For the minimal set, I think that you should run the main matrix of DB tests against the primary PG version you aim to support. And then you should test at least the core tests on the oldest and newest PG versions you want people to be able to use. This catches “are we using features that don't yet exist in some PG version” and “are we still using a feature that was removed”. Which are the main things you want to prevent. The other pieces of the matrix (like Java version or Ruby version) don't have much if any impact on Postgres behavior, so there's not much benefit to redoing the full matrix