Harvester CSI Driver

The Harvester Container Storage Interface (CSI) Driver provides a standard CSI interface used by guest Kubernetes clusters in Harvester. It connects to the host cluster and hot-plugs host volumes to the virtual machines (VMs) to provide native storage performance.

Deploying

Prerequisites

The Kubernetes cluster is built on top of Harvester virtual machines.
The Harvester virtual machines that run as guest Kubernetes nodes are in the same namespace.

Currently, the Harvester CSI driver only supports single-node read-write(RWO) volumes. Please follow the issue #1992 for future multi-node read-only(ROX) and read-write(RWX) support.

Deploying with Harvester RKE1 node driver

Select the Harvester(Out-of-tree) option.
Install Harvester CSI Driver from the Rancher marketplace.

Deploying with Harvester RKE2 node driver

When spinning up a Kubernetes cluster using Rancher RKE2 node driver, the Harvester CSI driver will be deployed automatically when Harvester cloud provider is selected.

Install CSI driver manually in the RKE2 cluster

If you prefer to install the Harvester CSI driver without enabling the Harvester cloud provider, you can refer to the following steps:

Prerequisites of manual install

Ensure that you have the following prerequisites in place:

You have kubectl and jq installed on your system.
You have the kubeconfig file for your bare-metal Harvester cluster. You can find the kubeconfig file from one of the Harvester management nodes in the /etc/rancher/rke2/rke2.yaml path.
```
  export KUBECONFIG=/path/to/your/harvester-kubeconfig
```

Perform the following steps to deploy the Harvester CSI driver manually:

Deploy Harvester CSI driver

Generate the cloud-config. You can generate the cloud-config file using the generate_addon_csi.sh script. It is available on the harvester/harvester-csi-driver repo.

<serviceaccount name> usually corresponds to your guest cluster name, and <namespace> should match the machine pool’s namespace.

./generate_addon_csi.sh <serviceaccount name> <namespace> RKE2

The generated output will be similar to the following one:

 ########## cloud-config ############
 apiVersion: v1
 clusters:
 - cluster: <token>
     server: https://<YOUR HOST HARVESTER VIP>:6443
   name: default
 contexts:
 - context:
     cluster: default
     namespace: default
     user: rke2-guest-01-default-default
   name: rke2-guest-01-default-default
 current-context: rke2-guest-01-default-default
 kind: Config
 preferences: {}
 users:
 - name: rke2-guest-01-default-default
   user:
     token: <token>

 ########## cloud-init user data ############
 write_files:
   - encoding: b64
     content: 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
     owner: root:root
     path: /var/lib/rancher/rke2/etc/config-files/cloud-provider-config
     permissions: '0644'

Copy and paste the cloud-init user data content to Machine Pools > Show Advanced > User Data.

The cloud-provider-config file will be created after you apply the cloud-init user data above. You can find it on the guest Kubernetes nodes at the path /var/lib/rancher/rke2/etc/config-files/cloud-provider-config.
Configure the Cloud Provider either to Default - RKE2 Embedded or External.
Select Create to create your RKE2 cluster.
Once the RKE2 cluster is ready, install the Harvester CSI Driver chart from the Rancher marketplace. You do not need to change the cloud-config path by default.

If you prefer not to install the Harvester CSI driver using Rancher (Apps > Charts), you can use Helm instead. The Harvester CSI driver is packaged as a Helm chart. For more information, see https://charts.harvesterhci.io.

By following the above steps, you should be able to see those CSI driver pods are up and running on the kube-system namespace, and you can verify it by provisioning a new PVC using the default StorageClass harvester on your RKE2 cluster.

Deploying with Harvester K3s node driver

You can follow the Deploy Harvester CSI driver steps described in the RKE2 section.

The only difference is in generating the cloud-init config where you need to specify the provider type as k3s:

./generate_addon_csi.sh <serviceaccount name> <namespace> k3s

Customize the Default StorageClass

The Harvester CSI driver provides the interface for defining the default StorageClass. If the default StorageClass in unspecified, the Harvester CSI driver uses the default StorageClass of the host Harvester cluster.

You can use the parameter host-storage-class to customize the default StorageClass.

Create a StorageClass for the host Harvester cluster.

Example:
Deploy the CSI driver with the parameter host-storage-class.

Example:
Verify that the Harvester CSI driver is ready.
1. On the PersistentVolumeClaims screen, create a PVC. Select Use a Storage Class to provision a new Persistent Volume and specify the StorageClass you created.
  
  Example:
2. Once the PVC is created, note the name of the provisioned volume and verify that the status is Bound.
  
  Example:
3. On the Volumes screen, verify that the volume was provisioned using the StorageClass that you created.
  
  Example:

Passthrough Custom StorageClass

Beginning with Harvester CSI driver v0.1.15, it’s possible to create a PersistentVolumeClaim (PVC) using a different Harvester StorageClass on the guest Kubernetes Cluster.

