Skip to main content
Nebius AI Cloud offers persistent storage for pods in Managed Service for Kubernetes clusters. In this tutorial, you will mount a Compute disk as a persistent volume to a pod in your cluster.

​
Background

By default, files created or modified by a container in a Kubernetes pod are lost when the container crashes, or is stopped or restarted. Persistent volumes help you avoid this: they keep the files for the lifetime of the pod, regardless of the states of its individual containers to the pod. These volumes are mounted to a node that is running the pod. In Nebius AI Cloud, Managed Service for Kubernetes offers a native solution for persistent volumes in your clusters. You can mount disks from another Nebius AI Cloud service, Compute, to your pods as persistent volumes. Compute provides multiple disk types that vary in performance, reliability and price, and you can choose between them depending on your use case. For more details, see the Compute documentation. Mounting disks to pods relies on support for Container Storage Interface (CSI), which is enabled on newer Managed Kubernetes clusters and can be enabled for older clusters (see prerequisites below). A disk can be used on one node at a time. If you need to share data between pods and nodes, use Compute shared filesystems instead of disks. See Mounting shared filesystems to pods in a Managed Service for Kubernetes® cluster for details.

​
Costs

Nebius AI Cloud only charges for a Compute disk. For more details, see the Compute pricing.

​
Prerequisites

  1. Create a Managed Kubernetes cluster or choose an existing one.
  2. If you are working with an existing cluster, verify that it supports CSI.
  3. Install kubectl and connect to the cluster.
Nebius AI Cloud disks are created automatically; you do not need to create them beforehand.

​
Verify that your cluster supports CSI

If your Managed Kubernetes cluster was created on or after January 7, 2025, it already supports CSI and no additional configuration is required. Proceed to the tutorial’s steps. If your cluster was created before January 7, 2025, enable CSI support in it:
  1. Get IDs of the node groups in the cluster:
    1. In Managed Kubernetes, click on your cluster.
    2. Switch to the Node groups tab.
    3. For each node group, find its ID under its name and click https://mintcdn.com/nebius-ai-cloud/1Ha0sWR6e1mnIaHS/_assets/button-copy.svg?fit=max&auto=format&n=1Ha0sWR6e1mnIaHS&q=85&s=78fcf2cb844ad8eda4598bfbb10b3680 to copy it.
  2. Upgrade each node group: 
    nebius mk8s node-group upgrade --id <node_group_ID> --latest-infra-version
  3. Contact support or your solution architect to enable CSI support.

​
Steps

​
Create a storage class for custom provisioning (optional)

Before using Compute disks as persistent volumes, you can choose how disks are provisioned: the disk type, the filesystem type (ext4 or xfs) and when a disk should be created (together with a persistent volume claim or when the claim is used by a pod). To do this, define a custom Kubernetes storage class.
The default storage class offered by Managed Kubernetes, compute-csi-default-sc, creates a Network SSD disk with an ext4 filesystem, and does so when a pod that uses a persistent volume claim with this storage class is created. If this configuration suits you, you can skip this step and proceed to creating a persistent volume claim.
To create a custom storage class:
  1. Create a manifest, e.g., storage-class.yaml, that defines the storage class. For example: 
    apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: storage-test-class
provisioner: compute.csi.nebius.com
volumeBindingMode: WaitForFirstConsumer
parameters:
  csi.storage.k8s.io/fstype: xfs
  type: "NETWORK_SSD"
    • .metadata.name — The name of the storage class. Use it when referring to the class in persistent volume claims (.spec.storageClassName).
    • .provisioner — The volume provisioner. For Nebius AI Cloud disks, the provisioner is compute.csi.nebius.com.
    • .volumeBindingMode — The volume binding mode that determines when a Nebius AI Cloud disk is created for a persistent volume claim of this storage class. Supported values are WaitForFirstCustomer (the disk is created together with a pod that uses the claim) or Immediate (the disk is created together with the claim, even if no pods use it yet).
      Disks created in the Immediate binding mode may start to be charged and to count towards quotas before being actually used in pods. To avoid this, use the WaitForFirstCustomer binding mode.
    • .parameters."csi.storage.k8s.io/fstype" — The filesystem for disks: ext4 or xfs.
    • .parameters.type — The Nebius AI Cloud disk type. See the list of disk types and their IDs. Write IDs in uppercase, for example, NETWORK_SSD, NETWORK_SSD_IO_M3.
  2. Apply the manifest to your cluster: 
    kubectl apply -f storage-class.yaml

