Container Service for Kubernetes:Create a group snapshot for multiple Cloud Disk volumes

Create a VolumeGroupSnapshotClass

A default VolumeGroupSnapshotClass named alibabacloud-disk-group-snapshot is already available in the cluster. To use a custom class, create group-snapshot-class-demo.yaml:

apiVersion: groupsnapshot.storage.k8s.io/v1alpha1
kind: VolumeGroupSnapshotClass
metadata:
  name: group-snapshot-class-demo
deletionPolicy: Delete
driver: diskplugin.csi.alibabacloud.com

The deletionPolicy field controls what happens to the underlying ECS snapshot-consistent group when you delete the VolumeGroupSnapshot:

Apply the class:

kubectl apply -f group-snapshot-class-demo.yaml

Create a VolumeGroupSnapshot

Create group-snapshot-demo.yaml to snapshot both PVCs mounted to the MySQL application:

apiVersion: groupsnapshot.storage.k8s.io/v1alpha1
kind: VolumeGroupSnapshot
metadata:
  name: group-snapshot-demo
  namespace: default
spec:
  source:
    selector:
      matchLabels:
        app: mysql           # Selects all PVCs with this label
  volumeGroupSnapshotClassName: group-snapshot-class-demo

The source.selector field uses a label selector to identify the PVCs to back up. PVCs created from a StatefulSet's volumeClaimTemplates are automatically labeled with the StatefulSet's pod labels. If you created your PVCs manually, add the corresponding labels before running this step.

Apply the VolumeGroupSnapshot:

kubectl apply -f group-snapshot-demo.yaml

Option 1: Restore a single volume

The following example restores the disk volume attached to the disk-mysql-0 pod.

Create disk-mysql-0-copy.yaml. The dataSource field tells CSI to provision a new disk from the specified VolumeSnapshot:

kind: PersistentVolumeClaim
apiVersion: v1
metadata:
  name: disk-mysql-0-copy
spec:
  volumeMode: Filesystem
  accessModes:
  - ReadWriteOnce
  resources:
    requests:
      storage: 20Gi
  storageClassName: "alicloud-disk-essd"
  dataSource:
    name: snapshot-1c2c5bcaf47ee2bffcc5b2f52dff65a4aacaaea38032c05d75acd536f7xxxxxx-2024-11-27-6.4.7  # [1]
    kind: VolumeSnapshot                                                                                  # [2]
    apiGroup: snapshot.storage.k8s.io                                                                    # [3]

• [1] The name of the VolumeSnapshot that backs up disk-mysql-0 (from the kubectl describe vgs output above).

• [2] The resource kind — always VolumeSnapshot when restoring from a snapshot.

• [3] The API group for the CSI snapshot resources.

Apply the PVC:

kubectl apply -f disk-mysql-0-copy.yaml

Verify the PVC is bound:

kubectl get pvc disk-mysql-0-copy

Expected output:

disk-mysql-0-copy   Bound    d-2ze0iwwqg0s6b0xxxxxx   20Gi   RWO   alicloud-disk-essd   <unset>   69s

To confirm the disk was created from the snapshot, go to the ECS console, choose Storage & Snapshots > Block Storage in the left-side navigation pane, click the Cloud Disk tab, and search for d-2ze0iwwqg0s6b0xxxxxx. Open the disk details page to see that it was created from a snapshot.

Option 2: Restore all volumes in one batch

Use the generate_pvc.sh script to automatically generate PVC manifests from a VolumeGroupSnapshot and apply them all at once.

1. Install the jq command-line tool:

   • CentOS: yum install jq

   • Ubuntu: apt-get install jq

2. Scale the MySQL StatefulSet to zero replicas to stop all writes before restoring:

   kubectl scale sts mysql --replicas=0

3. Delete the existing PVCs:

   kubectl delete pvc data-mysql-0 data-mysql-1

4. Create generate_pvc.sh with the following content: <details><summary>generate_pvc.sh</summary>

   Parameter	Description	Example
   1	Namespace of the VolumeGroupSnapshot	default
   2	Name of the VolumeGroupSnapshot	group-snapshot-demo
   3	StorageClass name for the restored PVCs	alicloud-disk-essd
   4	Disk capacity	20Gi
   5	Path to the kubeconfig file	.kube/config
   6	Output path for the generated PVC YAML	./output.yaml

