On 2019-May-30, Alvaro Herrera wrote:
> One thing I noticed while writing it, though, is that worker_spi uses
> the postgres database, instead of the contrib_regression database that
> was created for it. And we create a schema and a table there. This is
> going to get some eyebrows raised, I think, so I'll look into fixing
> that as a bugfix before getting this commit in.
Another thing I noticed when fixing *this*, in turn, is that if you load
worker_spi in shared_preload_libraries then the contrib_regression
database doesn't exist by the point that runs, so those workers fail to
start. The dynamic one does start in the configured database.
I guess we could just ignore the failures and just rely on the dynamic
worker.
I ended up with these two patches. I'm not sure about pushing
separately. It seems pointless to backport the "fix" to back branches
anyway.
--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services