HPC batch system on Kubernetes?

The HPC facilities I manage run SLURM, and we happily run JupyterHub with batchspawner on top. However my experience with z2jh has shown me that kubernetes can host much more diverse workloads than classic batch submission systems. For those running jupyterhub and batchspawner, are you aware of, and/or have you considered, replacing your classic batch submission system with something like Kueue?

One could theoretically substitute SLURM for Kqueue, and then run batchspawner+Kqueue or z2jh or both.

2 Likes