Preemption lets a higher-priority Pod obtain capacity by proactively removing lower-priority Pods instead of remaining Pending until resources become free. Configure it through a PriorityClass and assign that class to the workload. section 5.1.3
Kubernetes applies priority and preemption in several related decisions:
- The scheduler considers higher-priority Pending Pods first. section 5.1.3
- Preemption removes lower-priority running Pods when no capacity is available and the higher-priority Pod cannot otherwise be scheduled. section 5.1.3
- A PriorityClass with preemptionPolicy: Never can rank workloads without allowing them to displace lower-priority Pods. section 5.1.3
- Managed workloads such as Deployments recreate removed Pods; standalone Pods are not rescheduled automatically. section 5.1.3
Keeping capacity ready for urgent workloads
Run low-priority placeholder Pods in a Deployment to reserve headroom. When a regular Pod with a higher priority arrives, it preempts the placeholders, allowing rapid scheduling. Set placeholder Pods to a lower priority and preemptionPolicy: Never; use multiple replicas sized at least as large as the largest regular Pod. section 6.3.2
Listing 5.5 Chapter05/5.1.3_Priority/priorityclass-preemption.yaml
apiVersion: scheduling.k8s.io/v1 kind: PriorityClass metadata: name: high-priority-preemption value: 1000000 preemptionPolicy: PreemptLowerPriority #1 globalDefault: false description: "Critical services." #1 This priority class will cause eviction of lower priority Pods if there is no available capacity in the cluster.
Inspect a PriorityClass configured to preempt lower-priority Pods when cluster capacity is unavailable. section 5.1.3
Listing 6.10 Chapter06/6.3.2_PlaceholderPod/placeholder-deploy.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: placeholder
spec:
replicas: 10 #1
selector:
matchLabels:
pod: placeholder-pod
template:
metadata:
labels:
pod: placeholder-pod
spec:
priorityClassName: placeholder-priority #2
terminationGracePeriodSeconds: 0 #3
containers:
- name: ubuntu
image: ubuntu
command: ["sleep"] #4
args: ["infinity"] #4
resources:
requests:
cpu: 200m #5
memory: 250Mi #5
Inspect the Deployment pattern for low-priority placeholder Pods that provide spare scheduling capacity. section 6.3.2
