model-writtenrun 4152 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Feb - Apr 2022 · 8 entries
- Vignesh C proposed filtering changes carrying replication origins to prevent endless two-way replication in #1, while identifying initial table copy as a separate problem because heap tuples do not retain equivalent origin information.
- Hayato Kuroda tested the initial patch and requested ALTER support in #2, leading Vignesh to add alteration support, documentation and tests in #5.
- Dilip Kumar questioned how filtering would support larger replication rings in #501209, while Amit Kapila argued for a simpler first implementation in #23; Peter Smith favored publication-level control, but Amit defended per-subscription choice in #27.
- Ashutosh Bapat objected to storing filtering policy in replication slots in #26, and Vignesh removed the slot machinery while adding pg_dump and psql support in #33.
- Following Amit Kapila's copy_data=false/true/force proposal in #4, Vignesh added a publisher-side pg_subscription_rel check in #36, abandoning the proposed extra replicated-data catalog flag as explained in #37.
- Ashutosh Bapat showed that the existing origin-filter callback was sufficient in #42, so #45 removed the new callback and decoding-core changes and implemented filtering through pgoutput.
- Peter Smith challenged the force option and sought clearer setup instructions in #51, while Amit Kapila defended its use with mixed-table publications in #53; subsequent revisions expanded the examples and tests.
- Peter Smith's multi-node testing exposed duplicate subscription-table entries in #63, which #64 fixed by selecting distinct relation IDs instead of subscription states.
May - Jun 2022 · 6 entries
- The revisions through #80 refined third-node tests, deterministic query output, copy-data handling and setup documentation, and Peter Smith reported both patches satisfactory in #81.
- Amit Kapila rejected an unnecessary protocol-version bump in #84 and identified the need to quiesce nodes during setup in #86, prompting compatibility, testing and documentation changes in #88.
- Amit Kapila adopted an extensible origin=local/any interface in #105, and Vignesh implemented it in #108, separating the existing-functionality test, origin feature, initial-copy safeguard and setup guide into four patches.
- Shi yu and Peter Smith identified contradictory locking instructions in #106 and #107, leading Vignesh to remove the unsupported example of joining a new node already containing data, as confirmed in #113.
- Peter Smith's review in #116 led to caching the parsed filtering choice, while default-value reviews in #119 and #125 produced consistent non-null origin=any handling by #129.
- Shi yu found that REFRESH PUBLICATION incorrectly rejected existing subscription tables in #146; #147 initially exempted ready tables, and Amit's review in #165 led #166 to exempt all already subscribed tables.
Jul - Aug 2022 · 13 entries
- Álvaro Herrera questioned the additional test_decoding test's value in #155, and Vignesh removed it as redundant in #162, without committing that separate test patch.
- Amit Kapila supplied a comments and error-message fix-up in #168, which Vignesh incorporated into the full series #169.
- Dilip Kumar explained that locally generated changes can still carry an explicitly assigned origin in #173, so #175 replaced the misleading value local with none.
- Dilip Kumar proposed reserving special names in #178, and #180 prohibited replication origins named any or none before #181 added case-insensitive option handling.
- Amit Kapila prepared the isolated feature patch in #182, Peter Smith corrected a comment in #183, and Vignesh supplied final formatting and test renaming in #184; this became 366283961a with only a catalog-version addition.
- Vignesh rebased the remaining initial-copy safeguard and setup guide in #186, and Jonathan Katz's testing and terminology concerns in #187 led #189 to describe replication between primaries and explicitly acknowledge conflicts and other restrictions.
- Jonathan Katz challenged copy_data=force and advocated a warning in #200, while Amit Kapila initially defended stronger protection for explicitly requested filtering in #203.
- Wang wei supplied documentation and translation improvements in #208, which Vignesh folded into #209 alongside further review feedback.
- Shi yu raised old-publisher compatibility in #220, and Peter Smith suggested rejecting unsupported filtering in #223, while Amit Kapila argued in #222 that an isolated initial-copy check could not make unsupported streaming filtering work.
- Masahiko Sawada objected to false positives and the usability cost of force in #226, and Zhijie Hou demonstrated a legitimate setup rejected by the error in #229, helping the discussion settle on warnings.
- Vignesh replaced the error and force option with a warning in #234, also replacing prescribed EXCLUSIVE locks with a requirement to prevent writes because those locks could block tablesync itself.
- Peter Smith's review in #238 led to warnings listing suspect tables in #242, before Amit Kapila's feedback in #249 prompted publication lists and diagnostic SQL in #250.
- Amit Kapila questioned expensive three-node tests in #240, and the test simplification through #250 was explained in #251: the evolved safeguard did not itself implement the broader node-joining scenarios.
Sep 2022 · 5 entries
- Reviews by Peter Smith, Shi yu, Wang wei and Amit Kapila refined publisher-side table exclusion, joins, documentation placement and memory handling across #257, #261, #267 and #273.
- Amit Kapila committed the warning component of #273 unchanged as 8756930190, leaving the separate replication-setup guide uncommitted.
- Vignesh reported an immediate buildfarm failure in #274, and Amit Kapila diagnosed in #275 that synchronization helpers accepted syncdone state even though dropping the table required ready state.
- Vignesh added explicit ready-state polling in #277, which became 0324651573 with only an explanatory comment edit.
- Amit Kapila recommended moving the remaining setup guide to a separate thread in #278, and Vignesh closed the commitfest entry while leaving that documentation pending in #279.
Jan 2023 · 1 entry
- Jonathan Katz later reproduced bidirectional synchronous-replication deadlocks in #280 and requested clearer synchronous_commit guidance in #282; Amit Kapila explained that the issue also occurred without origin filtering and recommended separate behavior-fix work in #283.
Aug 2023 · 3 entries
- Peter Eisentraut identified inaccurate wording that described publications as subscribing to other publications in #284, prompting Peter Smith's correction #286 and Amit Kapila's translation review in #287.
- Eisentraut committed the wording correction with further prose edits as 1e7ca1189c and confirmed completion in #290; the candidate list also includes its identical PostgreSQL 16 counterpart 842b659050, representing the same change.
- At the final confirmation #290, origin filtering, the initial-copy warning and both follow-up fixes had landed, while the broader setup guide and synchronous-replication deadlock questions had no completed resolution in this thread.