model-writtenrun 8771 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed letting the postmaster launch an auxiliary base-backup client before normal standby startup, using an initdb-created minimal directory and connection settings from postgresql.conf #1.
- Thomas Munro suggested the --replica option name and reported a Windows build problem, while Sergei Kornilov found EXEC_BACKEND failure, configuration overwriting, and missing concurrent WAL reception; restartability and safe cleanup also drew concerns #2 #3 #4 #12.
- Robert Haas proposed starting a replica directly through a postgres command-line option, but Peter retained a separate initialization step to reuse directory setup and fit service automation, while Tom Lane emphasized configuration, credentials, and service-tool compatibility #9 #10 #11.
- Peter's next revision retained the auxiliary-process design and added concurrent WAL reception, progress reporting, fsyncing, configuration-file exclusion, the --replica name, and a recovery test #13.
- Michael Paquier requested independently useful control-file diagnostics, switch-based state handling, and protection against WAL removal; Peter's revision also added tablespace symlinks without mapping, centralized directory syncing, and a pg_resetwal rename to avoid conflicting declarations #16 #17.
- After Michael agreed to extracting the diagnostic improvement, the unchanged StartupXLOG portion of the larger patch landed as b85e43feb3, leaving the backup client uncommitted #18 #17.
- Peter first demonstrated automatic temporary slots on the WAL sender, but Sergei questioned compatibility with pg_basebackup --no-slot; Peter then replaced that approach with receiver-controlled slots and two API prerequisites, which Michael considered a more natural use of the existing interface #19 #20 #21 #22.
- Masahiko Sawada reviewed the receiver series and requested that temporary slot names appear in pg_stat_wal_receiver; the optional-LSN prerequisite landed unchanged as c67a55da4e, and Peter resent the remaining patches with shared-memory slot bookkeeping #24 #25.
- Masahiko caught documentation that still claimed temporary slot names were hidden; Peter removed that stale hunk and committed temporary-slot support as 3297308278, alongside the unchanged backend-PID API patch as ee4ac46c8e #26 #27 #25.
- Peter returned to the backup client with a rebased remainder and clarified that its purpose was easier standby automation, with no plan to replace pg_basebackup #29 #30.
- Testing the backup client, Masahiko found that connections hung during the postmaster wait loop and that locally generated defaults could prevent standby startup when the primary required larger settings; he also questioned the signal-file handling #31.
- After temporary-slot support landed, Andres Freund warned that a slow standby could fill the primary's disk and that temporary slots did not protect across disconnects; Michael referenced another complaint, and Masahiko discussed slot exhaustion after failover while defending the protection during active replication #32 #34 #35.
- Andres's detailed backup-client review requested separate refactoring commits and raised unresolved concerns about startup polling, signal handling, failure recovery, and placing backup logic inside libpqwalreceiver #33.
- Peter extracted three independent patches, which landed unchanged: the pg_resetwal rename as 5f1b8260af, comment formatting as 9745f93afc, and InitControlFile extraction as 79c2385915 #36 #37.
- Peter marked the main backup-client proposal returned with feedback because he could not finish it and postmaster startup sequencing needed more thought; the prerequisite commits remained landed #38.
- The final message reopened temporary-slot semantics: Alvaro Herrera relayed Sergei's proposal from elsewhere to use primary_slot_name as a temporary-slot name when both settings were enabled, leaving that question unresolved in this thread #39.