model-writtenrun 8659 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut forwarded Pavan Deolasee's fix, developed with bug reporter Craig Ringer, to release the replication slot's lock on five SaveSlotToPath() error exits; low disk space had exposed a path where LOG-level failures could leave a walsender blocked indefinitely #1.
- Andres Freund questioned allowing slot-save failures without aborting a checkpoint, Alvaro Herrera suggested checkpoint continuity explained the behavior, and Robert Haas argued slot persistence need not be part of a checkpoint; the discussion produced no change to the error-level policy #2 #3 #5 #6.
- Thomas Munro caught a typo in the patch's comment, while Robert Haas raised a concern about wait-event cleanup and then withdrew it after Peter explained the existing handling #4 #6 #7 #8.
- Peter proposed applying and backpatching the focused fix because broader reorganization remained undefined and unsuitable for backpatching; Alvaro and Robert agreed, and Peter committed #1 with comment cleanup as 49bf81536e and announced the backpatches #9 #10 #11 #12.
- After the commit, Michael Paquier identified three new lock-release calls that could overwrite errno before error reporting, particularly with lock tracing enabled; Peter supplied a follow-up patch saving and restoring errno around those calls #13 #14.
- Michael endorsed the errno fix and preferred its consistency with surrounding code over moving lock release after logging; Peter's final reply reported completion, leaving no outstanding patch work in the thread, although the follow-up commit is not among the supplied candidates #15 #16.