Overview

The YuniKorn scheduler is able to run any Kubernetes workload. All that is required is to ensure that the schedulerName field of a Pod specification is set to yunikorn and an applicationId label is set to a unique value per application:

apiVersion: v1
kind: Pod
metadata:
  labels:
    app: sleep
    applicationId: "application-sleep-0001"
  name: sleep-app-1
spec:
  schedulerName: yunikorn
  containers:
    - name: sleep-30s
      image: "alpine:latest"
      command: ["sleep", "30"]
      resources:
        requests:
          cpu: "100m"
          memory: "100M"

Additionally, if the YuniKorn admission controller is present, the schedulerName field may be omitted as it will be set automatically on newly created pods.

Advanced Examples

Examples of more advanced use cases can be found here:

• Run NVIDIA GPU Jobs
• Run Spark Jobs
• Run Flink Jobs
• Run TensorFlow Jobs
• Run MPI Jobs