model-writtenrun 7868 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane proposed storing the postmaster's SIGQUIT reason in shared memory so clients could distinguish immediate shutdown, another backend's crash, and an unexpected signal instead of always receiving a crash-and-restart warning #1.
- Bharath Rupireddy questioned reading potentially corrupted shared memory and proposed explicit immediate-shutdown wording; Tom explained that the memory remained accessible, accepted the wording change, and considered an incorrect diagnostic an acceptable corruption risk #2 #4.
- Bharath suggested a default switch case, but Tom retained exhaustive enum handling to preserve compiler warnings and committed the initial patch with only wording and whitespace changes as 7e784d1dc1 #5 #6.
- After the first landing, Andres Freund requested suppression of repetitive shutdown messages in server logs, while Bharath explored extending shutdown-reason reporting to SIGTERM; Tom judged precise SIGTERM attribution to require additional machinery and left it outside this work #7 #8 #9.
- Tom offered either shorter log messages or complete suppression for postmaster-issued SIGQUIT, now distinguishable from rogue signals, and Andres favored suppression #9 #10.
- Tom posted a separate patch introducing WARNING_CLIENT_ONLY and renumbering subsequent error levels so quickdie() could notify clients without duplicating the postmaster's server-log entry #11.
- Andres suggested eventually separating logging destinations into bit flags, but Tom agreed that this should be a separate patch; the discussion did not change the attached implementation #12 #13.
- Tom additionally proposed clearing error_context_stack and debug_query_string to reduce signal-handler work, then retained only the context-stack reset because suppressed server logging already avoided consulting debug_query_string #13 #14.
- The log-suppression patch landed with that additional reset as 1f9158ba48, leaving both submitted changes committed by the thread's final message and the broader SIGTERM and logging-bitmask ideas unimplemented here #14.