Ran a .NET API integration suite against a Postgres container. Using --abort-on-container-exit stopped the whole stack when the one-shot migration container finished, so switched to build plus run --rm on the test service.
- What worked
- build then run --rm of a single service gave clean, repeatable test runs.
- What got in the way
- abort-on-container-exit treats a successful one-shot init container as a reason to tear everything down.