model-writtenrun 9778 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jun - Jul 2018 · 6 entries
- Haribabu Kommi proposed a separate pg_stat_statements_reset_query() function to observe fresh statistics for an optimized query without clearing every statement's history in #1.
- Sergei Kornilov identified the current-user limitation in #4, Euler Taveira advocated extending the existing reset function in #5, Robert Haas cautioned against confusing overloads in #7, and Rui DeSousa proposed filtering by user, database and query in #8.
- Haribabu adopted one reset function with three optional zero-valued parameters in #12, supporting an exact hash lookup when all keys are supplied and filtered removal otherwise.
- Sergei's reviews in #13 and #16 led to NULL defaults, a removed-entry return count, a backward-compatible C entry point, removal of missing-entry logging, and corrected NULL handling and documentation in #15 and #17; he subsequently reported successful testing in #18.
- Fujii Masao raised both the possibility of a valid zero utility-query hash and the inappropriate pg_read_all_stats reset privilege in #19, prompting a separate permissions patch in #27 that Haribabu corrected to revoke directly from that role in #28.
- Haribabu bundled the permissions prerequisite with the revised feature in #30, reserving zero for wildcard matching by replacing a zero utility-query hash with two and using UINT64CONST for the comparison.
Sep 2018 - Jan 2019 · 10 entries
- After the summer gap, Amit Kapila requested separate handling of the permissions issue in #31, Haribabu moved it to another thread in #32, and the feature resumed here as a rebased standalone attachment in #33.
- Michael Paquier's reviews in #34 and #36 led to corrected regression-role names, centralized initialization checks, simpler cleanup control flow, documentation markup, and masking of the variable initial reset count in #35 and #37.
- Amit questioned the compatibility cost and practical value of returning a count in #41 and #43; after an interim test simplification in #42, Haribabu restored the void return type in #50.
- Amit and Haribabu established that partial resets could leave query text for later garbage collection in #51 and #52, and Amit supplied simplified predicates and a separate unconditional-removal path in #53.
- Additional tests in #54 exposed an unintended broad reset when a query-ID subquery returned NULL, which Haribabu explained in #59; the test corrections in #57 and Amit's finishing changes in #60 initially brought the patch close to commitment.
- Alvaro Herrera halted that progression by warning that NULL-as-wildcard could erase unrelated production statistics in #61 and proposing separate strict functions in #63, while Michael suggested current-user and current-database defaults in #65.
- Amit proposed retaining one function but making it strict with explicit wildcard sentinels in #67, and discussion settled on zero for all three parameters, with Alvaro accepting that approach in #85 and Michael verifying its compatibility with utility-query hashing in #87 and #89.
- Haribabu implemented zero defaults and STRICT behavior with a NULL-input regression test in #91; Robert Haas subsequently preferred NULL wildcards in #93, but Peter Eisentraut's concrete missing-database example in #95 reinforced the chosen protection against unintended broad resets.
- At Amit's request in #98, Haribabu added documentation examples in #101 and refined their formatting in #103, after which Amit corrected the commit message's default-value description and posted the final attachment in #104.
- Amit committed #104 unchanged as 43cbedab8f and confirmed the push in the thread's final message #106, completing the selective-reset feature while the separate permissions work had already moved elsewhere.