Scheduling pods using pod topology spread constraints #

Pod topology spread constraints scheduling policies can be applied in a KWOK cluster.

For this particular scenario, the cluster has 4 nodes that span 2 regions. A pod topology spread constraint policy will be applied, so the pods are evenly scheduled across each region.

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 nodes #

kwokctl scale node --replicas 4

Label nodes #

Below are the node resource specifications:

Node: 4 worker nodes in the cluster
- Node 1:
  - Key = topology.kubernetes.io/region
  - Value = us-west-1
- Node 2:
  - Key = topology.kubernetes.io/region
  - Value = us-west-1
- Node 3:
  - Key = topology.kubernetes.io/region
  - Value = us-west-2
- Node 4:
  - Key = topology.kubernetes.io/region
  - Value = us-west-2

kubectl label nodes node-000000 topology.kubernetes.io/region=us-west-1
kubectl label nodes node-000001 topology.kubernetes.io/region=us-west-1
kubectl label nodes node-000002 topology.kubernetes.io/region=us-west-2
kubectl label nodes node-000003 topology.kubernetes.io/region=us-west-2

Create deployment #

deployment.yaml ↕

apiVersion: apps/v1
kind: Deployment
metadata:
  name: fake-app
spec:
  replicas: 4
  selector:
    matchLabels:
      app: fake-app
  template:
    metadata:
      labels:
        app: fake-app
    spec:
      topologySpreadConstraints:
      - maxSkew: 1
        topologyKey: "topology.kubernetes.io/region"
        whenUnsatisfiable: DoNotSchedule
        labelSelector:
          matchLabels:
            app: fake-app
      containers:
      - name: fake-container
        image: fake

kubectl apply -f deployment.yaml

The deployment creates 4 replicas, with its maxSkew set to 1.

Observe the topology spread #

kubectl get pod -o wide

NAME                        READY   STATUS    RESTARTS   AGE   IP         NODE          NOMINATED NODE   READINESS GATES
fake-app-685c7b9cc7-2n8lv   1/1     Running   0          5s    10.0.2.2   node-000002   <none>           <none>
fake-app-685c7b9cc7-2s2gf   1/1     Running   0          5s    10.0.1.1   node-000001   <none>           <none>
fake-app-685c7b9cc7-7zvm9   1/1     Running   0          5s    10.0.1.2   node-000001   <none>           <none>
fake-app-685c7b9cc7-fhntg   1/1     Running   0          5s    10.0.2.1   node-000002   <none>           <none>

All pods are evenly distributed among each node across the regions.

Delete the cluster #

kwokctl delete cluster

Conclusion #

This example demonstrates how to use KWOK to simulate a scheduling scenario based on setting a pod topology spread constraints policy.