Mount cloud disks

Step 1: Verify the AI node state

Run the following command to view the nodes:

kubectl get nodes

Make sure that the target AI node is in the Ready state. If GPUs are required for your business, also verify that the node has sufficient allocatable GPU resources.

Step 2: Create a PVC

AI node cloud disks use the following StorageClass:

aliclouddb-vnode-disk-essd

1. Create a file named ai-node-disk-pvc.yaml:

   apiVersion: v1
   kind: PersistentVolumeClaim
   metadata:
     name: ai-node-disk
     namespace: default
   spec:
     accessModes:
       - ReadWriteOnce
     resources:
       requests:
         storage: 100Gi
     storageClassName: aliclouddb-vnode-disk-essd
     volumeMode: Filesystem

   To request a different capacity, modify resources.requests.storage. The requested capacity cannot exceed the remaining capacity of the cloud disk storage plan.

2. Apply the PVC:

   kubectl apply -f ai-node-disk-pvc.yaml
   kubectl get pvc ai-node-disk -n default

   The expected PVC status is Bound. You can also run the following command to view the details:

   kubectl describe pvc ai-node-disk -n default

   Note

   AI node cloud disks are provided by the platform. A Kubernetes StorageClass object with the same name may not exist in the ACK cluster. Use whether the PVC enters the Bound state as the criterion for successful cloud disk binding.

Step 3: Create a Pod and mount the PVC

1. Create a file named ai-node-disk-pod.yaml and replace the following content:

   • Replace <AI_NODE_ID> with the target AI node ID.

   • Replace <IMAGE> with your business image.

   • If you use a different namespace, modify the namespace in the Pod, PVC, and commands accordingly.

   apiVersion: v1
   kind: Pod
   metadata:
     name: ai-node-disk-example
     namespace: default
   spec:
     restartPolicy: Never
     nodeSelector:
       rm.alibaba-inc.com/vnodeId: <AI_NODE_ID>
     tolerations:
       - key: virtual-kubelet.io/provider
         operator: Equal
         value: aliclouddb
         effect: NoSchedule
     containers:
       - name: app
         image: <IMAGE>
         command:
           - /bin/sh
           - -c
           - sleep 3600
         resources:
           requests:
             cpu: "1"
             memory: 2Gi
           limits:
             cpu: "1"
             memory: 2Gi
         volumeMounts:
           - name: data-disk
             mountPath: /data
     volumes:
       - name: data-disk
         persistentVolumeClaim:
           claimName: ai-node-disk

   If your business requires GPUs, add the following line to both resources.requests and resources.limits:

   nvidia.com/gpu: "1"

2. Apply the Pod and check the scheduling result:

   kubectl apply -f ai-node-disk-pod.yaml
   kubectl get pod ai-node-disk-example -n default -o wide

   Make sure that the Pod is in the Running state and that the NODE column shows the target AI node.

Verify that the cloud disk is mounted

Run the following commands to view the mounted device and file system:

kubectl exec ai-node-disk-example -n default -- findmnt -T /data
kubectl exec ai-node-disk-example -n default -- df -hT /data

If the output includes an independent block device, a file system type, and the /data mount point, the cloud disk is mounted.

Verify that the cloud disk is readable and writable

Run the following command to write and read a test file:

kubectl exec ai-node-disk-example -n default -- sh -c \
  'echo ai-node-disk-test > /data/persist-marker.txt && sync && cat /data/persist-marker.txt'

If ai-node-disk-test is returned, the cloud disk supports normal read/write operations.

Verify that data persists after the Pod is rebuilt

1. Delete the test Pod. Deleting the Pod does not delete the PVC:

   kubectl delete pod ai-node-disk-example -n default

2. Recreate the Pod:

   kubectl apply -f ai-node-disk-pod.yaml
   kubectl wait --for=condition=Ready pod/ai-node-disk-example -n default --timeout=300s

3. Read the test file:

   kubectl exec ai-node-disk-example -n default -- cat /data/persist-marker.txt

   If ai-node-disk-test is still returned, the data persists after the Pod is rebuilt.

The PVC remains in the Pending state

Check the following items in order:

• Whether a cloud disk storage plan is purchased.

• Whether the remaining capacity of the storage plan is sufficient.

• Whether the cloud disk mounting capability is enabled on the AI node.

• Whether storageClassName is set to aliclouddb-vnode-disk-essd.

• Whether the PVC and the target AI node are in the same zone.

• Whether PVC events indicate insufficient capacity, invalid parameters, or permission errors.

View the details:

kubectl describe pvc ai-node-disk -n default

The Pod is not scheduled to the target AI node

Check the following items:

• Whether rm.alibaba-inc.com/vnodeId is set to the correct AI node ID.

• Whether a toleration for the aliclouddb provider is configured.

• Whether the target AI node is in the Ready state.

• Whether the node has sufficient CPU, memory, and GPU resources to meet the Pod requests.

• Whether the Pod and the cloud disk are in the same zone.

View the Pod and node details:

kubectl describe pod ai-node-disk-example -n default
kubectl get node <AI_NODE_ID>