model-writtenrun 7635 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Stephen Frost opened the thread with a tested patch raising checkpoint_completion_target from 0.5 to 0.9, updating documentation, and removing a redundant recovery-test setting #1; when Peter Eisentraut questioned why not use a value closer to 1.0 #2, Stephen explained that checkpoints need time for work beyond dirty-buffer writes #3.
- Tom Lane proposed removing the setting entirely because unnecessary GUCs impose maintenance costs #5, and Stephen supplied a replacement patch using a compile-time constant of 0.9 #8, while declining to take on the separate logging-default changes proposed by Laurenz Albe #7.
- Alvaro Herrera requested documentation preserving discoverability of the removed setting #9, and Stephen initially waited for agreement on an obsolete-settings documentation approach #11; Michael Paquier then reviewed the removal patch and requested more explicit wording about the 90% target #12.
- Andres Freund cautioned against hard-coding an insufficiently evaluated value and described workloads where shorter checkpoints reduce periods of unpredictable latency #13 #15, while Tomas Vondra defended runtime configurability and suggested a possible dynamic target #14.
- Peter Eisentraut recommended returning to the original default-only change because kernel-flushing behavior could still require tuning #16, and Stephen restored that approach with a rebase and Michael's explicit 90% explanation #17, leaving GUC removal and dynamic targeting unimplemented.
- Tom Lane identified misleading wording implying that 0.9 covers the entire checkpoint interval #18, and Stephen revised the documentation to explain the allowance for completion overhead #19.
- After a two-month gap, David Steele reviewed the patch, suggested replacing the idiomatic word 'slop' with 'time', and recommended deferring broader configuration-setting documentation cleanup #22; Stephen incorporated the wording change and prepared to commit #23.
- Michael Paquier found that the CheckPointCompletionTarget initializer in src/backend/postmaster/checkpointer.c still said 0.5 #24, and Stephen added the missing update to 0.9 #25.
- Nathan Bossart proposed clearer explanations of checkpoint I/O and recovery tradeoffs #26, which Stephen incorporated into the final attachment #28.
- Stephen committed the final attachment unchanged apart from rebase metadata as bbcc4eb2e0 and announced completion in #30, leaving the setting configurable with a default of 0.9 and no remaining patch awaiting review in this thread.