model-writtenrun 4346 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Bowenshi reported that snapshot restoration could leave a stale next_phase_at value and trigger an assertion during logical replication on PostgreSQL 14.6 #1, then supplied a restoration fix and a Perl reproducer #2.
- Vignesh reproduced the failure on 14.6 but not HEAD #3; Masahiko Sawada explained the snapshot-restoration sequence, reproduced it using an injected sleep, confirmed the fix, and requested placing the assignment immediately after the consistency assertion #5.
- Bowenshi adopted that placement and incorporated the reproduction test into the revised patch #6, while Hayato Kuroda reported that reproduction on HEAD remained intermittent #7.
- Bowenshi reported that the fix removed the original failure but exposed a different assertion involving safeXid and snapshot xmin #8; Sawada connected that second failure to another discussion #9.
- After several idle months, Daniel Gustafsson identified obsolete test-module imports, an unused DBI dependency, and outdated test planning, and questioned whether the work belonged with the other assertion fix #10.
- Sawada clarified that this thread should address the original next_phase_at assertion and supplied a smaller patch that directly resets the field to InvalidTransactionId, leaving out the disputed regression test #11.
- Alexander Lakhin supplied concurrent reproduction steps #12 and then reduced them to a small SQL workload, verified that Sawada's patch prevented failures, and supported committing without a resource-intensive regression test #14.
- Gustafsson applied Sawada's final patch unchanged as development-branch commit 48efb2302b and its v16 backport eb27d3dc88, explaining that a regression test would be expensive and potentially brittle #15.
- Alexander Pyhalov subsequently reported the same issue on 14.8 and asked why older branches had not received the fix #17.
- At the thread's end, Gustafsson acknowledged the missed report and announced plans to backpatch to PostgreSQL 14 and 15 #18, with Sawada agreeing and reiterating that those versions were affected #19; the final messages do not confirm those additional backports.