   #!/bin/bash
   
   # Input parameters
   NAMESPACE=$1           # Namespace of the VolumeGroupSnapshot
   VGS_NAME=$2            # Name of the VolumeGroupSnapshot
   STORAGE_CLASS_NAME=$3  # StorageClass for the restored PVCs
   CAPACITY=$4            # Disk capacity (e.g., 20Gi)
   KUBECONFIG_PATH=$5     # Path to the kubeconfig file
   OUTPUT_FILE=$6         # Output path for the generated YAML
   
   # Get VolumeGroupSnapshot details
   VGS_INFO=$(kubectl --kubeconfig=${KUBECONFIG_PATH} -n ${NAMESPACE} get vgs ${VGS_NAME} -o json)
   
   # Validate that the PVC-snapshot mapping exists
   if ! echo ${VGS_INFO} | jq -e '.status.pvcVolumeSnapshotRefList' &>/dev/null; then
     echo "Error: .status.pvcVolumeSnapshotRefList not found in VolumeGroupSnapshot."
     exit 1
   fi
   
   # Extract PVC names and their corresponding snapshot names
   PVCS=($(echo ${VGS_INFO} | jq -r '.status.pvcVolumeSnapshotRefList[].persistentVolumeClaimRef.name'))
   SNAPSHOTS=($(echo ${VGS_INFO} | jq -r '.status.pvcVolumeSnapshotRefList[].volumeSnapshotRef.name'))
   
   # Clear the output file
   > ${OUTPUT_FILE}
   
   # Generate one PVC manifest per disk
   for i in "${!PVCS[@]}"; do
     PVC_NAME=${PVCS[$i]}
     SNAPSHOT_NAME=${SNAPSHOTS[$i]}
   
     cat <<EOF >> ${OUTPUT_FILE}
   ---
   kind: PersistentVolumeClaim
   apiVersion: v1
   metadata:
     name: ${PVC_NAME}
   spec:
     volumeMode: Filesystem
     accessModes:
     - ReadWriteOnce
     resources:
       requests:
         storage: ${CAPACITY}
     storageClassName: "${STORAGE_CLASS_NAME}"
     dataSource:
       name: ${SNAPSHOT_NAME}
       kind: VolumeSnapshot
       apiGroup: snapshot.storage.k8s.io
   EOF
   done
   
   echo "PVC YAML files have been written to ${OUTPUT_FILE}"

   </details> The script accepts six positional parameters:

5. Run the script to generate the PVC manifests:

   bash generate_pvc.sh default group-snapshot-demo alicloud-disk-essd 20Gi .kube/config ./output.yaml

   The generated output.yaml looks similar to:

   ---
   kind: PersistentVolumeClaim
   apiVersion: v1
   metadata:
     name: disk-mysql-0
   spec:
     volumeMode: Filesystem
     accessModes:
     - ReadWriteOnce
     resources:
       requests:
         storage: 20Gi
     storageClassName: "alicloud-disk-essd"
     dataSource:
       name: snapshot-1c2c5bcaf47ee2bffcc5b2f52dff65a4aacaaea38032c05d75acd536f7adb850-2024-11-27-6.4.7
       kind: VolumeSnapshot
       apiGroup: snapshot.storage.k8s.io
   ---
   kind: PersistentVolumeClaim
   apiVersion: v1
   metadata:
     name: disk-mysql-1
   spec:
     volumeMode: Filesystem
     accessModes:
     - ReadWriteOnce
     resources:
       requests:
         storage: 20Gi
     storageClassName: "alicloud-disk-essd"
     dataSource:
       name: snapshot-37a2fbf634d68cd2103f261313c2ed781fbd2bd52b5a0d0e0c0ef7c3396fea1c-2024-11-27-6.4.9
       kind: VolumeSnapshot
       apiGroup: snapshot.storage.k8s.io

   Edit output.yaml if needed, then apply it:

   kubectl apply -f output.yaml

6. Scale the StatefulSet back to two replicas:

   kubectl scale sts mysql --replicas=2

7. After the pods restart, verify that the data is restored:

   kubectl exec -it mysql-0 -- bash   # Open a shell in the pod
   mysql -uroot -p123456              # Connect to MySQL
   use test;
   select * from scores;

   Expected output:

   +------+-------+
   | name | score |
   +------+-------+
   | Amy  |    95 |
   +------+-------+