model-writtenrun 5885 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Michael Paquier proposed -N/--no-sync for pg_upgrade and enabled it in Unix and Windows upgrade tests to save a few seconds on slow systems, arguing that skipping the final initdb --sync-only call would not lose test coverage #1.
- Peter Eisentraut supported the patch while suggesting a global option to periodically test syncing, and Bruce Momjian encouraged examining other places that could use no-sync #2 #3.
- Michael committed the patch with only initialization ordering and formatting edits as 3d5ffccb6d, then identified other potential sync costs and proposed a PG_TEST_SYNC environment variable as future work #4.
- After the commit, Alvaro Herrera warned that disabling syncing could hide bugs, citing an earlier failure exposed by the pg_upgrade test, and asked whether some buildfarm animals could retain syncing #5.
- Michael questioned whether that earlier backend fsync failure depended on pg_upgrade's final directory flush and offered to revert the test.sh change pending a broader testing facility, while asking to retain the Windows optimization; the thread ended with the commit landed and this testing concern unresolved #6.