model-writtenrun 7976 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed making archive_cleanup_command, promote_trigger_file, recovery_end_command, and recovery_min_apply_delay reloadable in #1; Michael Paquier checked their process usage and documentation in #2, and Peter committed the patch unchanged as 13b89f96d0, announcing it in #3.
- Sergei Kornilov asked about extending this to restore_command in #4; Peter wanted to await the primary_conninfo work in #5, while Michael raised possible WAL receiver interference in #6 and Andres Freund pointed to the existing receiver shutdown synchronization in #7.
- After primary_conninfo became reloadable, Sergei returned with a separate restore_command patch in #8, arguing that archive restoration runs with the WAL receiver stopped.
- Michael identified restore_command use by logical WAL senders in #10, but Robert Haas explained in #11 that different processes using old and new commands need not compromise internal consistency.
- Anastasia Lubennikova raised archive-source changes, pg_rewind, and future prefetch interactions in #12; Kyotaro Horiguchi explained standby waiting behavior in #13, and Sergei argued in #14 that externally changing a restore script already permits the same archive-source hazards.
- Fujii Masao asked how an empty restore_command should behave after reload in #15, prompting Sergei to add recovery-parameter validation on reload in #16.
- Fujii rejected the added validation because it could terminate archive recovery on reload in #17, and Kyotaro examined related shutdown cases in #19 and #21; Sergei considered pausing recovery but favored avoiding special handling in #22.
- Fujii recommended returning to the original implementation in #23, and Sergei supplied its rebase in #24, abandoning reload-time validation while retaining the startup requirement for a nonempty restore_command.
- Fujii considered documenting the distinction between startup and reload requirements in #25; Kyotaro requested a warning about invalid commands in #26, then withdrew that request after checking the comparable behavior of recovery_end_command in #28.
- Fujii committed #24 unchanged as 942305a363 and announced it in #29; Sergei's thanks in #30 closed the thread with both changes landed and no remaining patch under review.