model-writtenrun 9881 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Sergei Kornilov revived earlier recovery-configuration work in #1 and attached it in #2, moving recovery settings into the GUC system, using recovery.signal and standby.signal to select recovery state, rejecting legacy recovery.conf, and making pg_basebackup -R append configuration to postgresql.auto.conf.
- The initial proposal retained restart-only settings and replaced five recovery-target settings with a type/value pair; Sergei kept it current through the July and September rebases in #4 and #7.
- Peter Eisentraut argued in #8 for preserving the existing recovery-target settings and the familiar trigger terminology, while Tom Lane identified unsafe interdependent GUC checks in #9; Andres Freund proposed checking dependencies at first use in #10, but Tom argued in #11 that this would sacrifice useful configuration-time validation.
- Sergei adopted the five existing recovery-target settings with separate GUC hooks and renamed the promotion setting to promote_trigger_file in #13, leaving broader target-setting reorganization for future work as agreed in #12.
- Peter's detailed review in #16 led Sergei to remove redundant recovery-parameter logging, unrelated restore-point and promotion-file renames, and translation problems, while fixing GUC defaults and test-helper naming in #17.
- Peter took over final preparation in #18, refining validation and signal-file handling, removing the obsolete recovery.conf sample, and integrating recovery documentation into the main server-configuration chapter.
- Michael Paquier questioned removing standby_mode in #19, but Peter explained in #20 that a persistent setting could accidentally turn a later PITR into standby recovery; Michael accepted that reasoning in #21, while suggesting that retaining completed signal files might help debugging.
- Peter committed #18 as 2dedf4d9a8 with light documentation edits and the additional Fujii Masao author credit requested by Michael, announcing completion in #22.
- After the commit, Stephen Frost warned in #24 that repeated pg_basebackup -R operations could accumulate duplicate settings in postgresql.auto.conf and that stale recovery targets could survive promotion, complicating automated restores and third-party backup tools.
- Tom Lane reported reproducible EXEC_BACKEND buildfarm failures in #25, suspecting an assumption that recovery state would be inherited through fork; the supplied thread contains no subsequent fix for that report.
- Andres proposed a separate overwriteable recovery configuration file in #26, Stephen supported preserving configuration separation in #28, and Petr Jelinek emphasized in #30 that overwriting such a file would avoid interfering with ALTER SYSTEM settings.
- Peter raised file-ordering, ownership, and location questions in #33, while David Steele proposed an automatically included temporary recovery file renamed after recovery in #34; Peter replied in #35 that persistent standby settings and temporary recovery targets might require different treatment.
- Stephen proposed replacing duplicate pg_basebackup settings, removing completed recovery targets, and documenting or providing configuration-editing tools in #36, then explained the value of persistent primary_conninfo and restore_command for symmetric HA configurations in #38.
- The core integration remained committed, but configuration lifecycle and backup-tool compatibility were unresolved when David renewed the discussion in #39 and #40, advocating an automatically included postgresql.recovery.conf that could be renamed after recovery.