LAB 11 – Persistent storage: Azure Disk & Azure Files¶

Story: CloudCart starts writing order logs and an uploads folder. Some data belongs to one pod (fast, block storage), some must be shared by all replicas. Choose the right storage class for each.


Duration	~35 min
Files	`statefulset-disk.yaml`, `pvc-files.yaml`, `deployment-shared.yaml`

	Azure Disk (`managed-csi*`)	Azure Files (`azurefile-csi*`)
Access mode	ReadWriteOnce – one node	ReadWriteMany – many pods/nodes
Use for	databases, queues, high IOPS	shared content, uploads, config
Zones	zonal (disk lives in one zone)	zone-redundant options

Step 1 – What's available?¶

source env.sh
kubectl get storageclass
kubectl describe storageclass managed-csi | grep -E "Provisioner|Parameters|ReclaimPolicy|VolumeBindingMode|AllowVolumeExpansion"
kubectl get csidrivers

Note WaitForFirstConsumer: the disk is created only when a pod is scheduled – in the pod's zone.

Step 2 – Azure Disk with a StatefulSet¶

envsubst < labs/LAB11-storage/statefulset-disk.yaml | kubectl apply -f -
kubectl -n cloudcart get pvc,pod -l app=orders -w        # PVCs Bound, pods Running, Ctrl+C
kubectl -n cloudcart exec orders-0 -- sh -c 'wget -qO- localhost:8080/visit; wget -qO- localhost:8080/visit'
kubectl -n cloudcart delete pod orders-0                # data must survive
kubectl -n cloudcart wait --for=condition=Ready pod/orders-0 --timeout=180s
kubectl -n cloudcart exec orders-0 -- wget -qO- localhost:8080/visit

✅ Checkpoint: the visit counter continues (3) after the pod was recreated – each replica (orders-0, orders-1) has its own disk.

az disk list -g $(az aks show -g $RG -n $CLUSTER --query nodeResourceGroup -o tsv) -o table

Step 3 – Expand a volume online¶

kubectl -n cloudcart patch pvc data-orders-0 -p '{"spec":{"resources":{"requests":{"storage":"10Gi"}}}}'
kubectl -n cloudcart get pvc data-orders-0 -w
kubectl -n cloudcart exec orders-0 -- df -h /data

Step 4 – Azure Files shared by several pods¶

kubectl apply -f labs/LAB11-storage/pvc-files.yaml
envsubst < labs/LAB11-storage/deployment-shared.yaml | kubectl apply -f -
kubectl -n cloudcart rollout status deploy/uploads
for p in $(kubectl -n cloudcart get pods -l app=uploads -o name); do kubectl -n cloudcart exec $p -- wget -qO- localhost:8080/visit; done

✅ Checkpoint: three pods on different nodes write to one file – the counter keeps growing across pods (ReadWriteMany).

Step 5 – Reclaim policy¶

kubectl get pv -o custom-columns=NAME:.metadata.name,CLAIM:.spec.claimRef.name,POLICY:.spec.persistentVolumeReclaimPolicy

Delete (default) removes the Azure disk/share with the PVC. For data that must outlive the cluster, create a StorageClass with reclaimPolicy: Retain – or use static provisioning of an existing disk.

Step 6 – Clean up¶

kubectl -n cloudcart delete deploy uploads
kubectl -n cloudcart delete statefulset orders
kubectl -n cloudcart delete pvc --all

Discussion¶

Backups: Azure Backup for AKS (cluster state + disk snapshots), Velero.
Azure Container Storage for very high-performance local NVMe / ephemeral disks.
Why are databases often better as managed Azure services than StatefulSets?

Troubleshooting¶

Symptom	Likely cause
PVC `Pending` with waiting for first consumer	normal – no pod uses it yet
`Multi-Attach error for volume`	an RWO disk is still attached to another node (old pod terminating)
pod `Pending`: volume node affinity conflict	zonal disk in a zone without a node of the pool
`Permission denied` writing to Azure Files	non-root container – check `fsGroup` / mount options

Files for this lab: Think-Cube/aks-labs/labs/LAB11-storage