Open reuben opened 4 years ago
How about something like this?
Would that work if I pit stop
a job?
I'll play around with the suggestions in that SO and close the issue if they solve my case. In particular trap
might be enough since it catches signals.
Ah - you are right - I just had the failure case in mind. No, am pretty sure it won't execute.
Doing things inside
/data/rw/pit
in interactive jobs is very painful because of sshfs, agit status
can take tens of seconds to complete. But doing things outside of the snakepit mounts means risking losing data if something goes wrong unexpectedly and your job gets stopped/killed, or if you stop it and forget to copy things first.If we could provide a cleanup script that is executed as part of job end, then I could copy any critical folders (checkpoints, etc) into networked folders to make sure nothing is forgotten.