mpi-operator
mpi-operator copied to clipboard
How can I use a different scheduler from volcano?
I want to use other scheduler (my own scheduler and etc). How can I configure mpi-operator to use other scheduler (like changing some parameter)?
You can set the schedulerName
within the Pod templates of your MPIJobs.
There is this command line flag https://github.com/kubeflow/mpi-operator/blob/993b010e05c48c6f6ef5a5180362ccd3e716982e/v2/cmd/mpi-operator/app/options/options.go#L65, however, it assumes that you want to use Volcano's PodGroup API, even if using a different scheduler.
There are also these open issues https://github.com/kubeflow/training-operator/issues/1519 to integrate with queuing systems instead. But I don't think anybody is working on them at the moment https://github.com/kubernetes-sigs/kueue/issues/65.
I would welcome contributions to mpi-operator and kueue, if you would like to contribute.
Thanks. I'll have a try.
And the job-manager
introduced by kueue
seems very promising. I've always wanted a job-level
manager (argo is not) instead of pod-level
manager and to avoid duplicating mature functionality in k8s.
/close
@alculquicondor: Closing this issue.
In response to this:
/close
Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.