Closed dylex closed 8 months ago
Hello @dylex SLURM batch script template is configurable and we can configure it within the jupyterhub_config.py
file. So you can set it like the way you want in your deployment in your config file instead of changing the default template?
Yes, we already changed this in our config, I just figured it should be the default to prevent others from running into this, too.
I think this would be a breaking change, and since its configurable already I think its not a PR this project ought to merge. I'll go for a close here to avoid leaving it hanging without a decision.
I think this could be a relevant documentation update, but I'm not sure. Thank you @dylex and @mahendrapaipuri for working this!!
We've had a few cases where requests for multiple GPUs ended up allocating multiple nodes, which obviously causes weird problems. Obviously we could configure our partition and options to prevent this, but since in general I can't think of any case where you'd want to run multiple tasks or nodes, it seems nice to constrain it here (PBS already seems to do this). Either
--ntasks=1
or--nodes=1
should be enough but both shouldn't hurt. (If I'm wrong and there's some use cases where you want -N2 or -n2 or something please correct me.)