This message was deleted.

Nothing is HA with only 1 instance.

Since K3s updates are so fast, with only around 10 seconds when the server node is not available, you can often get away with it. If any of the workloads running on the agents need to interact with the Kubernetes API then that will fail while the server is down, so have retries when accessing the Kubernetes API. I think a summary could be that a single server node is not recommended in production. I have a home-lab cluster with a single server node and 3 agent nodes that's been running for 4 years... YMMV

Yeah i guess i should've rephrased my question as sort-of HA 🙂 We already run single-node k3s "clusters" in production for our edge locations and during updates we barely notice any downtime as the update is so fast so our experience working with k3s is good. My initial question was born out of a use-case where we wan't to serve an HA application with as minimal deviation from our current edge setup

Creating a 3-node HA cluster with the embedded etcd is very easy. But now you need to manage your etcd backup and test your recovery routes. You managed the backup and recovery for the sqlite in the single-server cluster, right? So it's just a bit more learning, right? 🙂