model-writtenrun 5618 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Nov 2021 - Apr 2022 · 17 entries
- SATYANARAYANA NARLAPURAM opened the thread after finding that an exclusive backup's backup_label could make a restarted server replay old WAL or fail recovery; Michael Paquier recommended the existing non-exclusive backup API #1 #2.
- The discussion shifted from repairing exclusive backups to removing them: Tom Lane initially emphasized compatibility, Stephen Frost argued that the restart hazard and maintenance burden justified removal, and Tom clarified that he did not object to outright removal #5 #7 #11.
- David Steele explained the connection-management objections to removal and explored a marker-file alternative, but highlighted corruption risks when restore instructions were missed and ultimately preferred removing exclusive mode #13 #17 #19.
- Nathan Bossart revived David's older removal patch, reporting that it built and passed tests; it deleted exclusive-backup state and APIs, simplified documentation, and switched the logical-decoding timeline test to a cold filesystem backup #21.
- Andrew Dunstan questioned removing backup_fs_hot(), but Nathan and David concluded that preserving its sole use was not worthwhile because the non-exclusive API requires a persistent connection and the test worked with backup_fs_cold() #22 #26 #28 #29.
- After a February rebase, Ashutosh Sharma identified misleading comments and questioned duplicated cleanup logic; Nathan fixed the comments, declined the helper extraction, and explained why the pg_basebackup test now had to delete its bogus backup_label explicitly #30 #31 #34.
- Chapman Flack tested the patch successfully, found documentation omissions, and argued that removing the exclusive argument would unnecessarily break scripts already using non-exclusive backups #35 #36.
- Nathan found that selectively warning about an explicitly supplied exclusive argument was awkward, while David proposed renaming the functions to prevent older scripts from silently doing the wrong thing; the revised patch adopted pg_backup_start() and pg_backup_stop() without compatibility wrappers despite Chapman's preference for a transition period #39 #40 #46 #52.
- Chapman then identified a shell example that incorrectly used separate sessions for starting and stopping a backup; Nathan removed the example, and Stephen Frost and Magnus Hagander argued that an oversimplified replacement would encourage unreliable backups #53 #56 #58 #59.
- Chapman accepted removing the example and moved the broader question of documenting backup-tool requirements into another thread, leaving that work outside this patch #60 #62.
- Further review made the documentation examples use named arguments and extract the lsn field from the stop-backup result; Chapman then repeated installcheck-world, reread the documentation, and marked the patch ready for a committer #64 #65 #67 #68 #69.
- Stephen's committer review strengthened instructions to keep backup_label and tablespace_map out of the live data directory, clarified checkpoint timing, and removed obsolete pg_ctl warnings; Nathan incorporated these changes and David's subsequent comment typo correction #71 #72 #73 #74.
- Nathan subsequently simplified shutdown connection handling by replacing connsAllowed with a boolean and removing CAC_SUPERUSER; David reviewed the changes and reported successful pgBackRest regression tests #75 #76.
- Stephen posted the final revision with additional documentation, recovery comments, and consolidated recovery errors; David again reported successful pgBackRest regression tests, and Magnus and Nathan endorsed the result #77 #78 #79 #80.
- Stephen committed the final attachment unchanged as 39969e2a1e, removing exclusive backup mode for PostgreSQL 15, and announced completion in #82.
- Martín Marqués later reported a duplicated word in the new basebackup.c comment; Michael Paquier fixed it directly from the report in 83cca409ed and confirmed the correction #84 #85.
- The thread ended with Stephen acknowledging the typo fix and welcoming the removal for PostgreSQL 15; both changes had landed, with the broader backup-documentation discussion already moved elsewhere #86 #62.