First check to see where the Pods are running:
kubectl get po -l app=whoami -o wide
Now use the cordon command to mark the agent-1 node as unschedulable, so no new Pods will be allocated to it:
kubectl cordon k3d-lab-clusters-agent-1
Drain the node to remove all Pods, using the flag to ignore the Pods which are managed by DaemonSets (otherwise those Pods would not be removed):
kubectl drain k3d-lab-clusters-agent-1 --ignore-daemonsets --delete-emptydir-data
Check the Pod list now and you'll see the agent-1 Pods are terminated, and replacements get created; check the node list and you'll see the status of the cordoned node:
kubectl get po -l app=whoami -o wide
kubectl get nodes -o wide
When the work is finished on the server, it can be made available for scheduling again:
kubectl uncordon k3d-lab-clusters-agent-1
Bringing a node online again doesn't cause Pods to be redistributed - agent-1 is available again but it won't run any whoami Pods unless you restart the rollout so new Pods are scheduled:
kubectl rollout restart deploy whoami
kubectl get po -l app=whoami -o wide