Skip to content

Cluster Autoscaler Add-on

The Cluster Autoscaler add-on adds support for Cluster Autoscaler to an EKS cluster. Cluster Autoscaler is a tool that automatically adjusts the number of nodes in your cluster when:

  • pods fail due to insufficient resources, or
  • pods are rescheduled onto other nodes due to being in nodes that are underutilized for an extended period of time.


import 'source-map-support/register';
import * as cdk from 'aws-cdk-lib';
import * as blueprints from '@aws-quickstart/eks-blueprints';

const app = new cdk.App();

const addOn = new blueprints.addons.ClusterAutoScalerAddOn();

const blueprint = blueprints.EksBlueprint.builder()
  .build(app, 'my-stack-name');


  1. Configure proper IAM permissions (such as modify autoscaling groups, terminate instances, etc.) as a Policy.
  2. Configures IAM Role for Service Account (IRSA) with the generated policy.
  3. Resolves proper CA image to pull based on the Kubernetes version.
  4. Applies proper tags for discoverability to the EC2 instances.
  5. Supports standard helm configuration options.

The add-on automatically sets the following Helm Chart values, and it is highly recommended not to pass these values in (as it will result in a failed deployment): - cloudProvider - autoDiscovery.clusterName - awsRegion - rbac.serviceAccount.create -

Testing the scaling functionality

The following steps will help test and validate Cluster Autoscaler functionality in your cluster.

  1. Deploy a sample app as a deployment.
  2. Create a Horizontal Pod Autoscaler (HPA) resource.
  3. Generate load to trigger scaling.

Deploy a sample app

Take a note of the number of nodes available:

kubectl get nodes
NAME                                         STATUS   ROLES    AGE   VERSION   Ready    <none>   80m   v1.19.6-eks-49a6c0

The first step is to create a sample application via deployment and request 20m of CPU:

kubectl create deployment php-apache
kubectl set resources deploy php-apache --requests=cpu=20m 
kubectl expose deployment php-apache --port 80

You can see that there's 1 pod currently running:

kubectl get pod -l app=php-apache
NAME                          READY   STATUS    RESTARTS   AGE
php-apache-55c4584468-vsbl7   1/1     Running   0          63s

Create HPA resource

Now we can create Horizontal Pod Autoscaler resource with 50% CPU target utilization, and the minimum number of pods at 1 and max at 20:

kubectl autoscale deployment php-apache \
    --cpu-percent=50 \
    --min=1 \

You can verify by looking at the hpa resource:

kubectl get hpa
php-apache   Deployment/php-apache   10%/50%   1         50        2          52s

Generate load

With the resources created, you can generate load on the apache server with a busybox container:

kubectl --generator=run-pod/v1 run -i --tty load-generator --image=busybox /bin/sh

You can generate the actual load on the shell by running a while loop:

while true; do wget -q -O - http://php-apache; done

Verify that Cluster Autoscaler works

While the load is being generated, access another terminal to verify that HPA is working. The following command should return a list of many nods created (as many as 10):

kubectl get pods -l app=php-apache -o wide --watch

With more pods being created, you would expect more nodes to be created; you can access the Cluster Autoscaler logs to confirm:

kubectl -n kube-system logs -f deployment/blueprints-addon-cluster-autoscaler-aws-cluster-autoscaler

Lastly, you can list all the nodes and see that there are now multiple nodes:

kubectl get nodes
NAME                                         STATUS   ROLES    AGE   VERSION    Ready    <none>   73s   v1.19.6-eks-49a6c0   Ready    <none>   84m   v1.19.6-eks-49a6c0   Ready    <none>   46s   v1.19.6-eks-49a6c0   Ready    <none>   90s   v1.19.6-eks-49a6c0