Migration from k3d to k3s

[DataHearth]

Hello folks !
I'm wondering if there's a way to migrate a k3d cluster to a standard k3s cluster. Can I just add the docker nodes directly as server/workers nodes on the k3s by passing the correct command ? Or is it a little bit trickier to do this kind of migration ?
If this is a little more difficult (like keeping persistant storage, etc...), I thing I would rather go for a very small VM with a k3s server node on it and a bigger VM as a k3s worker node.

P.S: I'm fairly new with kubernetes in general, so if I'm missing something, I'll be glad to know it 😃 !

Thanks for reading 😄 .

[iwilltry42]

Hi @DataHearth , interesting use case 😁
Since you first created the k3d cluster, that means, that your data is in that cluster (sqlite/etcd in the server node(s)).
You can try to back that data up and re-import it in a new k3s cluster.
What's your goal that you want to achieve though? Moving the cluster completely to k3s? Growing hybrid cluster with k3d and k3s nodes?

[DataHearth]

well, my cluster is actually a single node k3d (because I don't have any other other baremetal server available :p). So I though I could keep everything and migrate data to a new HA k3s cluster that fits my needs when I have a some barebone servers available.
I was wondering if I could achieve that easily with some scripts or even a scalable configuration inside k3s's core.
But, it seems that it is not like I though :D.

[iwilltry42]

Okay (side-note: k3d allows you to have an "HA" cluster on a single node... for testing) 🤔
Actually, the problem here is, that you're switching from single-server to multi-server (HA).
By default, in a single-server setup, k3s uses sqlite as the backend database, while in an HA setup (first server started with the --cluster-init flag), it will switch to etcd. So you don't even have a way of backup/restore that you could go to persist your data.
However, the good thing about Kubernetes in general is, that you can "just" apply all your manifests again to achieve the same setup in a matter of minutes in a new cluster.
If you have data (e.g. a database) persisted in PersistentVolumes inside the existing k3d cluster, you can stop the pods and then use docker cp to get the files from the k3d container onto your local disk, from where you can then move it to the new cluster.
Surely tedious, but it would work.
If you were to move to a similar k3s setup like what you have right now in k3d, you could've (potentially) backed up the database, or simpler, exported the whole /var/lib/rancher/k3s directory, which keeps all state of k3s and used this to build the new plain k3s cluster.

[DataHearth]

Ok ! Thanks for the note about k3d 😃 . Didn't know how the HA on a docker cluster :p.
Well, okay. Migration isn't really "easy" to do. I'll make sure I backup everything needed and then migrate my stack :D.
Thanks for giving me a direction :).

Aside the question, that could be cool feature to see on k3d and/or k3s ! But that would be huge feature, so I'm not quite sure it is possible to implement :p.

Thank you bud !

[iwilltry42]

Sure thing, good luck with your new setup :)
That really would be a large feature request.. however, we have something with regards to keeping state around on our plan 👍
(not exactly for k3d -> k3s migration though)

P.S.: feel free to mark the question as answered, if you think it's good enough for you