Getting started with NFS Subdir External Provisioner and OVHcloud File Storage Service

Als Markdown ansehen

Find out how to use the NFS Subdir External Provisioner to provision Kubernetes persistent volumes from an OVHcloud File Storage Service NFS share.

Objective

This guide explains how to deploy the NFS Subdir External Provisioner on a Kubernetes cluster to dynamically provision persistent volumes from an OVHcloud File Storage Service NFS share.

This approach is compatible with OVHcloud Managed Kubernetes Service (MKS) and self-managed Kubernetes environments running on OVHcloud Public Cloud instances.

The NFS Subdir External Provisioner does not manage the lifecycle of the underlying NFS share. Instead, it automatically creates a dedicated subdirectory inside an existing share for each PersistentVolumeClaim (PVC), making storage available to pods with ReadWriteMany (RWX) access. This approach is simpler to operate than Manila CSI but trades off features such as snapshots and per-PVC capacity enforcement.

Architecture

When a PVC is created against the provisioner's StorageClass, the provisioner pod mounts the NFS share root, creates a subdirectory named after the PVC, and binds a PersistentVolume (PV) to it. Pods then mount only their subdirectory, not the entire share.

OVHcloud File Storage Service share
  └── /shares/share-<UUID>/                 ← NFS root (nfs.path)
        ├── ns1-pvc-a-pvc-<UUID>/           ← PVC "pvc-a" in namespace "ns1"
        ├── ns1-pvc-b-pvc-<UUID>/           ← PVC "pvc-b" in namespace "ns1"
        └── archived-ns1-pvc-c-pvc-<UUID>/  ← deleted PVC (archiveOnDelete=true)

Requirements

A Kubernetes cluster whose nodes are deployed on a private network with access to the File Storage Service share — compatible with OVHcloud Managed Kubernetes Service (MKS) and self-managed Kubernetes clusters running on OVHcloud Public Cloud instances
An OVHcloud File Storage Service NFS share in the same region and on the same private network as your cluster
Helm installed on your workstation — see the How to install Helm on OVHcloud Managed Kubernetes Service guide
Basic knowledge of operating a Kubernetes cluster — see the Deploying a Hello World application guide
Optional — OVHcloud CLI for command-line management of the share and ACL

Instructions

If you have not yet created a share, follow the File Storage Service getting started guide to create an NFS share on your private network.

Once the share is in available status, retrieve its NFS export path. The path follows this format:

<NFS_SERVER>:/shares/share-<UUID>

For example: 10.1.0.12:/shares/share-abc12345-def6-4abc-8def-123456abcdef

Split it into 2 components and note them for the following steps:

NFS_SERVER — the IP address part (e.g. 10.1.0.12)
NFS_PATH — the path part (e.g. /shares/share-abc12345-def6-4abc-8def-123456abcdef)

Via the OVHcloud Control Panel

Via the OVHcloud CLI

Go to Storage & backup > File Storage, click your share, and find the Mount Path on the General information tab.

# List shares to find the share ID
ovhcloud cloud storage file share list \
  --cloud-project <PROJECT_ID> \
  --region <REGION>

# Retrieve the share details including the NFS export path
ovhcloud cloud storage file share get <SHARE_ID> \
  --cloud-project <PROJECT_ID> \
  --region <REGION>

Note the NFS export path from the output and split it into NFS_SERVER and NFS_PATH as described above.

The provisioner pod and every node that will mount NFS volumes must be allowed to reach the share. Add an ACL entry covering your cluster's private network subnet.

Info

The File Storage Service is only accessible from OVHcloud IPs (Public Cloud instances, Managed Kubernetes Service). A single CIDR rule for your cluster subnet covers all nodes without listing individual IPs.

Via the OVHcloud Control Panel

Via the OVHcloud CLI

Click your share, go to the Access Control List (ACL) tab, then click Add a new access. Enter the CIDR range of your cluster's private network (e.g. 10.1.0.0/24) and select the Read and write permission.

Verify that the ACL entry is in active status before continuing.

ovhcloud cloud storage file share acl create <SHARE_ID> \
  --cloud-project <PROJECT_ID> \
  --region <REGION> \
  --access-level rw \
  --access-to <CIDR>

Verify the ACL status:

ovhcloud cloud storage file share acl list <SHARE_ID> \
  --cloud-project <PROJECT_ID> \
  --region <REGION>

Wait until the entry is in active status before continuing.

Step 3 - Install the NFS Subdir External Provisioner

Add the Helm repository:

helm repo add nfs-subdir-external-provisioner https://kubernetes-sigs.github.io/nfs-subdir-external-provisioner
helm repo update

Install the provisioner, replacing <NFS_SERVER> and <NFS_PATH> with the values retrieved in Step 1:

helm install nfs-subdir nfs-subdir-external-provisioner/nfs-subdir-external-provisioner \
  --namespace nfs-provisioner \
  --create-namespace \
  --set nfs.server=<NFS_SERVER> \
  --set nfs.path=<NFS_PATH> \
  --set storageClass.name=nfs-subdir \
  --set storageClass.reclaimPolicy=Delete \
  --set storageClass.archiveOnDelete=false

Info

archiveOnDelete=false permanently removes subdirectory data when the PVC is deleted. Set it to true to rename deleted PVC directories to archived-* instead, preserving the data at the cost of continued share consumption.

