This work is based on Ranchers Local Path Provisioner. The original code was changed to work with ZFS datasets instead of local directories.

Local ZFS Provisioner provides a way for Kubernetes to utilize the local storage on each node. Based on the user configuration, the Local ZFS Provisioner will create hostPath based persistent volumes on the nodes automatically. It utilizes the features introduced by Kubernetes Local Persistent Volume feature, but it is much simpler to use then the built-in local volume feature in Kubernetes.

The ZFS Provisioner is implemented as a hostPath provisioner that shedules pods targeted at specific nodes to provision or delete datasets to fullfill the requested Persistent Volume Claims. It is typically deployed as a Kubernetes Deployment.

Kubernetes v1.12+.

In this setup the directory /var/lib/local-zfs-provisioner will be used across all nodes as the base mount point for provisoned datasets. The provisioner will be installed in the kube-system namespace by default.

kubectl apply -f https://raw.githubusercontent.com/pavel-tarasenko/local-zfs-provisioner/master/deploy/rbac.yaml
kubectl apply -f https://raw.githubusercontent.com/pavel-tarasenko/local-zfs-provisioner/master/deploy/deployment.yaml

Create a suitable configmap and add it to the cluster. You will have to change this to work with your zfs pools and datasets.

kubectl apply -f https://raw.githubusercontent.com/pavel-tarasenko/local-zfs-provisioner/master/deploy/example-config.yaml

Create a hostPath backed Persistent Volume Claim and a pod that uses it:

kubectl create -f https://raw.githubusercontent.com/pavel-tarasenko/local-zfs-provisioner/master/example/pvc.yaml
kubectl create -f https://raw.githubusercontent.com/pavel-tarasenko/local-zfs-provisioner/master/example/pod.yaml

You should see that the PV has been created:

$ kubectl get pv
NAME                                       CAPACITY   ACCESS MODES   RECLAIM POLICY   STATUS   CLAIM                    STORAGECLASS   REASON   AGE
pvc-5fdc9d7f-2a27-11e9-8180-a4bf0112bd54   2Gi        RWO            Delete           Bound    default/local-zfs-pvc    local-zfs               10s

The PVC has been bound:

$ kubectl get pvc
NAME             STATUS   VOLUME                                     CAPACITY   ACCESS MODES   STORAGECLASS   AGE
local-zfs-pvc    Bound    pvc-5fdc9d7f-2a27-11e9-8180-a4bf0112bd54   2Gi        RWO            local-zfs      16s

And the Pod started running:

$ kubectl get pod
NAME              READY     STATUS    RESTARTS   AGE
volume-test-zfs   1/1       Running   0          3s

Write something into the pods volume:

kubectl exec volume-test-zfs -- sh -c "echo local-zfs-test > /data/test"

Now delete the pod again:

kubectl delete -f https://raw.githubusercontent.com/pavel-tarasenko/local-zfs-provisioner/master/example/pod.yaml

After confirming that the pod is gone, recreated it:

kubectl create -f https://raw.githubusercontent.com/pavel-tarasenko/local-zfs-provisioner/master/example/pod.yaml

Check the volume content:

$ kubectl exec volume-test-zfs cat /data/test
local-zfs-test

Delete the pod and the pvc:

kubectl delete -f https://raw.githubusercontent.com/pavel-tarasenko/local-zfs-provisioner/master/example/pod.yaml
kubectl delete -f https://raw.githubusercontent.com/pavel-tarasenko/local-zfs-provisioner/master/example/pvc.yaml

The volume content stored on the node will be automatically cleaned up. You can check the log of the local-zfs-provisioner-xxx pod for details.

You have now verified that the provisioner works as expected.

The configuration of the provisioner is a json file config.json, stored in a config map, e.g.:

kind: ConfigMap
apiVersion: v1
metadata:
  name: local-zfs-provisioner-config
  namespace: kube-system
data:
  config.json: |-
    {
        "nodeDatasetMap": [
            {
                "node": "DEFAULT_PATH_FOR_NON_LISTED_NODES",
                "dataset": "pool/data/local-zfs-provisioner"
            },
            {
                "node": "that-other-node",
                "dataset": "tank/local-zfs-provisioner"
            }
        ]
    }

nodeDatasetMap is the place where the user can customize where to store the data on each node.

1. If a node is not listed in the nodeDatasetMap map, and Kubernetes wants to create volume on it, the dataset specified in DEFAULT_PATH_FOR_NON_LISTED_NODES will be used for provisioning.
2. If a node is listed in the nodeDatasetMap map, the specified dataset will be used for provisioning.

The configuration must obey following rules:

1. config.json must be a valid json file.
2. A dataset name can not start with /.
3. No duplicate node allowed.

The provisioner supports automatic reloading of configuration. Users can change the configuration using kubectl apply or kubectl edit with config map local-zfs-provisioner-config.

When the provisioner detects configuration changes, it will try to load the new configuration.

If the reload fails due to some reason, the provisioner will report error in the log, and continue using the last valid configuration for provisioning in the meantime.

Before uninstallation, make sure that the PVs created by the provisioner have already been deleted. Use kubectl get pv and make sure no PV with StorageClass local-zfs.

To uninstall, execute:

kubectl delete -f https://raw.githubusercontent.com/pavel-tarasenko/local-zfs-provisioner/master/deploy/deployment.yaml
kubectl delete -f https://raw.githubusercontent.com/pavel-tarasenko/local-zfs-provisioner/master/deploy/rbac.yaml

Copyright (c) 2014-2018 Rancher Labs, Inc.

Copyright (c) 2019 Steven Armstrong

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.