API-Server unreachable with 1/3 node down, HA?

Hi there,
I tested rancher with custom nodes (bare metal) on ubuntu.

  • 3 nodes with ALL roles, 1 node with only worker role
  • rancher is running on a seperate machine.
  • no significant load

this is running fine until one of the master nodes goes down. Rancher GUI shows “etc unavailable” and “node down” (expected). But also the kube-API is completely dead now.
kubectl get nodes returns:
error: the server doesn’t have a resource type “nodes”

kubectl version returns Client version and
Error from server (ServiceUnavailable): the server is currently unable to handle the request

my expectation was that the API server (and etcd nodes) will keep working with one of three nodes down, because all components are available on the other 2 nodes. If the API server goes down with one node missing, then this is not the HA I expected.

Did I miss something?