​
Create a persistent volume claim

Persistent volume claims are requests that pods make for persistent volumes. These requests include what kind of storage should be provided (in our case, a Compute disk with the default or a custom storage class) and how large it should be. To create a persistent volume claim:
  1. Create a manifest, e.g., pvc.yaml, that defines the claim. For example: 
    apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: storage-test-claim
spec:
  accessModes:
    - ReadWriteOnce
  storageClassName: compute-csi-default-sc
  resources:
    requests:
      storage: 4Gi
    • .metadata.name — The name of your claim. Use it when referring to the claim in pod configurations.
    • .spec.accessModes — An array of access modes that determine how the persistent volume can be mounted to nodes. The only mode supported for Nebius AI Cloud block storage is ReadWriteOnce; the volume can be mounted as read-write by a single node. For storage that can be mounted to multiple nodes (ReadWriteMany), see Mounting shared filesystems to pods in a Managed Service for Kubernetes® cluster.
    • .spec.storageClassName — The storage class of the volume. The default storage class for Nebius AI Cloud block storage, compute-csi-default-sc, creates a Network SSD disk with an ext4 filesystem. To use another disk type or filesystem, create a custom storage class by using the Nebius AI Cloud provisioner and specify its name here.
    • .spec.resources.requests.storage — The volume size.
  2. Apply the manifest to your cluster: 
    kubectl apply -f pvc.yaml

​
Mount a volume to a pod and test it

When a persistent volume claim is used in a pod’s specification, the Compute disk created for the claim is mounted to the pod as a persistent volume.
A disk created for a claim is mounted to each pod that uses the claim, and is deleted together with the claim. Do not use a claim on multiple pods at one time, as a disk can only be mounted to a single node. For a solution that shares storage between nodes, see Mounting shared filesystems to pods in a Managed Service for Kubernetes® cluster.
To create a pod with a volume and test it:
  1. Create a manifest, e.g. pod.yaml, that defines the pod. For example, in the manifest below, a volume named persistent-storage, created from the test-claim persistent volume claim, is mounted to the pod’s container at /data: 
    apiVersion: v1
kind: Pod
metadata:
  name: storage-test-app
spec:
  volumes:
    - name: persistent-storage
      persistentVolumeClaim:
        claimName: storage-test-claim
  containers:
    - name: app
      image: centos
      command: ["/bin/sh"]
      args: ["-c", "while true; do echo $(date -u) >> /data/out.txt; sleep 5; done"]
      volumeMounts:
        - name: persistent-storage
          mountPath: /data
    • .spec.volumes[0].name: The name of the volume that is created for the pod.
    • .spec.volumes[0].persistentVolumeClaim.claimName: The name of the persistent volume claim that is used to create a volume.
    • .spec.containers[0].volumeMounts[0].name: The same volume name as in .spec.volumes[0].name.
    • .spec.containers[0].volumeMounts[0].mountPath: The mount point for the volume.
    You can configure volumes in Kubernetes resources that manage pods, like deployments or stateful sets, in the same way, by using sub-fields of .spec.template.spec.volumes and .spec.template.spec.containers.
  2. Apply the manifest to your cluster: 
    kubectl apply -f pod.yaml
  3. Wait until the pod is ready: 
    kubectl wait --for condition=Ready=true pod/storage-test-app
  4. Test that the container has written into /data/out.txt: 
    kubectl exec storage-test-app -- cat /data/out.txt

​
How to delete the created resources

The created Compute disk is chargeable. If you do not need it, delete the pod and the persistent volume claim, so that the disk is deleted and Nebius AI Cloud does not charge for it: 
kubectl delete pod/storage-test-app pvc/storage-test-claim

​
See also