Mar - Apr 2020

8 commits
Add perl2host call missing from a new test file.
4 Apr 2020Noah Mischmaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1
LLM analysis, 100% surewritten from the discussion, no matching patch

Adds the missing TestLib::perl2host conversion in src/test/recovery/t/018_wal_optimize.pl after the buildfarm failure reported in #242; no indexed attachment contains this fix.

Skip WAL for new relfilenodes, under wal_level=minimal.
4 Apr 2020Noah Mischmaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1
LLM analysis, 100% suremodified before commitmoderate changesfrom patchset #2313 of 9 reviewers saw the final version

Combines the four patches in #231, with the changes described in #232: pending-sync transfer moves from execParallel.c to parallel.c, serialization is refactored, and workers stop performing commit-time sync/FPI work. The commit also bumps XLOG_PAGE_MAGIC.

  • Robert Haassaw an earlier versionLLM analysis, 99% sureReviewed WAL-check costs, UPDATE semantics, locking, and truncation durability in #41, #129, #168, and #183, before substantial later revisions.
  • Andrew Dunstansaw an earlier versionLLM analysis, 100% sureReviewed the broad patch's recovery code in #114, prompting removal of its heap_xlog_update leniency in #115.
  • Andres Freundsaw the final versionLLM analysis, 99% sureExplicitly credited as a reviewer of earlier designs by c6b92041d3; no qualifying attached-patch review was found, so final denotes the commit-credit exception.
  • Alvaro Herrerasaw an earlier versionLLM analysis, 100% sureSupplied comment and naming improvements and challenged relcache-state durability in #71, leading to a different storage design.
  • Tom Lanesaw an earlier versionLLM analysis, 100% sureExplained in #74 why strong locks cannot prevent relcache invalidation, prompting restoration of authoritative state outside relcache.
  • Fujii Masaosaw the final versionLLM analysis, 99% sureExplicitly credited as a reviewer of earlier designs by c6b92041d3; the visible contributions precede attachments, so final denotes the commit-credit exception.
  • Simon Riggssaw an earlier versionLLM analysis, 99% sureReviewed Heikki's replacement implementation in #40, questioning repeated pending-sync lookups and hint-bit costs.
  • Thomas Munrosaw an earlier versionLLM analysis, 99% sureReported a concrete compile failure in #85 and a patched regression failure in #212, both corrected in subsequent versions.
  • Noah Mischsaw the final versioncommittercommitted the patch
Revert "Skip WAL for new relfilenodes, under wal_level=minimal."
22 Mar 2020Noah Mischmaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1+5 morereverts 1 commit here
LLM analysis, 100% surewritten from the discussion, no matching patch

Reverts cb2fd7eac2 as announced in #230, because parallel workers lacked WAL-skipping state and the recovery test assumed LP64; the attachment there is diagnostic instrumentation, not a revert patch.

Back-patch log_newpage_range().
21 Mar 2020Noah MischREL_11_STABLE11.8REL_10_STABLE10.13REL9_6_STABLE9.6.18REL9_5_STABLE9.5.22
LLM analysis, 99% surewritten from the discussion, no matching patch

The backport need arises in #225, and #227 chooses XLOG_FPI_MULTI so old readers fail instead of silently replaying one page. No indexed patchset contains this older-branch log_newpage_range implementation and record-type adaptation.

In log_newpage_range(), heed forkNum and page_std arguments.
21 Mar 2020Noah Mischmaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1+1 more
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #223

Exactly matches the standalone log_newpage_range argument fix in #223: ReadBufferExtended honors forkNum, and REGBUF_STANDARD depends on page_std. This is one unchanged member of the series, authored by Noah in its header.

During heap rebuild, lock any TOAST index until end of transaction.
21 Mar 2020Noah Mischmaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1+5 more
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #223

Exactly matches the standalone TOAST-lock member of #223: toast_get_valid_index retains locks, and finish_heap_swap uses NoLock. It selects one unchanged member of the larger series; that member's header identifies Noah as its author.

Skip WAL for new relfilenodes, under wal_level=minimal.
21 Mar 2020Noah Mischmaster13beta1REL_19_STABLE13beta1REL_18_STABLE13beta1REL_17_STABLE13beta1REL_16_STABLE13beta1REL_15_STABLE13beta1REL_14_STABLE13beta1REL_13_STABLE13beta1+5 more
LLM analysis, 100% surepart of a posted patchsetmoderate changesfrom patchset #2233 of 9 reviewers saw the final version

Commits the main WAL-skipping member of Noah's forwarded series #223, explicitly crediting Kyotaro as its author. Subsequent edits in #225 and #227 move index-state restoration to ATExecAddIndex, repair recovery tests, and update GUC integration; standalone members land separately.

  • Robert Haassaw an earlier versionLLM analysis, 99% sureReviewed WAL-check costs, UPDATE semantics, locking, and truncation durability in #41, #129, #168, and #183, before substantial later revisions.
  • Andrew Dunstansaw an earlier versionLLM analysis, 100% sureReviewed recovery behavior in #114, after which #115 removed the questioned heap_xlog_update changes.
  • Andres Freundsaw the final versionLLM analysis, 99% sureExplicitly credited as a reviewer of earlier designs by cb2fd7eac2; no qualifying attached-patch review was found, so final denotes the commit-credit exception.
  • Alvaro Herrerasaw an earlier versionLLM analysis, 100% sureReviewed and annotated the relcache implementation in #71, raising the state-lifetime problem that changed its architecture.
  • Tom Lanesaw an earlier versionLLM analysis, 100% sureRejected the implementation's reliance on locks to preserve relcache state in #74, leading to an authoritative backend hash.
  • Fujii Masaosaw the final versionLLM analysis, 99% sureExplicitly credited as a reviewer of earlier designs by cb2fd7eac2; the visible contributions precede attachments, so final denotes the commit-credit exception.
  • Simon Riggssaw an earlier versionLLM analysis, 99% sureExamined Heikki's replacement patch in #40 and proposed cheaper pending-sync checks while questioning hint-bit overhead.
  • Thomas Munrosaw an earlier versionLLM analysis, 99% sureIdentified the missing heap_insert change in #85 and reported the patched regression failure in #212 that Noah fixed in #215.
  • Noah Mischsaw the final versioncommittercommitted the patch
Back-patch src/test/recovery and PostgresNode from 9.6 to 9.5.
19 Mar 2020Noah MischREL9_5_STABLE9.5.22
LLM analysis, 100% surewritten from the discussion, no matching patch

Kyotaro identified missing back-branch TAP infrastructure in #225; Noah chose the broader recovery-suite backport in #227. The commit expressly says it did not use Kyotaro's patch, omits the newer synchronous-replication test, and substitutes hot_standby for replica.

Credits across 8 commits

Author
Kyotaro HoriguchiLLM2
Noah MischLLM2
Reviewer
Noah Mischcommitter2
Committer
Noah Misch8

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