model-writtenrun 4537 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Zsolt Parragi reported that database commands permanently unregistered interruptible background workers even when the command failed, and proposed a shared interrupt flag that preserved restart behavior in #1.
- Aya Iwata questioned repeated reconnection attempts after a successful DROP DATABASE in #2; Zsolt argued that existing behavior already allowed such attempts and that a worker might reconnect elsewhere in #3.
- Xuneng Zhou proposed direct signaling without changing worker registrations in #4, and Michael Paquier favored that simpler design despite the PID-reuse risk, while preserving extension authors' choice of restart policy, in #8 and #9.
- Michael supplied the direct-signaling implementation in #13, collecting PIDs under shared locks and signaling after unlocking, and changed the test interface to accept a restart interval so zero-delay restarts could speed testing.
- Xuneng identified a PID-inequality assumption and incomplete exit-status documentation in #15; Zsolt incorporated those corrections, restored total_slots, and corrected who sends the signal in #16.
- Inspired by shorter test delays, Xuneng separately submitted logical replication permission-test speedups in #17; no landing or further disposition of that independent patch is shown in this thread.
- Xuneng added backend-start-time checks to handle PID reuse in #18, but Michael rejected the added test complexity as disproportionate to the risk in #19, and Xuneng accepted that decision in #20.
- Aya reported successful build and regression testing and requested an explanation of why direct signaling enables restarts in #21; Zsolt returned to the simpler test and made only two comment edits in #22, which Xuneng approved with a further wording suggestion in #23.
- Michael committed #22 as 5bf919736b, with matching code, tests, and documentation and only minor commit-message edits, then announced the backpatch through PostgreSQL 19 in #24.
- After Nathan Bossart asked whether the release open item could be resolved in #25, Michael confirmed that it was closed in #26, completing the background-worker fix.