model-writtenrun 5940 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andy Fan reported that a crash could undo sequence advances from an uncommitted transaction and proposed an inline XLogFlush call to protect values used outside the database in #1.
- David G. Johnston connected the report to an earlier unresolved documentation discussion in #2, while Laurenz Albe opposed the flushing overhead and recommended documenting the need to commit in #3; Andy clarified that sequence WAL logging batches values in #4, but Laurenz maintained his objection in #5.
- Nathan Bossart supported documentation and floated optional stronger persistence in #6; the discussion then distinguished sequence caching from durability, with Jeremy Schneider asserting that even committed sequence-only transactions could be unsafe in #8.
- Tomas Vondra pointed out that the reported reproduction never committed in #10, then reported unsuccessful attempts to reproduce a committed-value loss and explained why commit flushes sequence WAL in #12; Andy agreed with that durability analysis in #13.
- Jeremy Schneider withdrew one earlier reproduction because closing psql aborts an open transaction, while leaving a separate concurrent I/O-suspension reproduction for further investigation in #14.
- Tom Lane confirmed the WAL and transaction-ID handling, rejected a behavior change, and proposed replacing the misleading promise that nextval is never rolled back with explicit crash-persistence guidance in #15; his attached patch also clarified setval visibility and persistence in #16.
- Nathan Bossart approved the patch and accepted requiring commit in #17; Andy emphasized that uncommitted sequence advances can nevertheless persist through page flushing in #18, and Laurenz Albe approved the documentation in #19.
- Tom Lane committed #16 with a redundant sentence removed as 4ac452e228 and announced completion in #20; the documentation landed without a runtime change, while Jeremy's remaining reproduction question received no resolution in this thread.