model-writtenrun 5330 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Nathan Bossart proposed removing durable archive-status renames after measuring substantial fsync overhead; his initial patch used rename() in the archiver and moved directory synchronization into the checkpoint-side completion check to protect WAL recycling after crashes #1.
- After reproducing duplicate hard links during interrupted WAL recycling #2, Nathan replaced the checkpoint synchronization approach with a change from durable_rename_excl() to durable_rename() when installing recycled WAL segments #3.
- Kyotaro Horiguchi questioned crash safety, extra checkpointer work, and documentation that discouraged archiving the same file twice #4; Nathan explained that the revised recycling fix avoided extra checkpoint work and prevented re-archiving a file being concurrently reused #5.
- Nathan measured approximately 43 percent less archiving overhead with the revised patch on a larger test system using a minimal archive module #6.
- Nathan moved the independently backpatchable WAL recycling fix to another thread and retained only the archive-status rename optimization #7; Robert Haas considered that change reasonable but said it had missed the feature freeze and would need to wait for PostgreSQL 16 #8.
- When Fujii Masao approved the optimization in July #10, Robert raised the need for clearer re-archiving documentation #11, and David Steele agreed while emphasizing that repeated archival was already possible #12.
- Nathan introduced documentation allowing success for identical existing archives or failure requiring administrator intervention #14, then added the requirement that an existing archive be fully persisted and incorporated David's wording cleanup #17 #18 #19.
- Kyotaro requested consistent modal wording #20, which Nathan incorporated while combining the documentation and code patches #22; David accepted the storage-neutral persistence wording without additional implementation detail #23.
- Fujii committed the code component of #22 unchanged as 756e221db6 and the documentation component with a small markup adjustment as 3cabe45a81, then suggested that basic_archive-specific documentation could be improved separately #25.
- Noah Misch objected that explicitly permitting failure for identical persisted archives invited dangerous manual intervention #27; Nathan supplied a follow-up recommending success and removing that alternative #28, which David endorsed as clearer #29 and Noah accepted #30.
- Noah committed Nathan's follow-up #28 with only paragraph reflow as d263ced225, crediting David's review.
- Peter Eisentraut subsequently pointed out that the documentation addressed archive_library but omitted archive_command #31, and proposed covering both while explaining manual recovery when duplicate handling was unavailable #34.
- Noah opposed restoring language that encouraged manual workarounds #35, David warned against describing duplicate archival as rare #36, and Robert argued that explaining the consequences of inadequate handling could still help users #37 #38.
- All attached work had landed by the thread's end, but documentation covering archive_command and the consequences of inadequate duplicate handling remained unresolved; Noah accepted the value of explaining dangerous manual fixes, without a further patch or commit in this thread #39.