Feb 2022

4 commits
Split xlog.c into xlog.c and xlogrecovery.c.
16 Feb 2022Heikki Linnakangasmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% suremodified before commitmoderate changesfrom patchset #210 of 3 reviewers saw the final version

Matches the main split in #21, including the folded ApplyWalRecord and xlogrecovery_redo work. The committed version additionally clears backupStartPoint and backupEndPoint after ReachedEndOfBackup in xlogrecovery.c, adds a storage/fd.h include, and adjusts comments.

  • Andres Freundsaw an earlier versiontrailerIn #2 and #6, he requested incremental patches, per-record extraction, and changes to recovery shutdown structure; interface changes and recovery logic fixes followed.
  • Kyotaro Horiguchisaw an earlier versiontrailerIn #12, he reviewed includes, naming, recovery-state ownership, and ApplyWalRecord, before later redo restructuring and timeline and backup-state fixes.
  • Robert Haassaw an earlier versiontrailerIn #14, he analyzed the startup ordering and recovery boundaries underpinning the split, before subsequent interface and recovery logic changes.
Move code around in StartupXLOG().
16 Feb 2022Heikki Linnakangasmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surelanded as postedfrom patchset #210 of 3 reviewers saw the final version

The StartupXLOG reordering patch in #21 and the commit have identical stable patch IDs, including the function-boundary comments and performedWalRecovery condition.

  • Andres Freundsaw an earlier versiontrailerIn #2 and #6, he requested separation of movement from behavior changes and questioned exitArchiveRecovery restructuring, before further startup logic revisions.
  • Kyotaro Horiguchisaw an earlier versiontrailerIn #12, he examined PublishStartupProcessInformation, unlogged-relation initialization, and recoveryStopReason placement, before the later performedWalRecovery condition was introduced.
  • Robert Haassaw an earlier versiontrailerIn #14, he reviewed individual StartupXLOG reorderings and questioned control-file timing and archive-recovery assertions, before the later performedWalRecovery condition was introduced.
Refactor setting XLP_FIRST_IS_OVERWRITE_CONTRECORD.
16 Feb 2022Heikki Linnakangasmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surelanded as postedfrom patchset #210 of 1 reviewer saw the final version

The overwrite-contrecord prerequisite in #21 has the same stable patch ID as the commit. Both set the page-header flag before inserting the record and verify its insertion position.

  • Robert Haassaw an earlier versiontrailerIn #19, he questioned insertion-lock safety, flushing, and parameter shadowing, prompting substantive flag-ordering, locking, and validation changes in #20.
Add test case for an archive recovery corner case.
14 Feb 2022Heikki Linnakangasmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1Also in: TAP test to cover "EndOfLogTLI != replayTLI" case (not this thread)
LLM analysis, 100% suremodified before commitlight changesfrom patchset #210 of 2 reviewers saw the final version

The commit matches src/test/recovery/t/028_pitr_timelines.pl in #21, removing an initial blank line and replacing the fixed test count with done_testing(). I disagree with attributing its origin to discussion in thread #45144: that thread explicitly selected Heikki's alternative, and this thread's matching attachment dated January 25 is newer than its last attachment dated January 20.

  • Amul Sulno review foundtrailerThe commit credits his review and separate test, but this thread contains no review activity by him.
  • Michael Paquierno review foundtrailerIn #22, he points to the overlapping submission in thread #45144 without assessing or testing the attached patch.

Jul 2021

4 commits
Move InRecovery and standbyState global vars to xlogutils.c.
31 Jul 2021Heikki Linnakangasmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #51 of 1 reviewer saw the final version

Matches the InRecovery and standbyState relocation in #5, with only a wording adjustment in the xlogutils.c comment and a rewritten commit explanation.

  • Andres Freundsaw the final versiontrailerIn #6, he reviewed the preliminary cleanups and discussed cross-translation-unit access costs; only explanatory wording changed before commitment.
Extract code to describe recovery stop reason to a function.
31 Jul 2021Heikki Linnakangasmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surelanded as postedfrom patchset #51 of 1 reviewer saw the final version

The getRecoveryStopReason extraction in src/backend/access/transam/xlog.c has the same stable patch ID as the corresponding patch in #5.

  • Andres Freundsaw the final versiontrailerIn #6, he endorsed the preliminary patches and suggested a future switch-based implementation for recovery-target handling.
Remove unnecessary 'restoredFromArchive' global variable.
31 Jul 2021Heikki Linnakangasmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surelanded as postedfrom patchset #51 of 1 reviewer saw the final version

The restoredFromArchive removal has the same stable patch ID as the corresponding patch in #5, removing the unused global and its assignments from xlog.c.

  • Andres Freundsaw the final versiontrailerIn #6, he endorsed the preliminary cleanups and recommended applying them immediately, which Heikki confirmed in #7.
Don't use O_SYNC or similar when opening signal file to fsync it.
31 Jul 2021Heikki Linnakangasmaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surelanded as postedfrom patchset #51 of 1 reviewer saw the final version

The signal-file opening cleanup has the same stable patch ID as the corresponding patch in #5, removing get_sync_bit(sync_method) from both BasicOpenFilePerm calls.

  • Andres Freundsaw the final versiontrailerIn #6, he endorsed this preliminary cleanup and recommended committing the simple initial patches.

Credits across 8 commits

Author
Heikki LinnakangasLLM8
Reviewer
Andres Freund6
Robert Haas3
Kyotaro Horiguchi2
Amul Sul1
Michael Paquier1
Committer
Heikki Linnakangas8

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