Priority Classes

A critical payment processing service goes Pending because the cluster is full of lower-priority batch Jobs. Without priority classes, all Pods are equal and the scheduler has no mechanism to make room. With priority classes, the scheduler can evict lower-priority Pods to schedule higher-priority ones.

A PriorityClass is a cluster-level object that maps a name to an integer priority value. Pods reference a PriorityClass by name. Higher values mean higher priority. The scheduler uses priority both for scheduling decisions and for preemption: evicting lower-priority Pods to free capacity for higher-priority ones.

Creating a PriorityClass

nano priority-classes.yaml

apiVersion: scheduling.k8s.io/v1
kind: PriorityClass
metadata:
  name: critical-workload
value: 1000000
globalDefault: false
description: 'For production-critical services'
---
apiVersion: scheduling.k8s.io/v1
kind: PriorityClass
metadata:
  name: batch-workload
value: 100
globalDefault: false
description: 'For batch and background processing'

kubectl apply -f priority-classes.yaml
kubectl get priorityclasses

The built-in system classes appear too: system-cluster-critical (value: 2000000000) and system-node-critical (value: 2000001000). These are used by Kubernetes system components and have very high values to prevent them from being preempted.

Assigning priority to a Pod

nano high-priority-pod.yaml

apiVersion: v1
kind: Pod
metadata:
  name: high-priority-pod
spec:
  priorityClassName: critical-workload
  containers:
    - name: app
      image: busybox:1.36
      command: ['sh', '-c', 'sleep 3600']
      resources:
        requests:
          cpu: '100m'
          memory: '64Mi'

kubectl apply -f high-priority-pod.yaml
kubectl get pod high-priority-pod -o jsonpath='{.spec.priority}'

The numeric value 1000000 is shown. This is the value that the scheduler uses for comparison.

How preemption works

When a Pod cannot be scheduled because of insufficient resources, the scheduler checks if preemption can help: are there lower-priority Pods whose eviction would free enough capacity? If yes, the scheduler evicts them, making room for the higher-priority Pod.

Preempted Pods receive a graceful termination signal. They are not killed immediately. They have terminationGracePeriodSeconds (default 30 seconds) to shut down cleanly.

The globalDefault field

A PriorityClass with globalDefault: true is automatically assigned to any Pod that does not specify a priorityClassName. Only one PriorityClass can have globalDefault: true at a time. This is useful for ensuring all Pods in the cluster have a priority value even if developers forget to set one.

kubectl get pod kube-apiserver-sim-control-plane -n kube-system -o jsonpath='{.spec.priorityClassName}'

The output shows system-cluster-critical. This is why control plane Pods are never preempted by user workloads.

kubectl delete pod high-priority-pod
kubectl delete priorityclass critical-workload batch-workload

Priority classes are the mechanism that ensures critical services survive resource pressure. Assign them carefully, restrict access to high-value classes, and pair them with resource quotas. The next lesson covers multiple schedulers, a less common but exam-relevant scenario.