[slurm-users] Additional Container Runtime (`scrun`) Commands

[Jesse Farebrother]

Hi,

I just finished testing out the new container support in 23.02. It’s really exciting the progress that’s been made so far!

Sadly, most tooling doesn’t respect the OCI runtime specification and instead we live in a reality where most people define the spec in terms of the major implementations like runc. This will be an obsticle to getting support for tools that build upon the major OCI-compatible runtimes as they all support `exec` amongst other commands.  `exec` was once implemented as part of the runtime specification but was removed in 2016 (https://github.com/opencontainers/runtime-spec/pull/388). I believe `exec` might be the biggest blocker currently and I was wondering if there were plans to support expanded functionality like `exec` in `scrun`?

Perhaps it might even be possible to implement `exec` with existing infrastructure. You should be able to implement `exec` with `state` and `start` (e.g., https://github.com/opencontainers/runtime-spec/pull/391/files). From there is it possible to run a job step inside of an existing job (perhaps using `srun —jobid $JOB_ID_FROM_CONTAINER_ID`?) to start the new process?

Jesse
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20230314/44b57474/attachment.htm>