Autoscaled Kubernetes bundle

This bundle is a minimal deployment of an autoscaled Kubernetes using the Elastisys CharmScaler.

Overview

The current version autoscales the number of k8s workers running in your cluster depending on the avarage CPU usage over the worker machines.

Getting started

Deploy the bundle

juju deploy autoscaled-kubernetes

Configure the CharmScaler to give it access to the Juju model.

For more details on the CharmScaler setup process, read this

Minimal config.yaml example:

charmscaler:
  juju_api_endpoint: "[API address]:17070"
  juju_model_uuid: "[uuid]"
  juju_username: "[username]"
  juju_password: "[password]"

Apply the configuration

juju config charmscaler --file config.yaml

Wait for the deployment to settle

watch -c juju status --color

Fetch the Kubernetes config and the kubectl binary

mkdir ~/.kube
juju scp kubernetes-master/0:config ~/.kube/config
juju scp kubernetes-master/0:kubectl ./kubectl

Make sure things are running

./kubectl cluster-info
./kubectl get nodes

Autoscale

As a simple proof of concept create the loadgenerator Replication Controller.

./kubectl create -f https://cdn.rawgit.com/elastisys/bundle-autoscaled-kubernetes/master/loadgenerator.yaml

It will start off by deploying one pod which runs the stress tool.

To increase the load simply scale the replication controller. One pod consumes one CPU core.

./kubectl scale rc loadgenerator --replicas X

By default the CharmScaler will scale the k8s worker at 80% CPU usage, you can change this to, for example 50%, like this

juju config charmscaler scaling_cpu_max=50

Wait a few minutes and watch your Kubernetes deployment grow!

---

For more details read the documentation of the Kubernetes core bundle and the CharmScaler.