Scheduling pods using pod priority and preemption #

Pod priority and preemption scheduling policies can be applied in a KWOK cluster. For this particular scenario, the cluster will be limited to a particular resource range, then pod priority and preemption policy policies will be used to evict low-priority pods.

This image shows you what you should expect when testing this scenario. You can follow the step-by-step guide after seeing this.

Prerequisites #

• KWOK must be installed on the machine. See installation.
• Install kubectl

Create cluster #

kwokctl create cluster

View clusters #

This ensures that the cluster was created successfully.

kwokctl get clusters

Create a node #

kwokctl scale node --replicas 1 --param '.allocatable.cpu="4000m"'

Create priority classes (low and high) #

apiVersion: scheduling.k8s.io/v1
kind: PriorityClass
metadata:
  name: low-priority
value: 100
globalDefault: false
description: "This priority class is for low-priority pods."
---
apiVersion: scheduling.k8s.io/v1
kind: PriorityClass
metadata:
  name: high-priority
value: 200
globalDefault: false
description: "This priority class is for high-priority pods."

kubectl apply -f priority-classes.yaml

Deploy a low-priority Pod #

This allows any pod that matches this priority class to consume 3 CPUs off the node.

apiVersion: v1
kind: Pod
metadata:
  name: low-priority-pod
spec:
  priorityClassName: low-priority
  containers:
  - name: low-priority-container
    image: fake
    resources:
      requests:
        cpu: 3000m
      limits:
        cpu: 3000m

kubectl apply -f low-priority-pod.yaml

Deploy a high-priority pod #

This allows any pod that matches this priority class to consume 3 CPUs off the node. Since the node has a maximum of 4 CPUs, pods with a low-priority class that consumes over 1 CPU will be preempted.

apiVersion: v1
kind: Pod
metadata:
  name: high-priority-pod
spec:
  priorityClassName: high-priority
  containers:
  - name: high-priority-container
    image: fake
    resources:
      requests:
        cpu: 3000m
      limits:
        cpu: 3000m

kubectl apply -f high-priority-pod.yaml

Observe Preemption #

Now, observe the preemption process. The higher priority pod will preempt the lower priority pod due to limited resources.

kubectl get pods

NAME                READY   STATUS    RESTARTS   AGE
high-priority-pod   1/1     Running   0          5s

You should see that the low-priority pod has been terminated and the high-priority pod has been scheduled.

See more details about the preemption event #

kubectl describe pod high-priority-pod | awk '/Events:/,/pod to node/'

Events:
  Type     Reason            Age   From               Message
  ----     ------            ----  ----               -------
  Warning  FailedScheduling  11m   default-scheduler  0/1 nodes are available: 1 Insufficient cpu.
  Normal   Scheduled         11m   default-scheduler  Successfully assigned default/high-priority-pod to node-000000

The event log indicates that the high-priority pod preempted another pod due to its higher priority.

Delete the cluster #

kwokctl delete cluster

Conclusion #

This example demonstrates how to use KWOK to simulate a scheduling scenario based on setting a priority and preemption policy.