This message was deleted.
# harvester
a
This message was deleted.
g
i guess you could.. but it would need a bit of manual work: https://github.com/harvester/cloud-provider-harvester
i have not tried it in a while.. but it should work with any k8s distro
s
Thank you for pointing me in a direction. I’m struggling on how to do this without using Rancher. I’m relying on using Terraform to spin up and configure new clusters on Harvester. Reading how to deploy the CSI Driver, it says the Harvester Cloud Provider is optional if you’re just interested in the CSI Driver, but which method should I be pursuing? Deploying to RKE1, RKE2, or K3s? I know this is likely outside of the recommended use case for Harvester, but I feel like Harvester has the potential to be a great Standalone platform for hosting any k8s flavor on VMs without needing Rancher, especially if there was an easy way to give those guest clusters access to the underlying Longhorn storage with the CSI Driver. Am I crazy? 😄
g
i am not sure what is involved in k0s but you would need to spin up the cluster with correct cloud-provider=external flags
once the vm's are bootstrapped and k0s is installed.. you will need to manually go and install the cloud-provider manifests
but that will be after you have generated a kubeconfig with access to namespace where the vm's exist
s
Thank you. I think maybe I’m getting somewhere. I started down the Cloud Provider road, but it was feeling wrong since I’m only interested in the CSI Driver. Your advice seems to be taking me in the right direction for that as well though. I cloned the
harvester-csi-driver
git repository and ran the
generate_addon_csi.sh
. According to the instructions I’m supposed to copy/paste the
cloud-init user data
into the User Data field through Rancher, but since I’m not using Rancher, I’m having to improvise. I copied the
Secret
manifest from the generated file and applied it to my k0s cluster, after running the
deployment.yaml
file the
harvester-csi-plugin
was created and was in
ContainerCreating
, then I saw it was looking for a
/var/lib/kubelet/plugins_registry
folder on the host, so I created that folder. Now the pods are in CrashLoopBackOff status. The logs say
level=fatal msg="Error running CSI driver: flag provided but not defined: -drivername"
. What’s supposed to be in the
plugins_registry
folder that I had to create?
Looking at more logs, under the
csi-controller
pod there is a
csi-provisioner
container that has an error of
unknown flag: --default-fstype
I’ve switched to using RKE2, and trying to deploy the CSI Driver without Rancher Server. I ended up in exactly the same place. Same errors happening. One thing that was different was when I generated the config using the RKE2 parameter, it didn’t generate a
harvester-csi-config
manifest in the output, but the
harvester-csi-plugin
was still looking for that Secret, so I ran the generator again without `RKE2`at the end so I could get that Secret.