model-writtenrun 2841 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Jelte Fennema-Nio proposed preserving dbname in pg_basebackup and pg_receivewal connection strings so proxies could route physical replication connections, while preventing repeated connection-string expansion for security reasons #1.
- Thom Brown questioned taking backups through PgBouncer #2; Euler Taveira explained the transparent-proxy and restricted-access use cases #3, and Jelte added failover, load-balancing, and database-move scenarios #4.
- After Thom requested documentation of normal server behavior #5, Jelte added explanations for both utilities in #6; Tristen Raab's successful testing and documentation feedback #7 led to clearer wording in #8.
- Jim Jones verified that the requested database name reached PgBouncer but encountered a proxy crash #9; Jelte fixed that separate PgBouncer problem #10, and Jim then confirmed a successful backup #11.
- Jelte submitted final documentation refinements in #12, and Daniel Gustafsson reviewed the feature favorably #13 before committing it as cca97ce6a6 with shorter documentation and unchanged implementation, announcing completion in #14.
- More than a year later, Tom Lane traced bug #18685 to the feature's removal of the default 'replication' dbname when no connection string was supplied, which broke matching password-file entries, and posted a one-line correction #15.
- Jelte approved the correction and suggested clarifying that both connection inputs can be NULL #16, while Daniel independently endorsed the fix and requested explanatory commentary #17.
- Tom incorporated the comment and corrected the pg_basebackup documentation to explain that dbname affects password-file lookup #18; Jelte approved and requested equivalent pg_receivewal documentation #19, which Tom agreed to add in the thread's final message #20.
- After the final message, Tom committed #18 with the additional pg_receivewal documentation as e2a9129093 and backpatched the regression fix to PostgreSQL 17, completing both phases of the thread.