We just copy a schema run test on a copy, and drop it afterwards. It can run in parallel if you name schemas randomly, or by test name. You only need one DB,

We just copy a schema run test on a copy, and drop it afterwards. It can run in parallel if you name schemas randomly, or by test name. You only need one DB, no need to even re-connect, start DB cluster or whatever. It can all work with a single persistent connection. And it's much less resource intensive.