Adding nodes from a Supercomputing Cluster

I would like to run Kubernetes on a local VM/host and add compute nodes allocated via slurm job allocation on the remote supercomputer. The compute nodes on the remote supercomputer are accessible on the local host by first login to the supercomputer through ssh, which ends up on the login node of supercomputer, and then ssh to the compute node.

Is there already some tool that can be leveraged for this kind of usage, or if are there some relevant directions on how to achieve it.

Sylabs was working on something that played well with slurm before their acquisition, but afaik nothing like that exists today. You could look into extending Virtual Kubelet, but there is nothing out there that won’t require a good amount of work.

1 Like

Thanks Bob,

I was looking at Torque-Operator paper based on sylabs / wlm-operator and investigating its usefulness. I will take a look at Virtual Kubelet.

I just recalled a talk from KubeCon last year that might be useful for you - they’re using pbspro for scheduling though:

2 Likes

Thanks Bob for sharing the URL. It seems they are designing cluster completely based on K8s. I am not sure about their current status and experiences so far.