This message was deleted.

You can disable all the control-plane components on a server and run it as an etcd-only node. But if your etcd is io-bound that doesn't generally help since the control plane itself generates almost no io. What would help most is to put additional iops behind etcd.

Is there documentation on how to do this, i didnt see anything specifically relating to this on my few read throughs? Aside, our limitations are coming from cpu scheduling on the hypervisor. The API and etcd are both heavy resource consumers, with a fixed physical node size, bigger VMs don't improve performance past a certain point, IO scheduling is the first thing to fall down on shared hardware