Verify the provisioner pod is running:

kubectl -n nfs-provisioner get pods

$ kubectl -n nfs-provisioner get pods
NAME                                                          READY   STATUS    RESTARTS   AGE
nfs-subdir-nfs-subdir-external-provisioner-6d4bfc8b4-xkjlp   1/1     Running   0          30s

Step 4 - Inspect the StorageClass

The Helm chart creates a StorageClass automatically. Inspect it to confirm the configuration:

kubectl get storageclass nfs-subdir -o yaml

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: nfs-subdir
provisioner: cluster.local/nfs-subdir-nfs-subdir-external-provisioner
parameters:
  archiveOnDelete: "false"
reclaimPolicy: Delete
volumeBindingMode: Immediate

volumeBindingMode: Immediate means a PV is provisioned as soon as the PVC is created, regardless of whether any pod has claimed it yet.

Step 5 - Create PersistentVolumeClaims

Create a PVC using the nfs-subdir StorageClass:

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: nfs-shared-pvc
spec:
  accessModes:
    - ReadWriteMany
  storageClassName: nfs-subdir
  resources:
    requests:
      storage: 5Gi

Apply it:

kubectl apply -f nfs-pvc.yaml

Warning

The storage request is not enforced by the NFS Subdir External Provisioner — it is recorded as metadata only. All PVCs share the full capacity of the underlying File Storage Service share. Size your share according to your total expected usage across all PVCs.

Verify the PVC is bound:

kubectl get pvc nfs-shared-pvc

$ kubectl get pvc nfs-shared-pvc
NAME             STATUS   VOLUME                                     CAPACITY   ACCESS MODES   STORAGECLASS   AGE
nfs-shared-pvc   Bound    pvc-1a2b3c4d-5e6f-7890-abcd-ef1234567890   5Gi        RWX            nfs-subdir     5s

Step 6 - Test with a shared workload

Deploy 2 Nginx replicas that mount the same PVC to verify ReadWriteMany access:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: nfs-test
spec:
  replicas: 2
  selector:
    matchLabels:
      app: nfs-test
  template:
    metadata:
      labels:
        app: nfs-test
    spec:
      volumes:
        - name: shared-data
          persistentVolumeClaim:
            claimName: nfs-shared-pvc
      containers:
        - name: nginx
          image: nginx:stable
          volumeMounts:
            - name: shared-data
              mountPath: /usr/share/nginx/html

kubectl apply -f nfs-deployment.yaml
kubectl get pods -l app=nfs-test

$ kubectl get pods -l app=nfs-test
NAME                        READY   STATUS    RESTARTS   AGE
nfs-test-74c5d6d8c6-fvbnm   1/1     Running   0          20s
nfs-test-74c5d6d8c6-qwxyz   1/1     Running   0          20s

Write a file from the first pod and confirm it is visible from the second:

POD1=$(kubectl get pods -l app=nfs-test -o jsonpath='{.items[0].metadata.name}')
POD2=$(kubectl get pods -l app=nfs-test -o jsonpath='{.items[1].metadata.name}')

kubectl exec "$POD1" -- sh -c 'echo "shared via NFS" > /usr/share/nginx/html/index.html'
kubectl exec "$POD2" -- cat /usr/share/nginx/html/index.html

shared via NFS

Both pods read and write the same subdirectory on the NFS share, confirming RWX access across nodes.

Limitations

No capacity enforcement: The storage field in a PVC is metadata only. All PVCs from a given StorageClass share the total capacity of the underlying NFS share.
No volume snapshots: The provisioner does not support the VolumeSnapshot API. Use Manila CSI if you need snapshot support.
No volume resize: PVCs cannot be resized after creation.
No ReadWriteOnce quota isolation: For single-pod ReadWriteOnce use cases with capacity enforcement, OVHcloud Block Storage (csi-cinder-high-speed) is more suitable.
Archived directories accumulate: With archiveOnDelete=true, deleted PVC data is renamed but not removed. Capacity is consumed until you manually clean the archived directories.

Best practices

One Helm release per environment: Deploy separate provisioner instances (e.g. nfs-subdir-dev, nfs-subdir-prod) pointing to different shares or paths to avoid cross-environment data mixing.
Use pathPattern for readable directory names: Set storageClass.pathPattern to ${.PVC.namespace}/${.PVC.name} to generate human-readable subdirectory structures. See the NFS Subdir External Provisioner documentation for the full list of template variables.
Choose archiveOnDelete deliberately: Decide up front whether deleted PVC data should be preserved (true) or removed (false). Mixing policies after data exists on the share can cause confusion.
Monitor share capacity: Because per-PVC quotas are not enforced, monitor your File Storage Service share size in the OVHcloud Control Panel or via the OVHcloud API and resize it before it fills up.
Match NFS mount options to File Storage Service capabilities: OVHcloud File Storage Service supports NFS v4. If you need to override mount options (e.g. nfsvers=4,hard,timeo=600), set them via storageClass.mountOptions in the Helm values.

Go further

For training or technical assistance implementing our solutions, contact your sales representative or visit our Professional Services page to request a quote and have your project analyzed by our experts.

Join our community of users.

War diese Seite hilfreich?