Do you recommend taking snapshots of master & worker servers?

[aidas-emersoft]

If you recreated the whole cluster from the snapshots and put them in the right private network, would it work?

[vitobotta]

You didn't specify what kind of snapshots you're talking about, so I'm assuming you mean the etcd snapshots that k3s can take automatically. If that's not what you meant, let me know, and I'll try to answer accordingly.

I plan to enable automatic etcd snapshots in hetzner-k3s, but I must admit, I don't find it particularly useful. I've never needed to restore a cluster from an etcd snapshot, but I suspect it could be quite problematic. Restoring a whole cluster from a snapshot would likely run into issues if the new nodes have different IPs and names. Plus, there might be other data discrepancies that could cause problems. It’s something that would require thorough testing before I trust the process.

I prefer using regular backups taken with Velero, along with automation. With the automation, I can easily create and set up a new cluster from scratch, including all necessary components like monitoring and logging. Then, I would just restore a Velero backup to get my apps back up and running. Since this repo is focused on Hetzner Cloud, Velero can back up Hetzner Cloud volumes using either Restic or Kopia, depending on your configuration. This way, you can restore your apps and data fully, including the data stored in volumes.

I think restoring apps and data into a new cluster is more reliable than trying to restore the same cluster from etcd snapshots.

[aidas-emersoft]

Thank you for the reply.

I meant server snapshots offered by Hetzner, but I believe you could run into the issues you've described like nodes gaining different IP addresses.

[vitobotta]

Ah, I see. A while ago, I tried using snapshots instead of regular images, and everything worked just fine. The only difference was that creating the nodes took a bit longer.