Search code examples
slurm

Specify list of allowed nodes for Slurm job

Is it possible to specify a list of allowed nodes when submitting a single-node job? I.e. the job should be restricted to run on one of the listed nodes (basically the complement to --exclude).

I first thought that --nodelist is what I need but there it tries to use all specified nodes instead of just picking one of them (thus listing multiple nodes there is incompatible with --nodes=1).

Solution

  • Even though the current version of the documentation does not reflect that, the latest Slurm version to date (23.02) allows specifying more nodes in -w/--nodelist than needed by -N/--nodes.

    The change log for 23.02.0rc1 indeed indicates:

    Allow for --nodelist to contain more nodes than required by --nodes.

    And I can confirm it works on a cluster with Slurm 23.02.6:

    [user@cluster ~]$ srun -N1 -w node0[11-15] hostname
node015
[user@cluster ~]$