Installing Longhorn

21 Dec 2021 11:28 raspberry-pi longhorn k3s

Installation with Helm, per https://longhorn.io/docs/1.2.3/deploy/install/install-with-helm/.

Add the repo

helm repo add longhorn https://charts.longhorn.io
helm repo update

Run the installation

helm install longhorn longhorn/longhorn --namespace longhorn-system --create-namespace

Make Longhorn the default storage class

Per https://rpi4cluster.com/k3s/k3s-storage-setting/#make-longhorn-the-default-storageclass:

$ kubectl get storageclass
NAME                   PROVISIONER             RECLAIMPOLICY   VOLUMEBINDINGMODE      ALLOWVOLUMEEXPANSION   AGE
local-path (default)   rancher.io/local-path   Delete          WaitForFirstConsumer   false                  20h
longhorn (default)     driver.longhorn.io      Delete          Immediate              true                   16m

$ kubectl patch storageclass local-path -p '{"metadata": {"annotations":{"storageclass.kubernetes.io/is-default-class":"false"}}}'

$ kubectl get storageclass
NAME                 PROVISIONER             RECLAIMPOLICY   VOLUMEBINDINGMODE      ALLOWVOLUMEEXPANSION   AGE
longhorn (default)   driver.longhorn.io      Delete          Immediate              true                   19m
local-path           rancher.io/local-path   Delete          WaitForFirstConsumer   false                  20h

Restarting a node (to install Ubuntu updates, e.g.) causes the local-path storage class to be re-registered as the default. See this issue. Consider disabling it entirely, or just be aware that occasionally you’ll get Internal error occurred: 2 default StorageClasses were found errors, and you’ll have to patch it again.

User Interface

The Longhorn frontend is accessible through a ClusterIP endpoint:

$ kubectl --namespace longhorn-system get service longhorn-frontend
NAME                TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)   AGE
longhorn-frontend   ClusterIP   10.43.174.121   <none>        80/TCP    9d

We can expose this in a number of different ways:

kubectl port-forward

$ kubectl --namespace longhorn-system port-forward --address 0.0.0.0 service/longhorn-frontend 5080:80

This makes the UI accessible at http://rpi401:5080/#/dashboard. If you’re running kubectl on one of the nodes, you’ll need the --address 0.0.0.0 option, but note that this is insecure: it exposes the UI to the local network. If you’re running kubectl on your desktop, you can omit the --address 0.0.0.0, and the port will be opened to localhost only.

ssh port forward

$ ssh -L5080:10.43.174.121:80 ubuntu@rpi401

This makes the UI accessible at http://localhost:5080/#/dashboard. It’s more secure (localhost only, and the traffic goes over SSH), but it requires figuring out the ClusterIP, rather than just specifying the service name.

Tip: You can use an SSH escape sequence to forward the port without disconnecting. Press Enter, then ~C, then L5080:10.43.174.121:80, then Enter.