model-writtenrun 9539 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Apr 2018 · 2 entries
- Fujii Masao proposed a per-table switch to prevent VACUUM truncation after observing response-time spikes from scanning large shared buffers under AccessExclusiveLock and unnecessary cycles of truncation and extension #1.
- Tom Lane preferred fixing buffer management instead of adding a permanent user-visible workaround, while Alvaro Herrera and Simon Riggs argued that production disruption and the exclusive lock itself justified the option; Andres Freund explained that the larger buffer-management redesign was unlikely to be ready for PostgreSQL 12 #2 #4 #6 #13.
Nov 2018 · 1 entry
- The proposal revived in November when Laurenz Albe highlighted standby replication conflicts and Alvaro Herrera reported continuing production problems without an alternative solution #24 #25.
Jan - May 2019 · 13 entries
- Tsunakawa, Takayuki supplied the first shrink_enabled implementation, immediately restored an omitted header change, and received successful build and behavioral test results from Kirk Jamison #29 #30 #31.
- Review by Julien Rouhaud and Michael Paquier led to the vacuum_shrink_enabled name and regression coverage, followed by moving the guard into should_attempt_truncation(), documenting autovacuum behavior, and replacing block-size-dependent checks #39 #40 #41 #45 #51.
- Masahiko Sawada reproduced failures under concurrent regression tests and suggested rolling back insertions; Tsunakawa adopted failed insertions so the test could remain parallel without depending on other transactions finishing #47 #49 #51.
- The discussion expanded to command-level and global controls: Robert Haas opposed a global GUC as an attractive nuisance, Michael Paquier favored a VACUUM override, and Tsunakawa kept those additions separate from the reloption patch #50 #53 #51.
- Robert Haas proposed consistent positive names and both command and relation options for truncation and index cleanup, while debate continued over persistent controls and fixing the underlying locking problem; the index-cleanup implementation proceeded separately #63 #66 #67 #68 #82 #84.
- Fujii requested TOAST support, which Tsunakawa added, and reviews by Sawada and Rouhaud then improved documentation formatting and clarified that truncation is only attempted #90 #91 #92 #93 #94.
- Fujii took over preparation for commit, rebasing the patch, changing shrink to truncate, simplifying documentation, and extending TOAST tests; he proposed landing the reloption before adding the command option #103.
- Near feature freeze, Sawada and Tom Lane challenged the vacuum_truncate_enabled spelling, Andres Freund emphasized the need to control autovacuum to avoid standby cancellations, and Fujii settled on vacuum_truncate #104 #109 #111 #114.
- Fujii committed Tsunakawa's reloption as 119dcfad98 from #103, with the shorter name and light documentation and formatting edits, then confirmed it was committed #116 #118.
- After Sawada raised the missing command option as an open item, Fujii posted a separate patch implementing VACUUM (TRUNCATE true/false), overriding the reloption when explicitly specified #119 #120.
- Julien tested the command-option patch and requested a documentation wording change; Kyotaro Horiguchi's suggestion to use SKIP_TRUNCATE was dropped after Fujii explained consistency with INDEX_CLEANUP #122 #125 #126 #127.
- Following several weeks as an open item, Sawada completed his review and Fujii committed #120 as b84dbc8eb8, preserving its code and tests and applying Julien's documentation correction #128 #129 #130 #131.
- The thread ended with both controls landed and Bruce Momjian updating the release notes; the broader buffer-management and locking redesigns remained outside these patches #132.