Config

Table of Contents

• Config

  • Table of Contents

  • Overview

  • Partitions

  • Converged/Hybrid

Overview

When using slurm-bridge, there are some configuration requirements and considerations to be made.

Partitions

Slurm bridge placeholder jobs are submitted to a default partition (e.g. slurm-bridge) in Slurm, or the one specified by slurmjob.slinky.slurm.net/partition on the workload. Any partition where these placeholder jobs are submitted to should only consist of Slurm nodes that map to Kubernetes nodes, be them nodes with co-located kubelet and slurmd, or Kubernetes nodes modeled as Slurm external nodes.

Converged/Hybrid

When you have kubelet and slurmd daemons running side-by-side, you may want to isolate their workloads to allow the physical node to dynamically switch workloads types.

This can be achieved by enabling MCS in Slurm.

# slurm.conf
...
MCSPlugin=mcs/label
MCSParameters=ondemand,ondemandselect