I don't know if this is related to

rke

, the rancher stuff built around kube-apiserver in

rancher/hyperkube

& etd in

rancher/coreos-etcd

, or some issue native to api-server and/or etcd, but: Every once in a while I observe that out of my 3 rke-provisioned master nodes, two of them start having heavy traffic from

kube-apiserver

one one master node to

etcd

on a different node (pretty much sustained 150-300MBPs send/receive). The only way to resolve this is to

docker restart kube-apiserver

on all the master nodes. The problem condition goes away for a few days and then spontaneously comes back until I bounce the kube-apiserver containers again. I've observed this in at least rke v0.2.4 & rke v0.2.5. Looking through the kube-apiserver and etcd logs don't seem to show anything unusual before/during the event Has anyone else noticed this behavior?