17.2. 由配额管理的资源

流程

1. 确定集群中某个节点中有多少 GPU 可用。例如：

   # oc describe node ip-172-31-27-209.us-west-2.compute.internal | egrep 'Capacity|Allocatable|gpu'
                       openshift.com/gpu-accelerator=true
   Capacity:
    nvidia.com/gpu:  2
   Allocatable:
    nvidia.com/gpu:  2
     nvidia.com/gpu  0           0

   本例中有 2 个 GPU 可用。

2. 在命名空间 nvidia 中设置配额。本例中配额为 1：

   # cat gpu-quota.yaml
   apiVersion: v1
   kind: ResourceQuota
   metadata:
     name: gpu-quota
     namespace: nvidia
   spec:
     hard:
       requests.nvidia.com/gpu: 1

3. 创建配额：

   # oc create -f gpu-quota.yaml
   resourcequota/gpu-quota created

4. 验证命名空间是否设置了正确的配额：

   # oc describe quota gpu-quota -n nvidia
   Name:                    gpu-quota
   Namespace:               nvidia
   Resource                 Used  Hard
   --------                 ----  ----
   requests.nvidia.com/gpu  0     1

5. 运行一个请求单个 GPU 的 Pod：

   # oc create pod gpu-pod.yaml

   apiVersion: v1
   kind: Pod
   metadata:
     generateName: gpu-pod-
     namespace: nvidia
   spec:
     restartPolicy: OnFailure
     containers:
     - name: rhel7-gpu-pod
       image: rhel7
       env:
         - name: NVIDIA_VISIBLE_DEVICES
           value: all
         - name: NVIDIA_DRIVER_CAPABILITIES
           value: "compute,utility"
         - name: NVIDIA_REQUIRE_CUDA
           value: "cuda>=5.0"
   
       command: ["sleep"]
       args: ["infinity"]
   
       resources:
         limits:
           nvidia.com/gpu: 1

6. 验证 Pod 是否在运行：

   # oc get pods
   NAME              READY     STATUS      RESTARTS   AGE
   gpu-pod-s46h7     1/1       Running     0          1m

7. 验证配额计数器 Used 是否正确：

   # oc describe quota gpu-quota -n nvidia
   Name:                    gpu-quota
   Namespace:               nvidia
   Resource                 Used  Hard
   --------                 ----  ----
   requests.nvidia.com/gpu  1     1

8. 尝试在 nvidia 命名空间中创建第二个 GPU Pod。从技术上讲，该节点支持 2 个 GPU，因为它有 2 个 GPU：

   # oc create -f gpu-pod.yaml
   Error from server (Forbidden): error when creating "gpu-pod.yaml": pods "gpu-pod-f7z2w" is forbidden: exceeded quota: gpu-quota, requested: requests.nvidia.com/gpu=1, used: requests.nvidia.com/gpu=1, limited: requests.nvidia.com/gpu=1

   应该会显示此 Forbidden 错误消息，因为您有设为 1 个 GPU 的配额，但这一 Pod 试图分配第二个 GPU，而这超过了配额。