model-writtenrun 9449 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Thunder reported that a shutdown signal during bootstrap caused PANIC when PostgreSQL tried to abort the already-committed bootstrap transaction, and proposed an inline guard skipping abort for non-normal transaction IDs #1.
- Kuntal Ghosh raised concerns about transaction cleanup, while Michael Paquier noted that bootstrap failure already requires reinitialization and that treating bootstrap transaction IDs as committed was correct #2 #3.
- Tom Lane redirected the fix toward bootstrap signal handling: die() incorrectly attempts transaction abort, whereas immediate termination lets initdb remove the incomplete data directory #4.
- Tom Lane posted the sole attached patch, assigning SIG_DFL to SIGHUP, SIGINT, SIGTERM and SIGQUIT, and demonstrated that interrupting bootstrap then produced a signal-termination message followed by initdb cleanup #6.
- Michael Paquier opposed suppressing abort-path errors for bootstrap transactions; Thunder explained that crashes and core dumps triggered operational complaints, and Tom distinguished external signals from internal errors that should still fail loudly #7 #8 #9.
- Michael Paquier accepted changing the signal handlers but questioned why bootstrap was being signaled, while Andres Freund identified shutdown, timeout and resource exhaustion as plausible causes that should not generate core files #10 #11.
- Tom Lane committed the attached patch with only a comment rewrite as fb489e4b31, declining backpatching because the longstanding behavior was essentially cosmetic, and closed the thread by confirming that the SIG_DFL fix had been pushed #12.