Node NotReady or needing replacement

A node stops reporting, or you need to take one out for maintenance without dropping traffic. All playbooks.

List nodes with their status, version and IP addresses.
```
kubectl get nodes -o wide
```
Check the node's conditions (MemoryPressure, DiskPressure, PIDPressure, Ready) and recent events.
```
kubectl describe node <node-name>
```
Stop new pods landing on it.
Cordon a node

Marks a node unschedulable; running pods stay where they are.
```
kubectl cordon <node-name>
```
nodes, maintenance
Move its pods elsewhere. PodDisruptionBudgets can make this wait; that is them doing their job.
Safely drain a node

Cordons the node and evicts its pods before maintenance, respecting PodDisruptionBudgets.

Caution. --delete-emptydir-data deletes the data in emptyDir volumes on that node.
```
kubectl drain <node-name> --ignore-daemonsets --delete-emptydir-data
```
node, maintenance, drain
With Karpenter, delete the node's NodeClaim instead: Karpenter drains the node and provisions a replacement.
```
kubectl delete nodeclaim <nodeclaim-name>
```
Watch the replacement come up.
Watch nodes being created and removed

Streams NodeClaim changes as Karpenter provisions and consolidates nodes.
```
kubectl get nodeclaims -w
```
nodeclaims, watch, consolidation