model-writtenrun 9147 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Aug - Oct 2017 · 4 entries
- Tatsuro Yamada proposed pg_stat_progress_cluster in #1, exposing CLUSTER phases and tuple counters while asking whether VACUUM FULL should share the view.
- Thomas Munro and Masahiko Sawada found incorrect regression expectations in #2 and #4; Yamada corrected them in #5, and Sawada confirmed that the tests passed in #6.
- Michael Paquier recommended one view with an explicit command indicator in #8, leading Yamada to add VACUUM FULL support in #10 while retaining the pg_stat_progress_cluster name.
- Robert Haas challenged the estimated total-tuple counter and the usefulness of coarse sorting phases in #13 and #15, then proposed reporting actual blocks read and tuples entering and leaving the sort in #21.
Nov 2017 · 1 entry
- Antonin Houska suggested planner-cost-based progress estimates in #22, but Tom Lane and Robert Haas argued in #23 and #25 that progress reporting should expose measured facts rather than potentially misleading predictions.
Aug 2018 · 1 entry
- After a long gap, Yamada revived the patch in #31, rebasing it, fixing documentation, and adopting Jeff Janes's cluster_index_relid naming suggestion from #19.
Nov - Dec 2018 · 1 entry
- Alvaro Herrera and Peter Geoghegan explored callbacks and trace_sort-style instrumentation in #37 and #38, while Yamada's December submission #42 remained a rebase without that additional instrumentation.
Feb - Mar 2019 · 7 entries
- Haas's detailed review #43 drove the simplification in #44: scan methods became distinct phases, estimated totals stopped being populated, and progress lifecycle handling moved into cluster_rel().
- Yamada followed further review #47 with separate code and documentation patches in #48, removing obsolete fields and duplicate lifecycle calls; extending the feature to rewriting ALTER TABLE operations remained outside the immediate scope.
- Haas explained the abstraction problem with CLUSTER-specific updates inside generic sorting code in #54, and Yamada explicitly deferred sort-internal progress instrumentation in #56.
- The revisions in #57 and #60 replaced the vacuumed-tuple counter with heap-block counters and an index rebuild count, then updated the accompanying documentation.
- After the next rebase, Kyotaro Horiguchi proposed a more compact, phase-dependent view and questioned materialized-view interactions in #64; the subsequent patch retained separate named counters.
- Haas supplied an integrated revision in #65, fixing documentation and regression failures, ending progress on early returns, and preserving counters across phases, while requesting a separate counter for tuples written.
- Yamada added heap_tuples_written in #67 and resent the identical patch with test results in #68; Haas landed it as 6f97457e0d with documentation corrections and a catalog-version bump, announcing the commit in #69.
Aug - Sep 2019 · 9 entries
- In August, Herrera reported that index_rebuild_count never advanced and CLUSTER disappeared from the progress view before completion in #72; Yamada bisected the regression to later REINDEX progress instrumentation in #74.
- Yamada considered restarting, saving and restoring, or storing multiple progress records in #76, and Sawada implemented a fixed-size stack of nested progress records in #77.
- Paquier objected to expanding shared backend state and redesigning progress reporting near the PostgreSQL 12 release in #78; Sawada agreed in #79 that nested reporting could instead be treated as a future feature.
- While examining that problem, Sawada found an independent AND-versus-OR error in progress cleanup and attached #80; Paquier committed it with a small guard restructuring as ae060a52b2 and confirmed backpatching through PostgreSQL 9.6 in #82.
- Haas argued in #84 that callers should control when progress updates are appropriate, while Paquier detailed shared-code-path collisions in #85; the discussion considered reverting affected features but produced no revert.
- Herrera proposed conditional REINDEX reporting in #90 and implemented it directly as 6212276e43, using REINDEXOPT_REPORT_PROGRESS to preserve CLUSTER state and announcing the fix in #97.
- Unaware that the targeted fix had already landed, Yamada posted a broader command-type-aware progress API in #99 and revised its guards in #102 following Paquier's feedback #100.
- Yamada tested the committed fix successfully in #107 and deferred his API redesign to PostgreSQL 13; Paquier's review #108 identified a remaining abort-handling flaw in that redesign that could leave stale progress state and prevent later commands from starting reports.
- At the thread's end, Paquier still questioned shared updates and lifecycle placement in #111, but Herrera explained in #112 that the disputed counter did not collide with CREATE INDEX fields and that moving initialization upward lacked the required table OID; the feature and both fixes remained committed, with richer nested progress reporting and the API redesign left for future work.