model-writtenrun 3816 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Mar - Apr 2021 · 4 entries
- Andrei Zubkov proposed per-statement entry timestamps so sampling tools could detect eviction and reinsertion even when counters had already surpassed their previous sampled values #1.
- Hayato Kuroda questioned timestamp overhead, while Andrei explained that timestamps were acquired only when allocating entries; timestamps were preferred over storing a deallocation counter because they also identify the collection interval #2 #3 #6.
- Julien Rouhaud explained PoWA's counter-decrease heuristic, and Andrei argued that explicit timestamps remove uncertainty for less frequent sampling #4 #5.
- Yuki Seino requested an extension-version bump, Andrei updated the patch for extension 1.10, and Chengxi Sun reported successful feature and installation testing #8 #14 #15.
Oct 2021 · 1 entry
- After an autumn rebase, Andrei identified that min/max values could not be reconstructed from snapshot differences, and Anton Melnikov supported separate resettable extrema while reporting successful tests #18 #19 #20.
Dec 2021 - Apr 2022 · 8 entries
- Andrei expanded the patch with four local min/max counters, a separate reset function, a reset timestamp, and the renamed stats_since field #21.
- Anton recommended auxiliary terminology and a separate view, while Andres Freund reported an assertion failure; Andrei incorporated the view and naming changes, fixed the failure, and added tests #23 #24 #25.
- Anton reviewed initialization and duplicated reset code, but Julien challenged the need to retain two sets of extrema; discussion considered NULL, NaN, and deferred-reset representations before settling on zero values after reset #26 #27 #29 #31 #33.
- Andrei removed the auxiliary counters and view, making existing min/max statistics independently resettable and recording their collection start in minmax_stats_since #33.
- Julien requested a shared reset implementation and one SQL reset function, leading Andrei to add the minmax_only argument and old-extension regression coverage #34 #35.
- A compiler warning led to unconditional timestamp acquisition, and subsequent discussion led to returning the precise reset timestamp so sampling tools could identify their own resets #36 #38 #41 #43.
- Andres objected to repeated reset code; Andrei first extracted a helper and then adopted Julien's suggested macro, while Greg Stark and Tom Lane identified the collation-dependent test failure #39 #43 #44 #46 #47 #48.
- The following revisions fixed test ordering, simplified the macro, passed its entry explicitly, and polished documentation; Julien declared the resulting patch ready for a committer #49 #52 #56 #57.
Jan - Mar 2023 · 4 entries
- After roughly nine months without a landing, Tomas Vondra questioned the second timestamp and reset complexity; Andrei explained that counters and means can be reconstructed without resets, while extrema need independently tracked resets to detect interference between samplers #59 #60.
- Andrei refreshed the patch for extension 1.11 and the reorganized regression suite, repairing Meson registration omissions and rebasing again around concurrent utility-query work #61 #62 #63 #65 #67 #68.
- Michael Paquier proposed extracting the widespread IS NOT NULL test conversion and removing unnecessary extension-membership commands, producing a preparatory test patch followed by the feature patch #69 #70.
- Sergei Kornilov caught missing privilege revocation for the new reset-function signature and requested explicit documentation of the four affected extrema; Andrei incorporated both corrections #72 #73.
Oct - Nov 2023 · 7 entries
- After another long gap, Andrei rebased onto the existing extension 1.11 changes and raised a question about its unchanged statistics-file header; the supplied thread shows no explicit resolution of that question #75.
- A discussion about removing inactive reviewers from Commitfest entries distinguished current work allocation from preserving historical review credit in commit messages #76 #77 #78 #79.
- Andrei Lepikhov questioned the second timestamp again, and Andrei Zubkov explained its role in detecting resets by competing sampling tools; Alena Rybakina reviewed top-level and nested-entry handling and prompted clearer reset comments #80 #81 #82 #83 #84 #85 #86.
- Following a final rebase and level-tracking test repair, Alexander Korotkov reviewed and republished Andrei's series with revised commit messages, and Julien explicitly endorsed it #87 #88 #89 #90.
- Alexander committed the preparatory test conversion from #89 as 6ab1dbd26b, fixing the commit-message typo identified by Andrei in #91 and acknowledged in #92.
- Alexander committed the feature patch from #89 as dc9f8a7983, with only formatting differences, delivering both timestamps, independently resettable extrema, and the reset timestamp return value.
- Both patches had landed by the last message, where Alexander Lakhin reported the remaining 'minimun' typo; no correction to that report is shown in the supplied thread #93.