Playing with HPA

This small demo is an example of how to use horizontal pod autoscaling for scaling out workloads by CPU usage metrics in Kubernetes.

The applications included:

• batch: for starting new jobs
• worker: for "calculating interest"
• counter: keeping track of how many jobs have reported being done.

All three are built as a single binary and they use the variable job to determine what to run.

The queue chosen for this project is NSQ.

The k8s/ directory contains yaml for deployment to a Kubernetes environment.

Before you run

Install metrics-server

Clone https://github.com/kubernetes-sigs/metrics-server.git

kubectl apply -f deploy/1.8+

Edit the deployment

kubectl edit deployment/metrics-server -n kube-system

add the flags within args:

 - --kubelet-insecure-tls
 - --kubelet-preferred-address-types=InternalIP,ExternalIP,Hostname

Save and exit

How to run

Build

Build the image and push it to a registry that your cluster has access to docker build -t MY_IMAGE:TAG .

Modify the k8s/*.yaml files' image bit to reference your image.

Deploy to K8s

kubectl apply -f k8s/

Port forward the batch bit

kubectl port-forward service/batch 8080

Open http://localhost:8080/ in the browser

Ctrl + C to stop it

Now forward the counter

Port forward the batch bit

kubectl port-forward service/counter 8080

Open http://localhost:8080/get in the browser

Observing

kubectl top pods
kubectl top nodes
kubectl logs -f <pods/BATCH_POD_ID>