model-writtenrun 9717 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Alvaro Herrera proposed propagating replica-identity changes to existing and newly created partitions, mapping index identities to child indexes, and displaying partitioned-table replica identities in psql, while leaving legacy inheritance unchanged #1; he supplied the initial implementation in #2.
- Testing index identities exposed that replaying the parent's command failed because child indexes have different names, so Alvaro replaced command-level recursion with internal execution-time recursion and added index-based tests #3.
- Dump restoration then exposed invalid parent indexes and temporarily missing child indexes, leading Alvaro to permit those intermediate states; creating partitions beneath an invalid replica-identity index remained unsupported #4.
- Dmitry Dolgov reported missing-index errors and deadlocks during concurrent partition creation and replica-identity changes, then supplied a nested-partition reproducer #5 #8.
- Robert Haas questioned changing existing children while also changing defaults for future partitions and called for consistent semantics across ALTER TABLE operations #9; Alvaro proposed honoring ONLY to distinguish those cases #10.
- The consistency discussion uncovered missing documentation for partitioned-table tablespaces #13, and Alvaro proposed wording explaining that SET TABLESPACE affects future partitions without moving existing data #14.
- Simon Riggs opposed recursive tablespace moves because of locking and storage costs #15 #17, while Robert favored a uniform recursion policy with ONLY providing an explicit opt-out and accepted the proposed documentation clarification #16.
- Alvaro committed the inline tablespace clarification as 2c31825fb9, adding the explicit TABLESPACE override qualification and a punctuation correction, and announced that this portion had landed #18.
- Discussion continued over OWNER, CLUSTER, TRIGGER, TABLESPACE, and identity-column recursion #19 #21; Peter Eisentraut explained that identity columns intentionally belonged only to the partition root because sharing their sequence across partitions was not supported #22.
- Alvaro announced an intention to commit replica-identity propagation separately #23, but Robert maintained his objection to piecemeal behavior changes #24, and Alvaro deferred the commit because Dmitry's reported problem remained uninvestigated #25.
- After Alvaro requested further opinions #27, Peter supported applying replica identity across the partition hierarchy #28 and Michael Paquier supported inheriting it when creating partitions #29, while the separate ownership debate continued #30.
- The thread ended with only the documentation change landed: Alvaro marked replica-identity propagation Returned with Feedback, citing unresolved concurrency problems and the need to delay dumped replica-identity commands until indexes existed in every partition, with an intention to resubmit in November #31.