[slurm-users] Using Podman containers with scrun (Slurm 23.02)

Joel Loo loo.joel93 at gmail.com
Wed May 31 06:22:17 UTC 2023


I am helping to manage a small 5-node cluster at a research lab (PhD
student), and am exploring using Podman containers on Slurm. I am
interested in taking advantage of the latest Slurm version's inbuilt
support for containerization (with scrun), but am confused as to how to
manage resources when using it.

During my explorations, I configured Podman with scrun based on these
instructions <https://slurm.schedmd.com/containers.html>, and attempted to
do the following:

   - Execute 'podman run' commands after configuring: These commands
   successfully started Slurm jobs and ran the containers in them. However,
   these jobs were started with the default resources.
   - Attempted to export SCRUN_* environment variables (e.g. export
   SCRUN_JOB_NODELIST=xxx && podman run --env 'SCRUN*' ...): This led to a
   generic 'Error: container create failed (no logs from conmon): EOF'.

Based on my readings of the Slurm documentation, it seems that exporting
the SCRUN environment variables is the current optimal manner of specifying
the resources that scrun should use. If this is true, would anyone have
suggestions on where my current attempt to pass the environment variables
is going wrong? If it is not true, what would be the right way to handle
resource management when running a job through podman?

Thanks for your help!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20230531/f7a8f8db/attachment-0001.htm>

More information about the slurm-users mailing list