model-writtenrun 5991 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Bharath Rupireddy proposed enabling log_checkpoints by default because checkpoint statistics are useful for diagnosis and are already collected even when logging is disabled #1.
- Tom Lane opposed routine log output from healthy servers, while Andres Freund and Jan Wieck argued that checkpoint logs are essential for diagnosing performance incidents and impose little logging overhead #2 #3 #6.
- Michael Banck suggested logging only WAL-triggered checkpoints, while Magnus Hagander and Daniel Gustafsson suggested exposing more information through pg_stat_bgwriter for monitoring #8 #9 #10.
- Robert Haas broadened the proposal to include log_autovacuum_min_duration=10min for retrospective performance diagnosis; Nikolay Samokhvalov preferred one minute, but Robert defended ten minutes as a useful threshold with bounded log volume #11 #17 #20.
- Related proposals included replacing checkpoint_warning with ordinary checkpoint reports, enabling log_lock_waits, and making log_checkpoints an enum, but these changes were not included in the submitted patch #19 #23 #26 #42.
- Tom Lane accepted logging sufficiently long autovacuum operations but continued opposing routine checkpoint logging; Robert Haas argued that time-triggered checkpoints can also explain incidents and that the expected log volume was small #29 #32 #33.
- Bruce Momjian proposed changes to rotation, message classification, and buffering before increasing default logging, while Robert Haas argued that broader logging improvements should not block this small increase #35 #36 #38.
- Bharath submitted the combined defaults patch and showed that it also produced checkpoint messages during initdb #42.
- Robert Haas requested suppressing checkpoint output during initdb, Fujii Masao identified the missing autovacuum global initializer, and Bharath incorporated both corrections into the second submission #43 #44 #45.
- Discussion continued about separating operational severity from existing log levels and adding classifications for filtering, but these remained broader proposals outside the defaults patch #48 #49 #50 #51 #52.
- Robert Haas approved the revised patch and subsequently committed it unchanged apart from rebasing as 64da07c41a, crediting Bharath as author and Fujii as reviewer; the thread ends with confirmation that the complete submitted change landed #53 #55.