model-writtenrun 6410 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Jeff Davis opened the thread by questioning why logical replication silently advances an old requested LSN to confirmed_flush_lsn, then identified misleading protocol documentation and the risk of hiding client errors or missing data in custom clients #1 #3.
- Amit Kapila explained that clients can acknowledge WAL that produces no logical changes, Robert Haas emphasized avoiding unnecessary durable progress writes, and Andres Freund confirmed that motivation and warned against breaking existing clients #2 #4 #5.
- Jeff proposed letting clients explicitly consult confirmed_flush_lsn; Robert preferred an optional START_REPLICATION flag over requiring SQL, while Andres argued that compatibility outweighed the proposed reliability improvement and that clients could already perform their own checks #6 #10 #11.
- With little support for changing behavior, Jeff proposed correcting the protocol documentation, expanding the code comment, and raising the diagnostic to WARNING; Robert supported the documentation and comment changes but opposed WARNING #14 #15.
- Jeff suggested LOG instead, Kyotaro Horiguchi supplied wording describing the already-streamed position and forwarding destination, and Amit accepted LOG because the event should be infrequent #16 #17 #18.
- Jeff attached a patch documenting that streaming starts at the greater of the requested LSN and confirmed_flush_lsn, advising clients to check their expectations, and incorporating the comment and LOG changes; Amit approved it in the thread's final message #19 #20.
- Jeff subsequently committed the complete patch with a missing comma repaired in the logging call as 14d474e079; the existing fast-forward behavior remained, and no behavioral change or optional protocol flag was included.