Apr 2023

6 commits
Fix assertion failure in heap_vacuum_rel
28 Apr 2023Daniel Gustafssonmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #902 of 2 reviewers saw the final version

Matches the assertion-to-reset change in src/backend/access/heap/vacuumlazy.c from #90, adding the recursive-call explanation requested in #92.

  • Melanie Plagemansaw the final versiontrailerApproved the fix and requested an explanatory comment in #92.
  • Daniel Gustafssonsaw the final versioncommittercommitted the patch
Fix vacuum_cost_delay check for balance calculation.
25 Apr 2023Daniel Gustafssonmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1REL_15_STABLE15.3REL_14_STABLE14.8REL_13_STABLE13.11+2 more
LLM analysis, 100% surelanded as postedfrom patchset #832 of 2 reviewers saw the final version

Exactly applies the single vacuum_cost_delay > 0 to >= 0 correction in table_recheck_autovac() from #83, modulo rebase context. Git confirms this is the development-branch commit, despite the contradictory backport metadata.

  • Melanie Plagemansaw the final versiontrailerVerified the historical bug and requested retaining cost_limit > 0 while changing delay to >= 0 in #78, which #83 implements.
  • Daniel Gustafssonsaw the final versiontrailerInvestigated the original fix in #77 and reported another reading and testing pass before committing in #84.
Fix autovacuum cost debug logging
20 Apr 2023Daniel Gustafssonmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #762 of 2 reviewers saw the final version

Takes only the logging attachment from #76, removing old/new cost comparisons and guarding logging with message_level_is_interesting(DEBUG2). The committer adds an explanatory comment and removes a blank line; the separate balancing fix is excluded.

  • Melanie Plagemansaw the final versionLLM analysis, 99% sureReviewed the missing initial-value logging and the proposed logging frequency in #78, arguing for less frequent output although the committed patch retained Sawada's approach.
  • Daniel Gustafssonsaw the final versioncommittercommitted the patch
Refresh cost-based delay params more frequently in autovacuum
6 Apr 2023Daniel Gustafssonmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #741 of 4 reviewers saw the final version

Commits the third component of #74, implementing frequent autovacuum reloads and worker-local cost balancing across vacuum.c, vacuumparallel.c, vacuumlazy.c, and autovacuum.c. Differences within that component are comments and whitespace, consistent with #75.

  • Masahiko Sawadasaw an earlier versiontrailerReviewed reload safety, failsafe handling, balancing, and logging through #65, before subsequent worker-state and logging logic changes.
  • Daniel Gustafssonsaw the final versiontrailerRequested the final logging lock adjustment in #73 and confirmed reading, testing, and committing #74 in #75.
  • Kyotaro Horiguchisaw an earlier versiontrailerReviewed balancing correctness, shared-state access, failsafe handling, and naming in #32, #37, and #52, before later worker-state and logging revisions.
  • Robert Haassaw an earlier versiontrailerReviewed update frequency and cost-limit calculation in #56 and supported retaining the worker-count design in #68, before subsequent local logic changes.
Separate vacuum cost variables from GUCs
6 Apr 2023Daniel Gustafssonmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #743 of 4 reviewers saw the final version

Commits the second component of #74, separating runtime vacuum_cost_delay/limit from the existing GUC variables and introducing VacuumUpdateCosts(). Its only changed-line difference is capitalizing GUCs in a comment.

  • Masahiko Sawadasaw the final versiontrailerReviewed initialization, symbol visibility, and worker checks in #43 and #48, then said the preparatory patches looked good apart from naming in #65.
  • Daniel Gustafssonsaw the final versiontrailerRequired preservation of exported GUC names in #53 and confirmed the final review, testing, and commit in #75.
  • Kyotaro Horiguchisaw an earlier versiontrailerProposed separating GUC backing variables from runtime cost values in #37, before the dedicated implementation and subsequent initialization changes.
  • Robert Haassaw the final versiontrailerExplicitly praised the understandable preparatory patches in #56, after which this component received naming and comment changes.
Make vacuum failsafe_active globally visible
6 Apr 2023Daniel Gustafssonmaster16beta1REL_19_STABLE16beta1REL_18_STABLE16beta1REL_17_STABLE16beta1REL_16_STABLE16beta1
LLM analysis, 100% surepart of a posted patchsetlight changesfrom patchset #744 of 4 reviewers saw the final version

Commits the complete first component of #74, promoting LVRelState->failsafe_active to VacuumFailsafeActive in vacuum.c and vacuum.h. Its added and removed lines match the attachment exactly; the other two series components landed separately.

  • Masahiko Sawadasaw the final versiontrailerReviewed failsafe state and reset requirements repeatedly, then said the preparatory patches looked good in #65.
  • Daniel Gustafssonsaw the final versiontrailerReviewed global-state documentation and symbol visibility in #26 and #49, then read, tested, and committed the final series in #75.
  • Kyotaro Horiguchisaw the final versiontrailerChallenged the ternary-state design and reviewed the final global's ownership comment in #52, which received only explanatory edits afterward.
  • Robert Haassaw the final versiontrailerPraised the preparatory patches as easy to understand in #56, and this component remained unchanged through #74.

Credits across 6 commits

Author
Masahiko Sawada3
Melanie Plageman3
Reviewer
Daniel Gustafssoncommitter6
Kyotaro Horiguchi3
Masahiko Sawada3
Robert Haas3
Melanie Plageman2
Reported by
John Naylor1
Committer
Daniel Gustafsson6