Horizontal Pod Autoscaler

The Kubernetes Horizontal Pod Autoscaler automatically scales the number of Pods in a deployment, replication controller, or replica set based on that resource's CPU utilization. This can help your applications scale out to meet increased demand or scale in when resources are not needed, thus freeing up your nodes for other applications. When you set a target CPU utilization percentage, the Horizontal Pod Autoscaler scales your application in or out to try to meet that target.

The Horizontal Pod Autoscaler is a standard API resource in Kubernetes that simply requires that a metrics source (such as the Kubernetes metrics server) is installed on your Amazon EKS cluster to work. You do not need to deploy or install the Horizontal Pod Autoscaler on your cluster to begin scaling your applications. For more information, see Horizontal Pod Autoscaler in the Kubernetes documentation.

Use this topic to prepare the Horizontal Pod Autoscaler for your Amazon EKS cluster and to verify that it is working with a sample application.

Note

This topic is based on the Horizontal Pod autoscaler walkthrough in the Kubernetes documentation.

Prerequisites

• You have an existing Amazon EKS cluster. If you don't, see Getting started with Amazon EKS.

• You have the Kubernetes Metrics Server installed. For more information, see Installing the Kubernetes Metrics Server.

• You are using a kubectl client that is configured to communicate with your Amazon EKS cluster.

Run a Horizontal Pod Autoscaler test application

In this section, you deploy a sample application to verify that the Horizontal Pod Autoscaler is working.

Note

This example is based on the Horizontal Pod autoscaler walkthrough in the Kubernetes documentation.

To test your Horizontal Pod Autoscaler installation

1. Deploy a simple Apache web server application with the following command.

   kubectl apply -f https://k8s.io/examples/application/php-apache.yaml
   

   This Apache web server Pod is given a 500 millicpu CPU limit and it is serving on port 80.

2. Create a Horizontal Pod Autoscaler resource for the php-apache deployment.

   kubectl autoscale deployment php-apache --cpu-percent=50 --min=1 --max=10
   

   This command creates an autoscaler that targets 50 percent CPU utilization for the deployment, with a minimum of one Pod and a maximum of ten Pods. When the average CPU load is lower than 50 percent, the autoscaler tries to reduce the number of Pods in the deployment, to a minimum of one. When the load is greater than 50 percent, the autoscaler tries to increase the number of Pods in the deployment, up to a maximum of ten. For more information, see How does a HorizontalPodAutoscaler work? in the Kubernetes documentation.

3. Describe the autoscaler with the following command to view its details.

   kubectl get hpa
   

   An example output is as follows.

   NAME         REFERENCE               TARGETS   MINPODS   MAXPODS   REPLICAS   AGE
   php-apache   Deployment/php-apache   0%/50%    1         10        1          51s

   As you can see, the current CPU load is 0%, because there's no load on the server yet. The Pod count is already at its lowest boundary (one), so it cannot scale in.

4. Create a load for the web server by running a container.

   kubectl run -i \
       --tty load-generator \
       --rm --image=busybox \
       --restart=Never \
       -- /bin/sh -c "while sleep 0.01; do wget -q -O- http://php-apache; done"
   

5. To watch the deployment scale out, periodically run the following command in a separate terminal from the terminal that you ran the previous step in.

   kubectl get hpa php-apache
   

   An example output is as follows.

   NAME         REFERENCE               TARGETS    MINPODS   MAXPODS   REPLICAS   AGE
   php-apache   Deployment/php-apache   250%/50%   1         10        5          4m44s

   It may take over a minute for the replica count to increase. As long as actual CPU percentage is higher than the target percentage, then the replica count increases, up to 10. In this case, it's 250%, so the number of REPLICAS continues to increase.

   Note

   It may take a few minutes before you see the replica count reach its maximum. If only 6 replicas, for example, are necessary for the CPU load to remain at or under 50%, then the load won't scale beyond 6 replicas.

6. Stop the load. In the terminal window you're generating the load in, stop the load by holding down the Ctrl+C keys. You can watch the replicas scale back to 1 by running the following command again in the terminal that you're watching the scaling in.

   kubectl get hpa
   

   An example output is as follows.

   NAME         REFERENCE               TARGETS   MINPODS   MAXPODS   REPLICAS   AGE
   php-apache   Deployment/php-apache   0%/50%    1         10        1          25m

   Note

   The default timeframe for scaling back down is five minutes, so it will take some time before you see the replica count reach 1 again, even when the current CPU percentage is 0 percent. The timeframe is modifiable. For more information, see Horizontal Pod Autoscaler in the Kubernetes documentation.

7. When you are done experimenting with your sample application, delete the php-apache resources.

   kubectl delete deployment.apps/php-apache service/php-apache horizontalpodautoscaler.autoscaling/php-apache