allardWhat this controller does (precisely)
Continuously: 1. Watches Nodes 2. If a Node is NotReady longer than a configurable timeout 3. Finds Pods on that node 4. Deletes only Pods that: • are owned by Deployments or StatefulSets • are NOT DaemonSets 5. Lets Kubernetes reschedule them on healthy nodes
It does not: • Drain nodes • Touch DaemonSets • Touch mirror/static pods • Interfere with kube-system (unless you allow it)
NOT_READY_GRACE_SECONDS=600 # default: 10 minutes