fast-socket-installer/fast-socket-installer.yaml

apiVersion: apps/v1
kind: DaemonSet
metadata:
  name: nccl-fastsocket-manual-installer
  namespace: kube-system
  labels:
    k8s-app: nccl-fastsocket-manual-installer
spec:
  selector:
    matchLabels:
      k8s-app: nccl-fastsocket-manual-installer
  updateStrategy:
    type: RollingUpdate
  template:
    metadata:
      labels:
        name: nccl-fastsocket-manual-installer
        k8s-app: nccl-fastsocket-manual-installer
    spec:
      affinity:
        nodeAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:
            nodeSelectorTerms:
            - matchExpressions:
              - key: cloud.google.com/gke-accelerator
                operator: Exists
              - key: cloud.google.com/gke-nccl-fastsocket
                operator: Exists
      tolerations:
      - operator: "Exists"
      hostNetwork: true
      hostPID: true
      volumes:
      - name: nvidia-install-lib64-dir-host
        hostPath:
          path: /home/kubernetes/bin/nvidia/lib64
          type: DirectoryOrCreate
      initContainers:
      - image: gcr.io/gke-release/fastsocket-installer@sha256:cb8dca70b5611769fd2e0e8eb9aebf81a89d4378537cff104775c873abf2d9c5
        name: nccl-fastsocket-installer
        command:
        - bash
        - -c
        - |
          cp /usr/lib/libnccl-net.so $NCCL_INSTALL_DIR/
        securityContext:
          privileged: true
        resources:
          limits:
            memory: 100Mi
        env:
        - name: NCCL_INSTALL_DIR
          value: /usr/local/nvidia/lib64
        volumeMounts:
        - name: nvidia-install-lib64-dir-host
          mountPath: /usr/local/nvidia/lib64
      containers:
      - image: "gke.gcr.io/pause:3.8@sha256:880e63f94b145e46f1b1082bb71b85e21f16b99b180b9996407d61240ceb9830"
        name: pause