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.
Clone https://github.com/kubernetes-sigs/metrics-server.git
kubectl apply -f deploy/1.8+kubectl edit deployment/metrics-server -n kube-systemadd the flags within args:
- --kubelet-insecure-tls
- --kubelet-preferred-address-types=InternalIP,ExternalIP,HostnameSave and exit
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.
kubectl apply -f k8s/Port forward the batch bit
kubectl port-forward service/batch 8080Open 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 8080Open http://localhost:8080/get in the browser
kubectl top pods
kubectl top nodes
kubectl logs -f <pods/BATCH_POD_ID>