Edit

Dynamically provision Azure Managed Lustre volumes for AKS

Dynamic provisioning creates a new Azure Managed Lustre file system when you create a persistent volume claim (PVC). A storage class defines the file system configuration, and the driver provisions the storage and binds a persistent volume (PV) to the claim.

To use a file system you already own, see Static provisioning.

Before you begin

  • Complete Install the Azure Managed Lustre CSI driver, including the verification steps for your chosen installation method. Don't install a second driver if one is already installed. You don't need an existing Lustre file system.
  • Prepare an existing resource group and a subnet for the new file system, separate from the AKS subnet and reachable from your workloads. See Plan network options for Azure Managed Lustre.
  • Grant the identity used by the CSI driver permission to manage file systems and access the subnet. Complete the dynamic provisioning permission requirements before creating the claim.
  • Configure kubectl for your cluster. Create the example PVC and pod in the same namespace.

Important

Creating the claim provisions an Azure resource that incurs charges. This example uses reclaimPolicy: Delete: deleting the claim also deletes the provisioned file system and its data. Use Retain instead if you need to keep the file system after deleting the claim.

Create a storage class

Create a file named azure-managed-lustre-dynamic-storage-class.yaml. Replace the placeholders with your target Azure region, file system resource group, and existing virtual network and Lustre subnet.

Choose a file system configuration and capacity supported in your region. The example uses AMLFS-Durable-Premium-125, availability zone 1, and a Sunday maintenance window. Adjust these values for your deployment. For supported settings, see Dynamic provisioning configuration.

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: azurelustre-dynamic
provisioner: azurelustre.csi.azure.com
parameters:
  sku-name: "AMLFS-Durable-Premium-125"
  zone: "1"
  maintenance-day-of-week: "Sunday"
  maintenance-time-of-day-utc: "22:00"
  location: "<AZURE_REGION>"
  resource-group-name: "<LUSTRE_RESOURCE_GROUP>"
  vnet-resource-group: "<VNET_RESOURCE_GROUP>"
  vnet-name: "<VNET_NAME>"
  subnet-name: "<LUSTRE_SUBNET_NAME>"
reclaimPolicy: Delete
volumeBindingMode: Immediate
mountOptions:
  - noatime
  - flock

Apply the storage class:

kubectl apply -f azure-managed-lustre-dynamic-storage-class.yaml

Create a persistent volume claim

Create a file named azure-managed-lustre-dynamic-pvc.yaml. Adjust the requested capacity for your workload and the selected file system configuration.

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: azurelustre-dynamic-pvc
spec:
  accessModes:
    - ReadWriteMany
  storageClassName: azurelustre-dynamic
  resources:
    requests:
      storage: 48Ti

Apply the claim and watch its status:

kubectl apply -f azure-managed-lustre-dynamic-pvc.yaml
kubectl get pvc azurelustre-dynamic-pvc --watch

File system provisioning can take several minutes. When the claim reaches Bound, press Ctrl+C and continue. If provisioning doesn't complete, inspect the events with kubectl describe pvc azurelustre-dynamic-pvc.

Mount and verify the volume

Create a file named azure-managed-lustre-dynamic-pod.yaml. Run the example on a supported node pool with the driver installed.

apiVersion: v1
kind: Pod
metadata:
  name: azurelustre-dynamic-test
spec:
  containers:
    - name: writer
      image: ubuntu:24.04
      command:
        - /bin/sh
        - -c
        - while true; do date >> /mnt/lustre/outfile; sleep 5; done
      volumeMounts:
        - name: lustre
          mountPath: /mnt/lustre
  volumes:
    - name: lustre
      persistentVolumeClaim:
        claimName: azurelustre-dynamic-pvc

Apply the pod and wait for it to become ready:

kubectl apply -f azure-managed-lustre-dynamic-pod.yaml
kubectl wait --for=condition=Ready pod/azurelustre-dynamic-test --timeout=10m

Verify that the file system is mounted and the pod is writing to it:

kubectl exec azurelustre-dynamic-test -- df -h /mnt/lustre
kubectl exec azurelustre-dynamic-test -- tail /mnt/lustre/outfile

Clean up the example

Warning

With this example's Delete reclaim policy, the following commands delete the provisioned file system and its data. Save any data you need before deleting the claim.

Stop the test pod and let its volume unmount before deleting the claim:

kubectl delete pod azurelustre-dynamic-test
kubectl delete pvc azurelustre-dynamic-pvc
kubectl delete storageclass azurelustre-dynamic

Confirm that the provisioned file system is deleted in Azure. If you chose Retain, the PV and file system remain and require manual cleanup when no longer needed. Keep the driver installed until volume cleanup completes.

Next steps

  • Last updated on