Pod Eviction on Down Node
When a node goes down, the Kube Controller Manager waits 5 minutes for it to come back up, if it does not, those pods part of a replica set will be started on other nodes, this time is known as the “pod eviction timeout” and is configurable by setting:
kube-controller-manager --pod-eviction-timeout=5m0s
Draining a Node
kubectl drain node-1
may need to force with
kubectl drain node-1 --ignore-daemonsets=true
There are many other flags that may need to be used like --force
Do maintenance then
kubectl uncordon node-1