model-writtenrun 797 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Seino Yuki proposed logging the blocking processes when FOR UPDATE NOWAIT fails, attaching an initial implementation that reused lock-wait diagnostics in proc.c #1.
- Fujii Masao identified compilation errors and warned that unconditional NOWAIT logging could be noisy #2; Seino then clarified that the existing diagnostics were triggered by deadlock_timeout rather than lock_timeout #4, and Fujii agreed to defer timeout and cancellation support #5.
- Seino first added a separate log_lock_nowaits switch #6, then followed Fujii's preference for extending log_lock_waits #8 by implementing off/on/fail/all values #10.
- Fujii reconsidered the configuration interface because extending log_lock_waits could make configuration complicated #11, leading Seino to rebase and introduce the separate log_lock_failure boolean while leaving timeout and cancellation support for later #13.
- Fujii observed that the patch also affected SKIP LOCKED and recommended moving reporting outside JoinWaitQueue() to avoid doing that work under an exclusive partition lock #14; Seino moved reporting into LockAcquireExtended() and documented SKIP LOCKED coverage #15.
- Concern about excessive SKIP LOCKED output prompted Seino to propose excluding it #17, Fujii to accept a NOWAIT-only first implementation #18, and Jelte Fennema-Nio to support that boundary because concurrent job queues commonly use SKIP LOCKED #19.
- Discussion of how to distinguish callers led Fujii to suggest a logging-control argument and recommend initially supporting only SELECT NOWAIT #23; Seino tested all four heap error paths and supplied a rework that passed collected lock details back to heap-level callers #25.
- After compilation and reporting fixes #27, Fujii replaced duplicated heap logging with centralized reporting controlled by a boolean argument to LockAcquireExtended() #28; Seino endorsed the simpler design #29, and Fujii posted the final cosmetic revision #30.
- Fujii committed the diff from #30 unchanged as 6d376c3b0d, crediting Seino as author and himself as co-author, and expanded the commit description to explain why SKIP LOCKED was excluded.
- After announcing the original landing, Fujii proposed extending logging to LOCK TABLE and other NOWAIT operations through ConditionalLockRelationOid() #31, but Seino identified unwanted autovacuum logging #32, so Fujii withdrew the extension pending further design work #33.
- Peter Eisentraut proposed renaming the setting to log_lock_failures for consistency with log_lock_waits #34; Fujii supplied the rename #36, Peter approved it #37, and it landed unchanged as 73bdcfab35, announced in #38.
- After nearly a year, ChangAo Chen revived the extension with a patch limited to LOCK NOWAIT, adding explicit logging control while keeping maintenance callers quiet #39.
- Fujii objected to adding a logging flag to the relation-lookup options and instead proposed broader relation-lock logging with log_lock_failures forcibly disabled in autovacuum workers #40.
- ChangAo's final revision removed the extra flag and enabled logging through the existing RVR_NOWAIT flag, retaining the narrower scope because whole-database REPACK and VACUUM SKIP LOCKED could still be noisy; the original feature and rename had landed, while this extension remained in review with a new CommitFest entry and its scope still under discussion #41.