model-writtenrun 7524 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Mar 2019 · 3 entries
- Yun Li requested a query identifier in pg_stat_activity to help correlate similar queries and their performance in #1.
- Tom Lane objected to imposing the cost and unstable semantics of pg_stat_statements hashing on core in #2 and #4, while Robert Haas argued for shared computation to avoid duplicated extension implementations in #3.
- Julien Rouhaud separated exposing Query.queryId from choosing its computation algorithm in #5, then posted an atomic-PGPROC implementation for pg_stat_activity and log_line_prefix in #12.
Jun - Sep 2019 · 3 entries
- Peter Geoghegan proposed showing identifiers in EXPLAIN in #24, but Julien initially deferred that work because the underlying statement bypassed the relevant hook and complicated normalization and accounting, as explained in #58.
- Robert Haas criticized the many atomic writes in #29, and Andres Freund explained that nested identifiers would disagree with top-level query text in #34 and #35; Julien consequently replaced that design with PgBackendStatus storage, top-level reporting, and parallel-worker propagation in #39.
- Michael Paquier's review in #45 exposed ambiguity around zero identifiers, reset behavior, and utility statements; Julien resolved these cases in #56 by computing utility identifiers earlier, preventing duplicate accounting, and reporting prepared executions from ExecutorStart.
Feb - Apr 2020 · 1 entry
- Repeated rebases kept the proposal current through 2020, from #57 through #68, while Tatsuro Yamada reported successful installcheck-world testing in #62.
Jul - Aug 2020 · 1 entry
- Atsushi Torikoshi found that log_statement could print a zero identifier despite identifiers appearing elsewhere in #66, and Julien explained in #67 that statement logging occurs before parse analysis makes the identifier available.
Oct 2020 · 4 entries
- Bruce Momjian reopened the design in October 2020, arguing that identifiers should work without loading pg_stat_statements in #69 and #77; Tom reiterated concerns about expectations of cross-version and cross-platform stability in #70.
- The discussion converged on a built-in algorithm with extension alternatives in #82 and #83, while Tom rejected an algorithm-version number because hashes already change across major releases in #85.
- Julien implemented that agreement as three patches in #87: core jumbling controlled by a GUC, activity and logging exposure, and verbose EXPLAIN output, with JumbleState passed to extensions through post_parse_analyze_hook.
- Julien measured roughly two percent overhead for a short query in #95, prompting Bruce to retain default-off computation in #96; discussion of per-session control and extension algorithms led to expanded documentation in #106.
Jan - May 2021 · 14 entries
- After a development gap, Julien rebased the series in #107 and #108; Tatsuro's attached regression failure in #109 proved to be a test-environment mistake, and he reported all tests passing in #110.
- Julien fixed EXPLAIN test normalization for negative identifiers in #112, then addressed renewed concerns about extension freedom by documenting the built-in computation switch and requiring providers not to overwrite an existing identifier in #138, following the agreement in #129 and #133.
- Bruce took responsibility for committing and rewrote the documentation in #139; Julien incorporated Zhihong's copyright correction and Alvaro's naming feedback into #152.
- Alvaro questioned the executor's dependence on pgstat in #153, and Julien explained in #154 that pgstat preserves the top-level identifier while executor state may describe a nested query; the resulting simplification retrieved the identifier directly inside ExecSerializePlan in #160.
- Julien rebased onto the backend-status file split in #166, then incorporated Nitin's condition and concurrency-documentation review from #168 and #172 into the final submission #174.
- Bruce landed the core-jumbling component of #174 as 5fd9dfa5f5 and combined its activity, logging, and EXPLAIN components into 4f0b0966c8, preserving the implementation and adding the required catalog-version bump.
- Justin Pryzby reported missing CSV output in #176, while Tom reported buildfarm failures in #177; Julien addressed both CSV support and JIT-dependent EXPLAIN test output in #179.
- Bruce applied the EXPLAIN repair and reposted the CSV remainder in #181; Tom corrected portable integer formatting in #182, and Julien caught misplaced separators and newlines in #184, producing #185, which landed unchanged as f57a2f5e03.
- Julien demonstrated that a third-party module could supply a different identifier algorithm while still using pg_stat_statements and the new infrastructure in #191.
- Thomas Munro identified an incorrect Datum conversion on 32-bit systems in #192; Julien bundled its correction with parallel-worker accounting protection in #193, incorporated Amit Kapila's redundant-reporting observation from #194, and Bruce confirmed application of the corrected series #196 in #197.
- The naming discussion settled on query_id for new interfaces while preserving pg_stat_statements.queryid for compatibility in #190 and #202; Bruce's cleanup #203 became #206 after Julien and Alvaro reviewed its argument and structure naming.
- The naming cleanup #206 landed as 9660834dd8, with a catalog-version bump added to the posted changes.
- Fujii Masao reported that the pg_stat_statements key documentation omitted toplevel in #209; Julien supplied #210, which Magnus Hagander committed as 7531fcb1fc after correcting a parameter name and rewriting the explanatory parenthesis.
- Fujii immediately identified that the rewritten parenthesis described the boolean behavior incorrectly in #213 and supplied wording alternatives in #215; Magnus applied the discussion-derived correction as f9e6d00df0, confirming it in #216.
Jul 2021 · 2 entries
- Peter Eisentraut found the remaining description of distinct keys ambiguous in #217; Magnus proposed explicitly describing combinations and top-level status in #219, leading to b4deefc39b.
- At the thread's final message #222, Magnus announced that last wording correction; the feature and attached follow-up work had landed, with no outstanding attached patch series identified.