model-writtenrun 896 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Baji Shaik reported that terminating REPACK (CONCURRENTLY) left its decoding worker and temporary replication slot alive because FATAL bypassed PG_FINALLY, and proposed an on_proc_exit termination callback in #1.
- Sami Imseih reproduced a crash caused by accessing the worker handle after its memory context had been destroyed, then supplied a before_shmem_exit revision that could safely terminate and wait for the worker in #2.
- Alvaro Herrera identified callback-slot exhaustion across repeated commands and recommended PG_ENSURE_ERROR_CLEANUP in #3; Baji adopted that scoped approach and reported successful regression, injection-point, termination, and repeated-command testing in #4.
- Alvaro committed #4 with naming and comment edits as 0160143ad9, then supplied a follow-up patch in #5 to release resources after incomplete worker initialization; he also declined the proposed dedicated termination test.
- Antonin Houska reviewed the follow-up in #6, requesting restoration of explicit shm_mq_detach() and clarification of the queue assertion failure; Alvaro agreed to restore the call and confirmed the failure explanation in #7.
- Alvaro committed the teardown patch from #5 with restored queue detachment, earlier queue-handle tracking, and revised cleanup ordering as 1588d89af2, then confirmed completion in #8, leaving both fixes landed.