #294 and #306, before later replay-wakeup logic changes.
  • Amit Kapilasaw the final versiontrailerProposed selective wakeups in #255 and reviewed the final replay guard in #320, which was incorporated with only light subsequent edits.
  • Masahiko Sawadasaw an earlier versiontrailerExamined replication-kind initialization and accepted its slot-creation justification in #280 and #282, before later wakeup corrections.
  • Andres Freundsaw the final versioncommittercommitted the patch
Handle logical slot conflicts on standby
8 Apr 2023Andres Freundmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 93% surepart of a posted patchsetmoderate changesfrom patchset #3373 of 7 reviewers saw the final version

Combines the recovery-conflict component and associated pg_stat_database_conflicts changes from #337, including catalog and statistics format bumps. #352 confirms squashing the statistics component; exact final attachment comparison is unavailable.

  • Drouvot, Bertrandsaw the final versiontrailerReviewed Andres's rewritten components in #335 and endorsed the latest series in #340.
  • Andres Freundsaw the final versiontrailerReworked shared-catalog conflicts and horizon handling in #334, polished the final series and committed it as announced in #352.
  • Robert Haassaw an earlier versiontrailerReviewed conflict policy, diagnostics and the evolving design in #2, #61 and #170, before the final invalidation rewrite.
  • Fabrízio de Royes Mellosaw an earlier versiontrailerTested and corrected earlier conflict handling and statistics through #117, #119, #121 and #149.
  • Bharath Rupireddysaw an earlier versiontrailerReviewed active-slot invalidation and requested monitoring and recovery guidance in #201, before the enum-based rewrite.
  • Amit Kapilasaw the final versiontrailerReviewed conflict and horizon handling and endorsed the final invalidation approach while requesting consistent errors in #346.
  • Alvaro Herrerasaw an earlier versiontrailerReported earlier conflict-test hangs and reviewed diagnostic translation in #103, #277 and #285, before the final structural rewrite.
Support invalidating replication slots due to horizon and wal_level
8 Apr 2023Andres Freundmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 93% surepart of a posted patchsetmoderate changesfrom patchset #3375 of 8 reviewers saw the final version

Takes the generalized horizon/wal_level invalidation infrastructure from #337 and combines its pg_replication_slots.conflicting projection with it. The predecessor patches and #352 establish the split and squash, although the final attachments were not extracted.

  • Drouvot, Bertrandsaw the final versiontrailerApproved the enum-based invalidation rewrite, supplied its explicit-cause API delta in #335, and reviewed the integrated series in #340.
  • Andres Freundsaw the final versiontrailerDesigned the final invalidation refactoring in #334, integrated follow-up changes in #337, and committed the result.
  • Robert Haassaw an earlier versiontrailerReviewed the earlier conflict policy and implementation in #2, #61 and #170, before persistent invalidation and enum/API reworking.
  • Fabrízio de Royes Mellosaw an earlier versiontrailerTested earlier slot-conflict behavior and reviewed its statistics and test fixes in #117, #119 and #149.
  • Bharath Rupireddysaw an earlier versiontrailerRequested observable invalidation status and operational guidance in #201, leading to the conflicting view column before the later rewrite.
  • Amit Kapilasaw the final versiontrailerReviewed invalidation predicates and effective horizons, then approved the final approach and requested consistent errors in #346.
  • Melanie Plagemansaw the final versiontrailerReviewed the final invalidation component's enum names, switch handling and diagnostics in #344, with light corrections confirmed in #345.
  • Alvaro Herrerasaw the final versiontrailerReviewed diagnostic translatability and explicitly accepted the final invalidation component in #343.
Prevent use of invalidated logical slot in CreateDecodingContext()
8 Apr 2023Andres Freundmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 95% surepart of a posted patchsetlight changesfrom patchset #3372 of 2 reviewers saw the final version

Extracts the centralized CreateDecodingContext validity-check component of #337. Against the available version in #334, the substantive diff is unchanged apart from renaming RS_INVAL_WAL to RS_INVAL_WAL_REMOVED.

  • Drouvot, Bertrandsaw the final versiontrailerExplicitly approved the centralized-check component in #335 and the integrated series in #340.
  • Melanie Plagemansaw the final versiontrailerExplicitly marked the centralized-check component as looking good in #344.
Replace replication slot's invalidated_at LSN with an enum
8 Apr 2023Andres Freundmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 94% surepart of a posted patchsetlight changesfrom patchset #3373 of 3 reviewers saw the final version

Extracts the invalidation-enum prerequisite from #337, whose message explicitly adds the SLOT_VERSION bump missing from #334. Final edits rename the WAL-removal cause and polish comments and formatting.

  • Drouvot, Bertrandsaw the final versiontrailerExplicitly approved replacing invalidated_at with an enum in #335 and endorsed the integrated series in #340.
  • Alvaro Herrerasaw the final versiontrailerApproved the enum prerequisite and explained why removing the unused invalidated_at information was reasonable in #343.
  • Melanie Plagemansaw the final versiontrailerRequested a more specific WAL invalidation enum name in #344, adopted as RS_INVAL_WAL_REMOVED in #345.
Pass down table relation into more index relation functions
2 Apr 2023Andres Freundmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #2542 of 4 reviewers saw the final version

Takes the first prerequisite from #254, passing heap Relation through index functions. Direct diff comparison shows only formatting and removal of a stray catalog/catalog.h include, also explained in #259.

  • Alvaro Herrerasaw an earlier versionLLM analysis, 96% sureRecommended retaining the table Relation instead of repeatedly opening it in #136, before several structural alternatives were explored.
  • Robert Haassaw an earlier versionLLM analysis, 96% sureIdentified deadlocks from table_open under buffer locks and argued for passing an existing Relation in #152, before the final plumbing implementation.
  • Melanie Plagemansaw the final versionLLM analysis, 94% sureReviewed heaprel argument placement and explanatory comments in #216, adopted in #225 with only light subsequent plumbing changes.
  • Andres Freundsaw the final versioncommittercommitted the patch
Assert only valid flag bits are passed to visibilitymap_set()
2 Apr 2023Andres Freundmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 99% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

Adds the stricter visibilitymap_set flags assertion identified as missing in #251. No attached patch contains this standalone assertion change.

  • Bertrand Drouvotsaw the final versionLLM analysis, 76% sureAccepted the visibility-map flag-validation and isolation approach in #254, although his response chiefly discusses masking WAL-only bits.

Mar 2019

1 commit
Compute XID horizon for page level index vacuum on primary.
26 Mar 2019Andres Freundmaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 moreAlso in: Pluggable Storage - Andres's take (part of a posted patchset)Also in: Computing the conflict xid for index page-level-vacuum on primary (not this thread)

Credits across 10 commits

Author
Bertrand Drouvot6
Andres Freund2 LLM5
Amit Khandekar3
Robert Haas1
Reviewer
Andres Freundcommitter5
Amit Kapilacommitter4
Bertrand Drouvot4
Robert Haas4
Alvaro Herrera3
Melanie Plageman3
Bharath Rupireddy2
Fabrízio de Royes Mello2
Jeff Davis1
Masahiko Sawada1
Reported by
Noah Misch1
Committer
Andres Freund9
Amit Kapila1

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