model-writtenrun 7220 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Maxim Orlov reported severe spinlock contention during concurrent analytical queries on a replica and proposed atomic threshold fields, reporting that sampled spinlock occurrences fell from 3486 to eight with his patch #1; Pavel Borisov supported the direction but requested direct throughput evidence #2.
- Robert Haas questioned why the atomic conversion retained a mutex and requested a clear synchronization model #3; Maxim then removed the mutex #4, but Robert warned that passing tests could not establish concurrency safety #5.
- Maxim concluded that reading the timestamp and transaction ID independently could calculate an incorrect cleanup horizon, so he restored locking for paired accesses and retained an atomic timestamp for independent reads #6; Robert still wanted a clearer safety explanation #7, which Maxim supplied by distinguishing the access patterns #8.
- Andres Freund redirected the discussion toward removing the feature, citing reproducible wrong query results, years without repairs, and performance costs even when the feature was disabled #9.
- Following agreement at a developer meeting, Thomas Munro posted a new removal patch for PostgreSQL 17, replacing the optimization effort with deletion of the feature, its extension, and its tests; he also raised whether error codes or extension compatibility stubs should remain #10.
- Thomas committed the removal from #10 with light comment edits and an adjustment for relocated wait-event definitions as f691f5b80a, removing the GUC, error code, and checking API, then announced completion while expressing hope for a future replacement #11; the related discussions in thread #42154 and thread #42242 preceded this thread's matching removal patch.
- Peter Geoghegan found unused snapshot arguments in nbtree and obsolete comments after the removal #12; Thomas confirmed these were accidental omissions #13.
- Thomas posted a follow-up removing unused snapshot arguments across access-method internals and stray comments, explaining that intervening code changes had discouraged a straight revert and reporting checks with clang's unused-parameter warnings #14.
- Peter approved the cleanup #15, which Thomas committed unchanged as 9f0602539d and confirmed pushed in the final message #16; both removal patches had landed, while the original atomic optimization was superseded.