model-writtenrun 5694 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Dec 2021 - Mar 2022 ยท 16 entries
- Bharath Rupireddy proposed adding checkpoint and REDO LSNs to checkpoint and restartpoint completion logs so historical checkpoint problems could be diagnosed without relying solely on current pg_controldata output #1.
- Michael Paquier questioned whether users could distinguish the two fields #4, prompting Bharath to submit alternative 'location' and 'REDO start location' formats alongside the original patch #6.
- Julien Rouhaud identified that the mutually exclusive attachments confused cfbot #8, so Bharath selected one patch and resent the alternatives as text attachments #16; a subsequent test failure led him to preserve the existing completion-message prefixes #17.
- Nathan Bossart initially favored 'start/end' labels #18, which Bharath implemented #22, but Nathan then reconsidered their ambiguity and recommended 'lsn/redo lsn' #23.
- Stephen Frost supported explicit LSN terminology and proposed making pg_controldata consistent #25, leading to a two-patch series #26 and an alignment correction requested by Nathan #27 and supplied in #28.
- Fujii Masao caught reversed LSN arguments and asked that the diagnostic fields follow the existing I/O statistics #30, producing the corrected order #31 and final field placement #33.
- Kyotaro Horiguchi approved the logging change and identified other user-facing uses of 'location' #34; Bharath added a broader terminology patch while deliberately preserving backup_label contents for compatibility #35.
- Fujii questioned duplicate REDO reporting and discrepancies between local and control-file pointers #36, then identified a promotion race that could remove WAL still required by the recorded restartpoint and leave the server unrecoverable #39.
- Kyotaro proposed updating checkpoint pointers even after promotion #40, and Fujii agreed that the minimum recovery point should then be cleared #44; Kyotaro supplied a separate recovery fix that also removed an assumed obsolete concurrent-checkpoint path #45.
- After Bharath reviewed the recovery fix #47, Kyotaro assembled five patches covering recovery correctness, LSN logging, pg_controldata terminology, broader wording, and hyphenation cleanup, while adding lock-safety comments to Bharath's logging patch #48.
- Fujii raised differences in older branches #49, prompting a PostgreSQL 14 recovery-fix variant #50; Kyotaro then corrected an erroneous OR condition in the development series #52.
- Further recovery review addressed redundant conditions and local minimum-recovery-point state #53, and Kyotaro supplied older-branch variants while investigating a pg_rewind test failure #55; he subsequently traced that failure to an accidentally removed condition #56 and confirmed the branch patches applied #59.
- Kyotaro refreshed the full series against current development sources and incorporated the revised recovery fix #60, then addressed Bharath's timestamp, locking-comment and documentation feedback #61 in #62.
- Michael asked that the unrelated work move to appropriately titled discussions #63, and Kyotaro agreed #64, split out the terminology patches #65, corrected their subject on resend #66, and left the standalone logging patch ready for a committer #67.
- Greg Stark examined the standalone patch and judged its reported test failure unrelated to the change #68; after a several-month gap, Fujii proposed qualifying the lock-safety comment #69, which Kyotaro accepted #70.
- Fujii committed #67 as 62c46eee22 with consolidated and clarified comments but unchanged executable additions, then confirmed the push in the thread's final message #71; the original logging feature was complete, while the recovery and wording work had moved to separate discussions.