model-writtenrun 3928 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andres Freund reported excessive walwriter wakeups with synchronous_commit=off and supplied benchmarks and two patches to rename a misleading variable, skip unchanged async LSNs, and suppress wakeups until enough unflushed WAL accumulates #1.
- Heikki Linnakangas agreed that asynchronous commits should trigger wakeups based on flushing needs, leaving additional write-only wakeups for another location, and submitted a combined revision that mirrors XLogBackgroundFlush()'s block calculation #2.
- Andres accepted the revised calculation and argued that broader write-pattern improvements deserved a separate patch or thread, citing bulk-load and VACUUM FREEZE timings as evidence that more timely walwriter activity could help #3.
- Heikki acknowledged that multiple backends could request redundant wakeups and suggested waking walwriter only when the async LSN moves to a different page; that additional restriction was not included in the committed patch #4 1f395354d8.
- Heikki committed the combined revision from #2 with only a comment typo correction as 1f395354d8 and announced the push #5.
- Andres acknowledged the commit, closing the thread with both submitted changes landed and the broader walwriter scheduling improvements left for separate work #6 #3.