Dask_jobqueue: Using Cluster from within container without e.g. sbatch

So I’m trying to run a SLURMCluster from within a singularity container. Obviously, once I actually want to do something, dask_jobqueue can’t find the sbatch executable.

I’ve seen this thread: https://groups.google.com/a/lbl.gov/g/singularity/c/IGvUup1aGXE and have tried to mount the path where the sbatch binary resides and add it into the PATH, but have quickly run into missing library problems.

Does anyone else have experience with this? I would imagine it must be a quite common problem.

Thanks for any hints!

For future reference, there’s also this (stale?) issue: Add RemoteSlurmJob to connect SLURMCluster to a remote Slurm cluster · Issue #502 · dask/dask-jobqueue · GitHub

Hi @poplarShift,

We’re doing that on our cluster, but with PBS. The only thing we do in the images that must use qsub equivalent to sbatch is installing the PBS Pro client and give it the configuration to our PBS Pro master servers.

I’m not sure but this might have required some privilege elevation (or linux capabilities) for the client to contact the masters.

Couldn’t you try to install Slurm client into your Singularity image?

1 Like