Hi All,
Just wanted to see if I am right about this:
OnDemand is running in a vm that does not have the scheduler installed on it (lsf in this case) so for OOD to submit/query lsf I will need to set “bin_overrides” in my cluster config to wrapper scripts for the various lsf commands that ssh into a login node that does have lsf to run.
Right?
Thanks,
Morgan
Correct, what you have described is the current suggested solution.
In OnDemand 1.7 or 1.8 (still undecided) we should support a config option to execute these commands via ssh on the specified host, at least for Slurm and LSF.
There is an issue open for this that targets the public 1.7.x release, but I don’t think that was intended as a promise: https://github.com/OSC/ood_core/issues/170.
Thank you both!
After reading through a fair amount of the ood_core source code, I can see it would be a bit of a pain to add this in (properly, not hacky). Wrapper scripts will be fine for now.
FWIW, The reason I am in this weird scenario is the cluster in question is running on Power9 architecture so to get OOD involved, I have a single x86 node that runs it. I figure sending the jobs to one of the power9 head-nodes is easier than trying to get lsf installed on the x86 node and playing nice, even if it is sshing into the head-node each time it needs to interact with it.
Best,
Morgan