Aug 2023

1 commit
Rename hook functions for debug_io_direct to match variable name.
24 Aug 2023Thomas Munromaster17beta1REL_19_STABLE17beta1REL_18_STABLE17beta1REL_17_STABLE17beta1REL_16_STABLE16rc1
LLM analysis, 100% surelanded as postedfrom patchset #1111 of 1 reviewer saw the final version

The three-file diff matches #111, including all eight added and eight removed lines renaming hooks, the backing variable, and header declarations. Thomas Munro accepted it in #113.

  • Thomas Munrosaw the final versioncommittercommitted the patch

May 2023

1 commit
Rename io_direct to debug_io_direct.
14 May 2023Thomas Munromaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #1071 of 3 reviewers saw the final version

Matches the GUC, documentation, and test rename in #107, with four additional error-string substitutions in src/backend/storage/file/fd.c. Thomas Munro explicitly describes that finishing change in #110.

  • Noah Mischsaw the final versiontrailerNoah proposed the debug_ prefix in #102 and approved the attached rename in #109, after which only error strings changed.
  • Justin Pryzbysaw an earlier versiontrailerJustin rejected the warning in #105 and advocated the replacement rename, while the commit explicitly credits review of the idea rather than the later patch.
  • Tom Lanesaw an earlier versiontrailerTom endorsed replacing the warning with a rename in #106, while the commit explicitly limits his credit to the idea rather than the later patch.

Apr 2023

4 commits
Skip the 004_io_direct.pl test if a pre-flight check fails.
13 Apr 2023Thomas Munromaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #582 of 2 reviewers saw the final version

Retains the filesystem pre-flight test from #58 but replaces the eval-based constant lookup and redundant import with defined &O_DIRECT. Comments, skip wording, and the commit explanation were also refined following #61 and #64.

  • Dagfinn Ilmari MannsÃ¥kersaw the final versiontrailerIlmari verified Perl compatibility and supplied the adopted defined &O_DIRECT simplification in #61, explaining its behavior further in #64.
  • Andrew Dunstansaw the final versiontrailerAndrew reviewed the Perl probe in #60 and discussed its constant-export behavior through #65, with only test simplifications following.
Use higher wal_level for 004_io_direct.pl.
8 Apr 2023Thomas Munromaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 100% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

Adds only wal_level = replica to src/test/modules/test_misc/t/004_io_direct.pl, exactly the workaround proposed in #20 and accepted in #21. The attachment in #19 instead implements pin limiting and does not contain this workaround.

  • Andres Freundsaw the final versionLLM analysis, 100% sureAndres diagnosed the buffer exhaustion in #17 and proposed the exact wal_level workaround in #20 before it was committed.
Add io_direct setting (developer-only).
8 Apr 2023Thomas Munromaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 100% suremodified before commitmoderate changesfrom patchset #130 of 2 reviewers saw the final version

Derives from the io_direct member of #13, adding BLCKSZ/XLOG_BLCKSZ rejection checks and restructuring check_io_direct() in src/backend/storage/file/fd.c. It also removes the unused _mdfd_open_flags() argument and edits documentation and formatting, as described in #14.

  • Justin Pryzbysaw an earlier versiontrailerJustin proposed consolidating the booleans and enabling CI exercise in #2, before the GUC implementation and validation were substantially revised.
  • Bharath Rupireddysaw an earlier versiontrailerBharath found GUC mutation and allocation-leak problems and supplied fixes in #12, which were incorporated before the later small-block validation changes.
Introduce PG_IO_ALIGN_SIZE and align all I/O buffers.
8 Apr 2023Thomas Munromaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 100% suremodified before commitmoderate changesfrom patchset #130 of 1 reviewer saw the final version

Matches the 26-file alignment member of #13 with three md.c assertions additionally guarded by PG_IO_ALIGN_SIZE <= BLCKSZ and the proposed XLOGShmemInit() alignment change omitted. These small-block adjustments and explanatory edits are documented in #14.

  • Justin Pryzbysaw an earlier versiontrailerJustin reported the BufFile alignment-sanitizer failure in #2, before the separate PGIOAlignedBlock design and subsequent small-block handling were introduced.

Credits across 6 commits

Author
Thomas Munro2 LLM4
Andres Freund2
Bharath Rupireddy1
Peter Eisentraut1
Reviewer
Justin Pryzby3
Andrew Dunstan1
Bharath Rupireddy1
Dagfinn Ilmari Mannsåker1
Noah Misch1
Thomas Munrocommitter1
Tom Lane1
Reported by
Andres Freund1
Christoph Berg1
Committer
Thomas Munro6

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