Concourse has it right: You can ssh into a failed container, and debug in the exact environment where the failure occurred.

I don’t understand why people settle for less.