Deploying with Kubernetes

This page is under construction and does not work yet. Help is appreciated. Currently, the move from docker-compose to Kubernetes has low priority.

Purpose

Deploying your Ampersand application to production can be done fast and frequent. For this purpose we are working on deployment on the Kubernetes platform. This allows you to deploy easily on your laptop (for testing), in your own data center (The minimum you need is a virtual machiine that runs kubernetes), or in the cloud (to benefit from the reliability and security provided by hosting providers). We use Kubernetes because it is open, free, well known, widely used, and is continuously being impoved by a large community. It lets you deploy fast and lets you upgrade the application without taking it offline.

Way of working

  1. A Kubernetes cluster is needed as the platform from which to launch and maintain the application. If it is there, that's fine. Otherwise you get one from a provider, from a data center, or you create one yourself (e.g. on your laptop). The cluster is built on top of virtual or physical machines and hosts and exposes the services that constitute the application.

  2. On the platform I need a registry from which Kubernetes can pull the image(s).

I tried it out for myself. I stole my inspiration from Kevin Smets' instruction. Thank you Kevin! I ran the following from my terminal (i.e. MacOS cli), just for the purpose of getting some hands-on experience...

Requirements

To make a Kubernetes cluster, I used Minikube. Minikube requires that VT-x/AMD-v virtualization is enabled in BIOS. To check that this is enabled on OSX / macOS, I ran:

sysctl -a | grep machdep.cpu.features | grep VMX

If there's output, it works!

Prerequisites

  • kubectl

  • docker (for Mac)

  • minikube

  • virtualbox

Verify

Start

This can take a while, expected output:

Great! You now have a running Kubernetes cluster locally. Minikube started a virtual machine for you, and a Kubernetes cluster is now running in that VM.

Check k8s

Should output something like:

Within minikube, a docker platform is running. However, on my Mac I have another docker daemon running. So which docker do we want to talk to? In this case that is definitely the docker daemon that is inside minikube...

Use minikube's built-in docker daemon

(You might add this line to .bash_profile or .zshrc or ... to use minikube's daemon by default. Or if you do not want to set this every time you open a new terminal).

If I want to talk to the docker daemon on my Mac, I have to revert back to it by running:

When running docker ps, it showed the following output:

The list was in fact a little bit longer: it contained 22 containers.

Build, deploy and run an image on your local k8s setup

To setup a local registry, so Kubernetes can pull the image(s) from there:

To test whether there is a local registry, ask docker:

Build

First of, store all files (Dockerfile, my-app.yml, index.html) in this gist locally in some new (empty) directory.

You can build the Dockerfile below locally if you want to follow this guide to the letter. Store the Dockerfile locally, preferably in an empty directory and run:

You should now have an image named 'my-app' locally, check by using docker images (or your own image of course). You can then publish it to your local docker registry:

Running docker images should now output the following:

Deploy and run from docker-compose.yml

I have a file called docker-compose.yml, which I have used to deploy RAP with docker-compose (rather than Kubernetes). I used kompose to transform that into the necessary Kubernetes .yaml files.

I deployed straight from the docker-compose.yml file:

However, I can also create the necessary Kubernetes files and deploy from kubectl:

This converts docker-compose.yml into the following files:

Having run kompose up to deploy the three RAP services, I now have the following situation on my minikube:

(I got even more information with the command kubectl get all --output=wide.)

So we see three pods (containers) running. Each of the pods is wrapped in a service of its own and exposed internally on the local network of this cluster. (In Kubernetes, nodes, pods and services all have their own IPs. In many cases, the node IPs, pod IPs, and some service IPs on a cluster will not be routable, so they will not be reachable from a machine outside the cluster, such as your desktop machine.)

To access these services from a browser, we now need to expose these services to localhost. I did this by exposing both services from the minikube platform:

Deploy and run from Kubernetes yml files

Store the file below my-app.yml on your system and run the following:

You should now see your pod and your service:

The configuration exposes my-app outside of the cluster, you can get the address to access it by running:

This should give an output like http://192.168.99.100:30304 (the port will most likely differ). Go there with your favorite browser, you should see "Hello world!". You just accessed your application from outside of your local Kubernetes cluster!

Kubernetes GUI

Delete deployment of my-app

You're now good to go and deploy other images!

Reset everything

Version

Last tested on 2017 October 20th macOS Sierra 10.12.6

PreviousGraduate StudentNextAuthentication and access management with OAuth

Last updated