Aug 2022

3 commits
When using the WAL-logged CREATE DATABASE strategy, bulk extend.
18 Aug 2022Robert Haasmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1REL_15_STABLE15beta4
LLM analysis, 100% surelanded as postedfrom patchset #2791 of 3 reviewers saw the final version

The bufmgr.c patch in #279 has the same stable patch ID as the commit: bulk extension with PGAlignedBlock followed by copying to explicit block numbers.

  • Andres Freundsaw an earlier versionLLM analysis, 88% sureReviewing the existing copying loop introduced by #194, he proposed bulk extension and avoiding P_NEW in #261, before the optimization was attached.
  • Tom Lanesaw an earlier versionLLM analysis, 85% sureIn #263 he endorsed extending through the last block and explained the zero-fill guarantee, reviewing the approach before its attached implementation.
  • Robert Haassaw the final versioncommittercommitted the patch
Avoid using a fake relcache entry to own an SmgrRelation.
12 Aug 2022Robert Haasmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1REL_15_STABLE15beta4
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #2764 of 4 reviewers saw the final version

Exactly matches the ownership-fix component of #276 by stable patch ID, excluding its bulk-extension component; the committed changes replace fake relcache ownership and explicitly close smgr entries.

  • Andres Freundsaw the final versionLLM analysis, 94% sureHe reviewed ownership and closure in #248 and #275, and the committed component restores the implementation already reviewed in #235, modulo rebase.
  • Tom Lanesaw the final versionLLM analysis, 88% sureHe examined fake-relcache removal in #251 and requested explicit closure in #273, restoring the closure behavior present in #235; the commit expressly credits his review.
  • Justin Pryzbysaw the final versionLLM analysis, 89% sureHe confirmed that the patch fixed access to freed memory in #238, and its ownership changes match the final component modulo rebase.
  • Robert Haassaw the final versioncommittercommitted the patch
Fix data-corruption hazard in WAL-logged CREATE DATABASE.
6 Aug 2022Tom Lanemaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1REL_15_STABLE15beta3
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #2701 of 3 reviewers saw the final version

Takes the first component of #270, removing empty-page skipping and locking source buffers in RelationCopyStorageUsingBuffer(); committer edits remove a blank line and release the destination before the source.

  • Andres Freundsaw an earlier versionLLM analysis, 88% sureReviewing the landed copying implementation, he confirmed the block-position corruption in #257 and examined source-buffer locking in #260 before the corrective patch was posted.
  • Robert Haassaw an earlier versionLLM analysis, 76% sureHe acknowledged the empty-page corruption and agreed with the missing-lock diagnosis in #265 and #266, before the attached fix.
  • Tom Lanesaw the final versioncommittercommitted the patch

Mar 2022

5 commits
initdb: When running CREATE DATABASE, use STRATEGY = WAL_COPY.
31 Mar 2022Robert Haasmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surelanded as postedfrom patchset #2212 of 2 reviewers saw the final version

The initdb.c diff has the same stable patch ID as #221, Robert's polished version of Dilip's #217; both actually select FILE_COPY despite the commit subject saying WAL_COPY.

  • Andres Freundsaw the final versionLLM analysis, 100% sureAfter reporting initdb performance and size regressions, he reviewed the follow-up and explicitly approved #221 in #222.
  • Robert Haassaw the final versioncommittercommitted the patch
Simplify a needlessly-complicated regular expression.
30 Mar 2022Robert Haasmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surelanded as postedfrom patchset #2131 of 1 reviewer saw the final version

Exactly matches #213 by stable patch ID, simplifying the invalid-strategy error regular expression in src/bin/scripts/t/020_createdb.pl.

  • Robert Haassaw the final versioncommittercommitted the patch
In 020_createdb.pl, change order of command-line arguments.
29 Mar 2022Robert Haasmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

Andres reported Windows argument-parsing failures in #203; Robert reordered the createdb test arguments and announced the fix in #204, without an attached patch.

  • Andres Freundsaw the final versionLLM analysis, 99% sureHis pre-commit CI report #203 identified the exact failing argument order that the discussion-derived fix corrects.
Add new block-by-block strategy for CREATE DATABASE.
29 Mar 2022Robert Haasmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surelanded as postedfrom patchset #1941 of 11 reviewers saw the final version

Stable patch IDs establish an exact match to Robert's posting #194, which credits Dilip's implementation. In #202, Robert explicitly chooses this version after reverting the unrelated dependency behind #198 and #199.

  • Ashutosh Sharmasaw an earlier versionLLM analysis, 99% sureHe benchmarked and reviewed locking, cleanup, tablespace handling, storage opening, and documentation, including #139, before the later fake-relcache and abort-cleanup changes.
  • Andres Freundsaw an earlier versionLLM analysis, 99% sureHe reviewed buffer and snapshot safety and reported the ICU test failure in #187, before the final storage-cleanup interface revision.
  • John Naylorsaw an earlier versionLLM analysis, 99% sureHe reviewed the six-component implementation in #41, raising cleanup, optimization-comment, and testing issues addressed in subsequent revisions.
  • Greg Nancarrowsaw an earlier versionLLM analysis, 99% sureHis reviews and tests in #43, #45, and #46 exposed buffer-release, version-file, and tablespace/fsync problems.
  • Neha Sharmasaw an earlier versionLLM analysis, 99% sureShe reported tablespace and cancellation crashes and tested the fixes in #58, #63, and #71, before ALTER DATABASE support was removed.
  • Julien Rouhaudsaw an earlier versionLLM analysis, 98% sureHe identified large-template and replication-bandwidth regressions in #2, motivating retention of the original file-copy strategy.
  • Heikki Linnakangassaw an earlier versionLLM analysis, 86% sureHe examined the copying design and large-template regression in #3, proposing alternative enumeration and buffering approaches.
  • Andrew Dunstansaw an earlier versionLLM analysis, 86% sureHe confirmed the importance of large-template copying in #6 and accepted the new default with an alternative strategy in #97.
  • Bruce Momjiansaw an earlier versionLLM analysis, 91% sureHis request for WAL_LOG rather than an ambiguous LOG spelling in #121 was adopted in the subsequent strategy interface.
  • Alvaro Herrerasaw an earlier versionLLM analysis, 97% sureHe proposed the enum-valued STRATEGY interface in #105, replacing the proposed boolean option in later revisions.
  • Robert Haassaw the final versioncommittercommitted the patch
Refactor code for reading and writing relation map files.
17 Mar 2022Robert Haasmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #1651 of 1 reviewer saw the final version

Exactly matches the first component of #165 by stable patch ID, also posted separately by Robert in #164; #166 confirms that only this prerequisite was committed.

  • Ashutosh Sharmasaw the final versionLLM analysis, 99% sureHe reviewed error-level handling and identified redundant shared-map path logic in #143, #148, and #163, followed only by the small cleanup incorporated into the committed component.

Credits across 8 commits

Author
Dilip KumarLLM6
Robert HaasLLM1
Reviewer
Robert Haascommitter5
Tom Lanecommitter1
Committer
Robert Haas7
Tom Lane1

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