pg_wal/.wal-g breaks pg_rewind

[thequailman]

Hello, I'm running wal-g with PostgreSQL 13. When I run pg_rewind -R -D /var/lib/postgresql/13/main --source-server=postgresql://myserver/postgres --restore-target-wal with wal-g as the archive_command and restore_command, I get a fatal error from pg_rewind:

pg_rewind: servers diverged at WAL location 18C/35000000 on timeline 4
pg_rewind: rewinding from last common checkpoint at 18C/34000060 on timeline 4
pg_rewind: fatal: could not remove file "/var/lib/postgresql/13/main/pg_wal/.wal-g/prefetch/running/000000050000018C00000036": No such file or directory

This also has the fun side effect of breaking the underlying cluster (no pg_control) and creating a new timeline in wal-g.

This issue was reported on the PostgreSQL mailing list (https://www.postgresql-archive.org/Concurrency-issue-in-pg-rewind-td6154319.html) but I don't think it was ever resolved. Basically, wal-g shouldn't be creating folders under pg_wal was what the maintainers recommended? Could you make the .wal-g prefetch directory configurable?

[x4m]

Thanks for reminding me. I've totally forgotten about it...