model-writtenrun 7643 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Aug - Sep 2016 · 2 entries
- Kartyshov Ivan introduced WAITLSN to let applications wait until an asynchronous standby had replayed required data, using a utility command to avoid acquiring a snapshot and shared latches to support timeouts and interruption #1.
- Craig Ringer explored how clients would obtain a commit identifier, while Thomas Munro proposed transaction-start integration and identified compiler errors, incorrect timeout arithmetic, synchronization problems, and excessive wakeups #2 #4.
Jan - Mar 2017 · 3 entries
- Ivan revised the core implementation with timeout fixes, explicit wait variants, and notification-throttling settings, then produced an extension using SQL functions and a replay hook at a customer's request #6 #10 #13.
- Masahiko Sawada questioned moving the feature into contrib, and Thomas Munro and Robert Haas argued that a function taking a snapshot before waiting could not provide the intended visibility at higher isolation levels; Ivan reposted the core-command branch #11 #12 #13 #16.
- Thomas's further review warned that throttled notifications could miss the last available WAL update and recommended standard backend latches and selective wakeups; Ivan subsequently replaced throttling with tracking of awaited LSNs and their minimum #14 #20 #21.
Aug - Nov 2017 · 2 entries
- Alexander Korotkov proposed ordinary TIMEOUT, INFINITE, and NOWAIT options instead of separate command-name suffixes, and Ivan implemented that syntax and removed incidental edits #22 #24 #25 #27.
- Ants Aasma and Robert Haas questioned exposing LSN ordering as the permanent visibility contract, with Ants proposing opaque state tokens and a WAIT FOR interface that could support future consistency mechanisms #23 #28 #29 #32.
Jan - Mar 2018 · 1 entry
- Simon Riggs proposed generalized event waiting with LSNs and timestamps, and Dmitry Ivanov asked whether multiple events meant ANY or ALL; the work nevertheless stalled without an updated implementation #36 #42 #45.
Mar - Apr 2020 · 11 entries
- After roughly two years, Ivan revived the work as WAIT FOR with timestamp handling and recovery tests, now using ordinary backend latches; Kyotaro Horiguchi questioned the syntax, advocated BEGIN integration, and again suggested queue-based waiting #49 #50.
- Ivan rebased the revived patch and implemented event lists with ANY/ALL handling, then added the waiting clause to BEGIN and START TRANSACTION while retaining standalone WAIT #53 #54 #56.
- Anna Akenteva tested the implementation and found misleading multi-event semantics, timeout conversion errors, broken indefinite waits, and a transaction-related crash, alongside missing null checks and node support #55 #57.
- Ivan invited Anna to work on the patch, and their joint revisions repaired parsing and timing, separated timeout events from LSN events, simplified shared storage, and expanded tests for both command forms #58 #59 #61.
- Alexander argued that multiple LSNs and timeouts introduced unnecessary ambiguity and that waiting inside an existing repeatable-read transaction could impede replay; Ivan consequently restricted the feature to one LSN and an optional timeout at transaction start #62 #64.
- Anna completed the narrowed design's node-copy and equality support and removed obsolete standalone-command machinery, while Amit Kapila identified remaining interruption and error-cleanup problems and questioned readiness to commit #66 #67 #68.
- Ivan and Anna addressed Amit's review with bounded polling, CHECK_FOR_INTERRUPTS, error-path cleanup, shared-state locking changes, and additional tests #72.
- Alexander committed that submission as 0f5ca02f53, adding naming and documentation cleanup plus substantive changes to minimum-LSN handling, latch registration checks, and polling intervals, and promised further review after the push #72 #73.
- The landing was immediately followed by renewed syntax discussion: Ivan proposed reusing existing keywords, Kyotaro suggested AFTER and WITHIN, and Anna favored BEGIN AFTER without a new LSN keyword #74 #76 #77.
- Tom Lane questioned coupling waiting to BEGIN and suggested a function or procedure, while Fujii Masao noted that a separate preceding transaction could avoid retaining its snapshot in the subsequent transaction #78 #80 #81.
- Andres Freund emphasized that a waiting function could itself hold a snapshot and cause recovery conflicts, and defended transaction-start syntax for client and pooler behavior; procedure support and timeout design remained unsettled #85 #87 #88 #89 #90.
Jul - Oct 2020 · 1 entry
- Anna later refreshed the submitted patch's terminology from master to primary, and Alexandra Pervushina took over the next revision, implementing BEGIN AFTER ... WITHIN with existing keywords #92 #94.
Nov 2020 · 1 entry
- Alexandra then replaced the transaction-start clause with three core pg_waitlsn functions, retaining a warning that calling them at higher isolation levels could acquire an unsuitable snapshot #95.
Jan 2021 · 1 entry
- Kyotaro supplied a native-procedure proof of concept that consolidated the waiting functions, but Ibrar Ahmed reported compilation failure because the attachment omitted commands/wait.h and other new implementation files #96 #97.
Mar 2021 · 1 entry
- Kyotaro closed the available discussion by reposting a rebased function patch and separating the procedure experiment into a text attachment; these alternatives remained awaiting review, with no second landing recorded in the supplied thread #98.