This message was deleted.

I always did three nodes with those three having control plane and etcd. Not sure specific reasoning for person below splitting them, but it's certainly true that etcd needs an odd number of nodes and you need >1 to be HA for things in general.

correct - that is what I have always done also. but I have never had so many worker/agent nodes (10 is the plan) so am wondering if 3 control plane / etcd is still enough for 10 worker/agents ?

I hazily recall either Rancher or vanilla kubernetes docs having recommendations on how many nodes required scaling up. I think 10 workers was still ok for a single control plane, so certainly wouldn't hurt to be prepared, but things like faster disks for where ETCD is writing might be more effective than more nodes if you do get some slowdowns.

good point - thanks. for disk, plan is to have dedicated datastore with SSDs for etcd / control plane nodes. worker/agent nodes are also SSDs but saperate datastore

I'd look for the official docs to verify before trusting my hazy memory. Might also look for the considerations for splitting etcd from control plane (those would be in vanilla kubernetes for certain).