Skip to content

Deploy NFS for Preloading Dataset

A Network File System (NFS) allows remote hosts to mount file systems over a network and interact with those file systems as though they are mounted locally. This enables system administrators to consolidate resources onto centralized servers on the network.

Dataset is a core feature provided by DCE 5.0 AI Lab. By abstracting the dependency on data throughout the entire lifecycle of MLOps into datasets, users can manage various types of data in datasets so that training tasks can directly use the data in the dataset.

When remote data is not within the worker cluster, datasets provide the capability to automatically preheat data, supporting data preloading from sources such as Git, S3, and HTTP to the local cluster.

A storage service supporting the ReadWriteMany mode is needed for preloading remote data for the dataset, and it is recommended to deploy NFS within the cluster.

This article mainly introduces how to quickly deploy an NFS service and add it as a StorageClass for the cluster.

Preparation

  • NFS by default uses the node's storage as a data caching point, so it is necessary to ensure that the disk itself has enough disk space.
  • The installation method uses Helm and Kubectl, please make sure they are already installed.

Deployment Steps

Several components need to be installed:

  • NFS Server
  • csi-driver-nfs
  • StorageClass

Initialize Namespace

All system components will be installed in the nfs namespace, so it is necessary to create this namespace first.

kubectl create namespace nfs

Install NFS Server

Here is a simple YAML deployment file that can be used directly.

Note

Be sure to check the image: and modify it to a domestic mirror based on the location of the cluster.

nfs-server.yaml
---
kind: Service
apiVersion: v1
metadata:
  name: nfs-server
  namespace: nfs
  labels:
    app: nfs-server
spec:
  type: ClusterIP
  selector:
    app: nfs-server
  ports:
    - name: tcp-2049
      port: 2049
      protocol: TCP
    - name: udp-111
      port: 111
      protocol: UDP
---
kind: Deployment
apiVersion: apps/v1
metadata:
  name: nfs-server
  namespace: nfs
spec:
  replicas: 1
  selector:
    matchLabels:
      app: nfs-server
  template:
    metadata:
      name: nfs-server
      labels:
        app: nfs-server
    spec:
      nodeSelector:
        "kubernetes.io/os": linux
      containers:
        - name: nfs-server
          image: itsthenetwork/nfs-server-alpine:latest
          env:
            - name: SHARED_DIRECTORY
              value: "/exports"
          volumeMounts:
            - mountPath: /exports
              name: nfs-vol
          securityContext:
            privileged: true
          ports:
            - name: tcp-2049
              containerPort: 2049
              protocol: TCP
            - name: udp-111
              containerPort: 111
              protocol: UDP
      volumes:
        - name: nfs-vol
          hostPath:
            path: /nfsdata  # (1)!
            type: DirectoryOrCreate
  1. Modify this to specify another path to store NFS shared data

Save the above YAML as nfs-server.yaml, then run the following commands for deployment:

kubectl -n nfs apply -f nfs-server.yaml

# Check the deployment result
kubectl -n nfs get pod,svc

Install csi-driver-nfs

Installing csi-driver-nfs requires the use of Helm, please ensure it is installed beforehand.

# Add Helm repository
helm repo add csi-driver-nfs https://mirror.ghproxy.com/https://raw.githubusercontent.com/kubernetes-csi/csi-driver-nfs/master/charts
helm repo update csi-driver-nfs

# Deploy csi-driver-nfs
# The parameters here mainly optimize the image address to accelerate downloads in China
helm upgrade --install csi-driver-nfs csi-driver-nfs/csi-driver-nfs \
    --set image.nfs.repository=k8s.m.daocloud.io/sig-storage/nfsplugin \
    --set image.csiProvisioner.repository=k8s.m.daocloud.io/sig-storage/csi-provisioner \
    --set image.livenessProbe.repository=k8s.m.daocloud.io/sig-storage/livenessprobe \
    --set image.nodeDriverRegistrar.repository=k8s.m.daocloud.io/sig-storage/csi-node-driver-registrar \
    --namespace nfs \
    --version v4.5.0

Warning

Not all images of csi-nfs-controller support helm parameters, so the image field of the deployment needs to be manually modified. Change image: registry.k8s.io to image: k8s.dockerproxy.com to accelerate downloads in China.

Create StorageClass

Save the following YAML as nfs-sc.yaml:

nfs-sc.yaml
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: nfs-csi
provisioner: nfs.csi.k8s.io
parameters:
  server: nfs-server.nfs.svc.cluster.local
  share: /
  # csi.storage.k8s.io/provisioner-secret is only needed for providing mountOptions in DeleteVolume
  # csi.storage.k8s.io/provisioner-secret-name: "mount-options"
  # csi.storage.k8s.io/provisioner-secret-namespace: "default"
reclaimPolicy: Delete
volumeBindingMode: Immediate
mountOptions:
  - nfsvers=4.1

then run the following command:

kubectl apply -f nfs-sc.yaml

Test

Create a dataset and set the dataset's associated storage class and preloading method to NFS to preheat remote data into the cluster.

After the dataset is successfully created, you can see that the dataset's status is preloading, and you can start using it after the preloading is completed.

FAQs

Missing Necessary NFS Client Software /sbin/mount

bad option; for several filesystems (e.g. nfs, cifs) you might need a /sbin/mount.<type> helper program.

On the nodes running Kubernetes, ensure that the NFS client is installed:

Run the following commands to install the NFS client:

sudo apt-get update
sudo apt-get install nfs-common

Run the following command to install the NFS client:

sudo yum install nfs-utils

Check the NFS server configuration to ensure that the NFS server is running and configured correctly. You can try mounting manually to test:

sudo mkdir -p /mnt/test
sudo mount -t nfs <nfs-server>:/nfsdata /mnt/test

Comments