Harvester CSI driver v0.1.15 is supported out of the box starting with the following RKE2 versions. For RKE1, manual installation of the CSI driver chart is required:

v1.23.16+rke2r1 and later
v1.24.10+rke2r1 and later
v1.25.6+rke2r1 and later
v1.26.1+rke2r1 and later
v1.27.1+rke2r1 and later

Prerequisites

Add the following prerequisites to your Harvester cluster to ensure the Harvester CSI driver displays error messages correctly. Proper RBAC settings are essential for error message visibility, especially when creating a PVC with a non-existent StorageClass, as shown in the image below:

Follow these steps to set up RBAC for error message visibility:

Create a new clusterrole named harvesterhci.io:csi-driver using the following manifest.

 apiVersion: rbac.authorization.k8s.io/v1
 kind: ClusterRole
 metadata:
   labels:
     app.kubernetes.io/component: apiserver
     app.kubernetes.io/name: harvester
     app.kubernetes.io/part-of: harvester
   name: harvesterhci.io:csi-driver
 rules:
 - apiGroups:
   - storage.k8s.io
   resources:
   - storageclasses
   verbs:
   - get
   - list
   - watch

Create a new clusterrolebinding associated with the clusterrole above with the relevant serviceaccount using the following manifest.

 apiVersion: rbac.authorization.k8s.io/v1
 kind: ClusterRoleBinding
 metadata:
   name: <namespace>-<serviceaccount name>
 roleRef:
   apiGroup: rbac.authorization.k8s.io
   kind: ClusterRole
   name: harvesterhci.io:csi-driver
 subjects:
 - kind: ServiceAccount
   name: <serviceaccount name>
   namespace: <namespace>

Make sure the serviceaccount name and namespace match your cloud provider settings. Perform the following steps to retrieve these details.

Find the rolebinding associated with your cloud provider:

 $ kubectl get rolebinding -A |grep harvesterhci.io:cloudprovider
 default                                 default-rke2-guest-01                                ClusterRole/harvesterhci.io:cloudprovider             7d1h

Extract the subjects information from this rolebinding:

 $ kubectl get rolebinding default-rke2-guest-01 -n default -o yaml |yq -e '.subjects'

Identify the ServiceAccount information:

 - kind: ServiceAccount
   name: rke2-guest-01
   namespace: default

Deploying

Now you can create a new StorageClass that you intend to use in your guest Kubernetes cluster.

For administrators, you can create a desired StorageClass (e.g., named replica-2) in your bare-metal Harvester cluster.

Then, on the guest Kubernetes cluster, create a new StorageClass associated with the StorageClass named replica-2 from the Harvester Cluster:

When choosing a Provisioner, select Harvester (CSI). The Host StorageClass parameter should match the StorageClass name created on the Harvester Cluster.
For guest Kubernetes owners, you may request that the Harvester cluster administrator create a new StorageClass.
If you leave the Host StorageClass field empty, the default StorageClass of the Harvester cluster will be used.

You can now create a PVC based on this new StorageClass, which utilizes the Host StorageClass to provision volumes on the bare-metal Harvester cluster.

RWX Volume Support

Prerequisites

Harvester v1.4 or later is installed on the host cluster.

You have created an RWX StorageClass on the host Harvester cluster.

On the Storage Class: Create screen, click Edit as YAML and specify the following:

kind: StorageClass
apiVersion: storage.k8s.io/v1
metadata:
  name: longhorn-rwx
provisioner: driver.longhorn.io
allowVolumeExpansion: true
reclaimPolicy: Delete
volumeBindingMode: Immediate
parameters:
  numberOfReplicas: "3"
  staleReplicaTimeout: "2880"
  fromBackup: ""
  fsType: "ext4"
  nfsOptions: "vers=4.2,noresvport,softerr,timeo=600,retrans=5"

The role-based access control (RBAC) settings are up-to-date.

RBAC authorization uses a specific Kubernetes API group to drive authorization decisions regarding access to computer or network resources.

The Harvester CSI driver requires the new RBAC settings to support RWX volumes. To check the RBAC settings, run the command kubectl get clusterrole harvesterhci.io:csi-driver -o yaml.

# kubectl get clusterrole harvesterhci.io:csi-driver -o yaml
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
...
  name: harvesterhci.io:csi-driver
...
rules:
- apiGroups:
  - storage.k8s.io
  resources:
  - storageclasses
  verbs:
  - get
  - list
  - watch
- apiGroups:
  - harvesterhci.io
  resources:
  - networkfilesystems
  - networkfilesystems/status
  verbs:
  - '*'
- apiGroups:
  - longhorn.io
  resources:
  - volumes
  - volumes/status
  verbs:
  - get
  - list

The networkfs-manager pods are running.

To check the status of the networkfs-manager pods, run the command kubectl get pods -n harvester-system | grep networkfs-manager.

Example:

# kubectl get pods -n harvester-system | grep networkfs-manager
harvester-networkfs-manager-2pxhm                       1/1     Running     4 (34m ago)    3h41m
harvester-networkfs-manager-8tst2                       1/1     Running     4 (37m ago)    3h41m
harvester-networkfs-manager-xvkgp                       1/1     Running     4 (37m ago)    3h41m

The Harvester CSI driver version is v0.1.20 or later.

Usage

Create a new StorageClass on the guest cluster.

On the StorageClass: Create screen, add a Host Storage Class parameter and specify the RWX StorageClass that you created on the host Harvester cluster.
Create an RWX PersistentVolumeClaim (PVC).

On the PersistentVolumeClaim: Create screen, configure the following settings:
- Volume Claim tab: Specify the new StorageClass.
- Customize tab: Select Many Nodes Read-Write.
Verify that the RWX PVC was created successfully.
Create two pods.

On the Pod: Create screen, specify the RWX PVC.