model-writtenrun 7254 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andres Freund proposed retiring vacuum_defer_cleanup_age because it complicated visibility horizons, lacked adequate tests and retained rows using an unreliable transaction-count proxy, initially suggesting deprecation for PostgreSQL 16 #1.
- Magnus Hagander supported removal, Alvaro Herrera suggested temporarily forcing the setting to zero, and Justin Pryzby questioned the need for a staged transition #2 #3 #4.
- Andres supplied a complete removal patch covering the implementation, configuration, pg_upgrade and documentation, arguing that carrying the complication forward was unnecessary #5.
- Daniel Gustafsson found documentation still referring to the removed setting; Andres proposed wording explaining that hot_standby_feedback needs a replication slot to preserve protection across disconnections, and Daniel endorsed it #6 #7 #8.
- Peter Geoghegan supported removal but suggested that a future replacement should constrain actual costs on the primary rather than transaction age; those ideas remained outside the removal patch #9.
- After Justin supported removal before the betas, Andres posted the revised patch with the documentation correction and sought release-team input #10 #11.
- Alvaro reviewed the revised implementation positively and proposed a broader reorganization of the replication-slot documentation #13.
- Jonathan S. Katz supported removal for PostgreSQL 16 based on low observed usage, operational risks and the availability of hot_standby_feedback, while Laurenz Albe and Robert Haas cautioned that some users could still depend on the setting #15 #16 #17 #19.
- Greg Stark raised disconnected archive-replay workloads, Alvaro suggested controlling replay for those cases, and Laurenz proposed recommending statement_timeout with hot_standby_feedback as a time-based alternative; Jonathan endorsed that recommendation #20 #21 #22 #24.
- Andres deferred the broader documentation rewrite because its merged explanation conflated WAL retention with feedback protection, and Alvaro accepted landing the existing version #26 #27.
- Phil Florent raised a possible use with pg_dirtyread for recovering mistaken DML, but the supplied thread contains no answer to that use case; its final message announces that removal was pushed #29 #30.
- The removal landed from the revised patch with identical file changes and an expanded commit message, completing the attached work #11 1118cd37eb.
- The deferred documentation cleanup originated with Alvaro's proposal #13, but its eventual January 2024 landing belongs to the later matching patchset #551103 in thread #49038, with light wording changes and a separate WAL-retention caution block, without backpatching 27d04ed531.