Default Kueue Usage Example

This example demonstrates how to use Kueue with HAMi vGPU resources. It includes a complete configuration that sets up ResourceFlavor, ClusterQueue, LocalQueue, and a sample Deployment that requests vGPU resources.

Before applying this example, ensure that HAMi and Kueue are installed, and Kueue is configured with ResourceTransformation enabled (see How to use Kueue on HAMi).

apiVersion: kueue.x-k8s.io/v1beta1
kind: ResourceFlavor
metadata:
  name: hami-flavor
spec:
  nodeLabels:
    gpu: "on"
---
apiVersion: kueue.x-k8s.io/v1beta1
kind: ClusterQueue
metadata:
  name: hami-queue
spec:
  namespaceSelector: {}
  resourceGroups:
  - coveredResources:
    - nvidia.com/gpu
    - nvidia.com/total-gpucores
    - nvidia.com/total-gpumem
    flavors:
    - name: hami-flavor
      resources:
      - name: nvidia.com/gpu
        nominalQuota: 20
      - name: nvidia.com/total-gpucores
        nominalQuota: 600
      - name: nvidia.com/total-gpumem
        nominalQuota: 20480
---
apiVersion: v1
kind: Namespace
metadata:
  name: kueue-test
---
apiVersion: kueue.x-k8s.io/v1beta1
kind: LocalQueue
metadata:
  name: user-queue
  namespace: kueue-test
spec:
  clusterQueue: "hami-queue"
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: gpu-burn
  namespace: kueue-test
  labels:
    kueue.x-k8s.io/queue-name: user-queue
spec:
  replicas: 1
  selector:
    matchLabels:
      app-name: gpu-burn
  template:
    metadata:
      labels:
        app-name: gpu-burn
    spec:
      containers:
      - args:
        - while :; do /app/gpu_burn 300 || true; sleep 300; done
        command:
        - /bin/sh
        - -lc
        image: oguzpastirmaci/gpu-burn:latest
        imagePullPolicy: IfNotPresent
        name: main
        resources:
          limits:
            nvidia.com/gpu: "2"        # requesting 2 vGPU instances
            nvidia.com/gpucores: "30"  # 30 cores per vGPU
            nvidia.com/gpumem: "1024"  # 1024 MiB per vGPU