Current dependencies on/for this PR:

• main
  • PR fix: send end of logs when dbfake completes job #11402 👈

This stack of pull requests is managed by Graphite.

Nice find. The fact that test-postgres was working fine should have been a signal to me about dbfake.

It's not a postgres vs inmemory db issue. The old "dbfake" was renamed to "dbmem" and now "dbfake" is a package for creating fake data in a database (and works on dbmem and postgres).

It's just that provisionerdserver publishes logs and end of logs, and that's what the CLI uses to stop waiting for a build to complete. The race was between completing the job before we start watching logs (in which case we don't watch for logs at all and the test passes), vs watching logs first (in which case we never see end of logs and the cli hangs until the context is canceled).