EKS/AKS 上的 Automation Suite 安装指南

在 Kubernetes 中启用已启用 MIG 的 GPU

• 选项 A：遵循云提供商的官方 GPU 设置文档：
  • Azure Kubernetes 服务 (AKS)
  • Amazon Elastic Kubernetes Service (Amazon EKS)
• 选项 B（可选）：直接部署 NVIDIA 设备插件：
  1. 新建命名空间：
     assignment

     kubectl create namespace gpu-resourceskubectl create namespace gpu-resources

  2. 应用以下配置，并将migEnabledPoolName替换为与您的 GPU 节点匹配的标签：
     assignment

     apiVersion: v1
     kind: Pod
     metadata:
       name: nvidia-device-plugin-pod
       namespace: gpu-resources
     spec:
      affinity:
         nodeAffinity:
           requiredDuringSchedulingIgnoredDuringExecution:
             nodeSelectorTerms:
             - matchExpressions:
               - key: agentpool
                 operator: In
                 values:
                 # To be changed to a selector that matches the GPU nodes
                 - migEnabledPoolName
      containers:
      - args:
        - --fail-on-init-error=false
        env:
        - name: MPS_ROOT
          value: /run/nvidia/mps
        - name: MIG_STRATEGY
           # We only support the single strategy for now
          value: single
        - name: NVIDIA_MIG_MONITOR_DEVICES
          value: all
        - name: NVIDIA_VISIBLE_DEVICES
          value: all
        - name: NVIDIA_DRIVER_CAPABILITIES
          value: compute,utility
        image: nvcr.io/nvidia/k8s-device-plugin:v0.17.3
        imagePullPolicy: IfNotPresent
        name: nvidia-device-plugin-ctr
        securityContext:
          allowPrivilegeEscalation: true
          capabilities:
            add:
            - SYS_ADMIN
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: File
        volumeMounts:
        - mountPath: /var/lib/kubelet/device-plugins
          name: device-plugin
      tolerations:
      - key: CriticalAddonsOnly
        operator: Exists
      - effect: NoSchedule
        key: nvidia.com/gpu
        operator: Exists
      terminationGracePeriodSeconds: 30
      volumes:
      - hostPath:
          path: /var/lib/kubelet/device-plugins
          type: ""
        name: device-pluginapiVersion: v1
     kind: Pod
     metadata:
       name: nvidia-device-plugin-pod
       namespace: gpu-resources
     spec:
      affinity:
         nodeAffinity:
           requiredDuringSchedulingIgnoredDuringExecution:
             nodeSelectorTerms:
             - matchExpressions:
               - key: agentpool
                 operator: In
                 values:
                 # To be changed to a selector that matches the GPU nodes
                 - migEnabledPoolName
      containers:
      - args:
        - --fail-on-init-error=false
        env:
        - name: MPS_ROOT
          value: /run/nvidia/mps
        - name: MIG_STRATEGY
           # We only support the single strategy for now
          value: single
        - name: NVIDIA_MIG_MONITOR_DEVICES
          value: all
        - name: NVIDIA_VISIBLE_DEVICES
          value: all
        - name: NVIDIA_DRIVER_CAPABILITIES
          value: compute,utility
        image: nvcr.io/nvidia/k8s-device-plugin:v0.17.3
        imagePullPolicy: IfNotPresent
        name: nvidia-device-plugin-ctr
        securityContext:
          allowPrivilegeEscalation: true
          capabilities:
            add:
            - SYS_ADMIN
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: File
        volumeMounts:
        - mountPath: /var/lib/kubelet/device-plugins
          name: device-plugin
      tolerations:
      - key: CriticalAddonsOnly
        operator: Exists
      - effect: NoSchedule
        key: nvidia.com/gpu
        operator: Exists
      terminationGracePeriodSeconds: 30
      volumes:
      - hostPath:
          path: /var/lib/kubelet/device-plugins
          type: ""
        name: device-plugin