How To Avoid Placement of More Than One Pod Replicas on Same Worker Node

Problem

• How to avoid assignment of multiple replicas of a pod on a single worker node?

• How to configure deployment to avoid co-location of replicas on a single node?

Environment

• Platform9 Managed Kubernetes - All Versions

• Kube-Scheduler

Procedure

• PodAntiAffinity can be configured to ensure that the scheduler does not co-locate replicas on a single node.

• Below mentioned is the snippet of PodAntiAffinity configured for a Nginx deployment with 3 pod replicas.

podAntiAffinity Configuration

# kubectl get deployment nginx-deployment -o yaml | grep -w "podAntiAffinity:" -A8
        podAntiAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:
          - labelSelector:
              matchExpressions:
              - key: app
                operator: In
                values:
                - nginx
            topologyKey: kubernetes.io/hostname

• This configuration makes sure that all 3 replicas of the pods are scheduled on three different nodes of the cluster.

kubectl output

# kubectl get po -o wide -n default
NAME                               READY   STATUS    RESTARTS    AGE       IP              NODE           
nginx-deployment-84c6674589-cxp55   1/1     Running   0          55s     10.20.152.138   10.128.228.29   
nginx-deployment-84c6674589-hzmnn   1/1     Running   0          55s     10.20.155.70    10.128.228.28   
nginx-deployment-84c6674589-vq4l2   1/1     Running   0          55s     10.20.225.7     10.128.226.44

Additional Information

Kubernetes Docs – Assigning Pods to Nodes