Vagif Gafar:
Dear all,
What is the normal/approved procedure to take a node out of service? How should I move running pods out to the other nodes? What about system pods like kube-dns - should we force these to be removed from the node?
Vagif Gafar:
Ok, @Ravi kiran but system pod(s) will remain there, is it normal to use kubectl drain <node_name> --ignore-daemonsets ?
Ravi Singh:
For draining to work, the application should have replicaset. Once a node is drained, all pods (with ReplicaSet) will be EVICTED from the node and node is marked as “SchedulingDisable”. So even if coreDNS is gone from master, the controller manager will start them on other node as it’s protected by replica-set. If you have static pods like kube-apiserver, kube-controller-manger and kube-scheduler, they can run on drained node. --ignore-daemonsets coz they are also protected by RS but as we know daemonsets can already be running on each node