Three filler Pods each reserve most of a node in peaks-rqm8q0:
sh
kubectl get pods -n peaks-rqm8q0 -o wide
kubectl describe node peaks-node-2 | grep -A5 "Allocated resources"Create a class for urgent work:
sh
kubectl create priorityclass high --value=1000 --description="urgent work"Then run a Pod that needs more CPU than any node has left, and give it the class:
sh
kubectl apply -n peaks-rqm8q0 -f - <<EOF
apiVersion: v1
kind: Pod
metadata:
name: urgent
spec:
priorityClassName: high
containers:
- name: worker
image: stackmonsters/monster-market-worker
resources:
requests:
cpu: 1500m
EOF
kubectl get pods -n peaks-rqm8q0 -o wide
kubectl get events -n peaks-rqm8q0 --field-selector reason=Preempted