This message was deleted.

Can you guys replicate this?

Thanks Tommaso. I haven't done much with eval but my first guess is it's by design... if not the best design. Will tag someone in the morning who has worked with this more

Thanks for getting back! Right now, I’ve implemented an ugly workaround: wrapping the run_evaluation function in a try/except block, and in case of failure, calling docker kill container_name

@wide-doctor-37830 is this an area you're familiar with?

Only tangentially. I usually use the evaluation scripts with the remote runtime, and part of that workflow involves calling a separate

cleanup_remote_runtimes.py

script that shuts down any hanging containers. I don't think the design is intentional so much as it's a pain point that was easier to work around than to fix. It's worth a deeper investigation though.

So it sounds like this is in the realm of "contributions welcome"