Options

• project [required]: Used for correct budgeting of compute time.

• partition [required]: specify the slurm partition

• nodes [required]: specify the number of nodes

• runtime [required]: specify the runtime in minutes

• kernel_argv [required]: copy & paste this from the kernel you want to run

• gpus [optional]: specify the number of GPUs, if the partition supports GPUs

• reservation [optional]: specify the slurm reservation, if you have one

• jobid [optional]: specify a pre-existing slurm allocation and start your kernel there. Without this, a new allocation for each kernel will be created.

• node [optional]: specify a node in your pre-existing allocation

Example kernel.json

.local/share/jupyter/kernels/slurmel/kernel.json

{
    "display_name": "Slurm Kernel",
    "language": "python",
    "metadata": {
        "debugger": true,
        "kernel_provisioner": {
            "config": {
                "gpus": "0",
                "nodes": "2",
                "partition": "batch",
                "project": "...",
                "reservation": "None",
                "runtime": 30,
                "jobid": "None",
                "node": "None",
                "kernel_argv": [
                    "~/.local/share/jupyter/kernels/my_kernel/kernel.sh",
                    "-f",
                    "{connection_file}"
                ]
            },
            "provisioner_name": "slurm-provisioner"
        }
    }
}