model-writtenrun 8485 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Vaibhav Dalvi reported that zero wal_receiver_status_interval could leave a stale feedback xmin blocking primary VACUUM on PG16 and later, and proposed sending invalid horizons with tests for reload and connection handling in #1; the separate case where Hot Standby is not yet active at connection time was explicitly excluded.
- Vaibhav prepared the submission for CommitFest by resending the unchanged master patch in #2, then responded to CFBot with a rebase and test renumbering in #3.
- Joao Rito reviewed and tested the rebased patch in #4, confirming that it fixed the VACUUM blockage and passed regression and recovery tests, but requested clearer zero-interval semantics and treatment of logical slots after demonstrating that releasing catalog_xmin could invalidate a standby logical slot.
- In #5, Vaibhav retained zero as disabling feedback rather than restoring PG14/15 behavior, documented the logical-slot consequence, centralized the enabling condition, marked the startup gap in code, and added catalog_xmin release and slot-invalidation tests; he reported the tests passing on master and PG16–19.
- Vaibhav declined a new startup warning to avoid adding messages across back branches and deferred a slotsync.c interval check to separate master-only work because reload tracking would also need changes in #5; the thread ends awaiting further review, with no linked commit and the early-connection stale-xmin gap still unresolved.