The gist is that a pipeline can be (re)started (usually, only failed pipelines are manually restarted, hoping for a better result as if the error was floating somewhere around the Ether and not our code most of the times) in debug mode.
After a pipeline fails in said debug mode, a
tmate client connection to a
tmate server (as configured in the
configuration files -
is open from withing the container that the failing step was running in and
connection instructions for
ssh-ing are displayed.
The error can then be debugged more intimately in the precise environment that made the bug appear.
Instead of repeating any more of what’s already been said, I suggest anybody interested reads the original kind-of-reference post by Brad.