nfs-provisioner

nfs-provisioner is an out-of-tree dynamic provisioner for Kubernetes 1.4. You can use it to quickly & easily deploy shared storage that works almost anywhere. Or it can help you write your own out-of-tree dynamic provisioner by serving as an example implementation of the requirements detailed in the proposal.

It works just like in-tree dynamic provisioners: a StorageClass object can specify an instance of nfs-provisioner to be its provisioner like it specifies in-tree provisioners such as GCE or AWS. Then, the instance of nfs-provisioner will watch for PersistentVolumeClaims that ask for the StorageClass and automatically create NFS-backed PersistentVolumes for them. For more information on how dynamic provisioning works, see the docs or this blog post.

Quickstart

Create a provisioner pod with the name matthew/nfs, by specifying the arg "-provisioner=matthew/nfs".

$ kubectl create -f deploy/kube-config/pod.yaml
pod "nfs-provisioner" created

Create a StorageClass named "matthew" with provisioner: matthew/nfs.

$ kubectl create -f deploy/kube-config/class.yaml
storageclass "matthew" created

Create a PersistentVolumeClaim with annotation volume.beta.kubernetes.io/storage-class: "matthew"

$ kubectl create -f deploy/kube-config/claim.yaml
persistentvolumeclaim "nfs" created

A PersistentVolume is provisioned for the PersistentVolumeClaim. Now the claim can be consumed by some pod(s) and the backing NFS storage read from or written to.

$ kubectl get pv
NAME                                       CAPACITY   ACCESSMODES   RECLAIMPOLICY   STATUS      CLAIM         REASON    AGE
pvc-dce84888-7a9d-11e6-b1ee-5254001e0c1b   1Mi        RWX           Delete          Bound       default/nfs             23s

Deleting the PersistentVolumeClaim will cause the provisioner to delete the PersistentVolume and its data.

Running

To deploy nfs-provisioner on a Kubernetes cluster see Deployment.

To use nfs-provisioner once it is deployed see Usage.

For information on running multiple instances of nfs-provisioner see Running Multiple Provisioners.

Implementation

The controller, the code for which is in the controller/ directory, watches PVCs and PVs to determine when to provision or delete volumes. It expects to receive an implementation of the Provisioner interface which has two methods: Provision and Delete. This NFS provisioner's implementation of the interface can be found under the volume/ directory.

So to create your own provisioner, you need to write your own implementation of the interface and pass it to the controller. Ideally you should be able to import the package to create the controller, without modifying any controller code. The passing in of the provisioner to the controller, and initialization of other things they might need (like a client for the Kubernetes API server), is done here in main.go.

Community

Kubernetes Storage SIG: https://github.com/kubernetes/community/tree/master/sig-storage

Roadmap

This is still alpha/experimental and will change to reflect the out-of-tree dynamic provisioner proposal

October

• Release
• Add CI & testing
• Support running as DaemonSet

November

• Add filesystem quotas for PV usage
• Support running as PetSet