model-writtenrun 8825 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Leif Gunnar Erlandsen proposed pausing when archived WAL runs out before a time-based recovery target, allowing administrators to supply missing WAL and restart recovery instead of accepting an unintended promotion #1.
- Peter Eisentraut requested support for other target types and initially suggested preserving the old behavior through a new recovery_target_action, while Laurenz Albe argued that premature promotion was not useful behavior to preserve #2 #3.
- Fujii Masao pointed out that standby mode already waits for more WAL and honors recovery targets; Peter considered simplifying the recovery modes, but Leif maintained that his proposal addressed administrators supplying missing archived WAL #4 #5 #6 #8.
- Peter then proposed a fatal error when recovery ends before its configured target; Leif extended his patch to XID, name and LSN targets, and subsequent discussion favored failure over a pause that operators might mistake for successful target recovery #9 #10 #11 #13 #14.
- Peter took over implementation with a reworked patch that replaced pausing with fatal errors and added documentation, test infrastructure and recovery tests #17.
- Kyotaro Horiguchi identified missing recovery-progress logging and a crash-recovery regression caused by retained target settings, while Leif also requested progress logging and questioned inclusion of immediate targets; Peter's next revision moved the check after logging and guarded it with ArchiveRecoveryRequested #18 #19 #20 #21.
- Kyotaro reviewed the revised code and documentation, caught a documentation typo, and supplied an additional patch separating test-node recovery-mode selection from streaming and restore setup #22.
- Peter committed #21 as dc788668bb with only the documentation typo corrected, declining Kyotaro's larger test-infrastructure refactor as out of scope and not clearly simpler #24.
- The recovery behavior change was landed by the thread's end, while Kyotaro's final request to document internal test helpers used externally remained unanswered in the available discussion #25.