Skip to content

Runners

Designed: SLURM and the laptop first, Kubernetes next

A runner is a small program that runs next to the compute. It starts the engine, watches the scheduler, and reports to the server.

  • SLURM and other HPC schedulers: the runner sits on a login node. It shows why jobs wait, what they used, and why they failed, and suggests what to change when a job ran out of memory or time.
  • Kubernetes: the runner runs in the cluster, with Kueue for queues.
  • A laptop for small runs and tests.
  • No inbound ports. The runner only makes outgoing HTTPS calls to the server.
  • Checks before launch: walltimes against partition limits, storage quotas, scratch purge rules.

Environments: SLURM and Kubernetes runners

Design details: schedulers and monitoring.