README.rst

Test driven infrastructure example

This is a sample project that achieve Test driven infrastructure.

Requirements:

• Easy to install
• Easy to use
• Speed testing
• Usable on normal hardware
• Usable on a CI server
• Should test a full deployment (i.e. on a not yet provisioned system).
• Should test an upgrade deployment (i.e. on an already provisioned system).

The stack

This is a brief description of the tools used in this project, but each has alternatives that you could use on your stack.

• Docker provide a fast system that is pretty close to production one and it is easy to install on a modern Linux distro. Alternatives: Qemu, Xen, VMWare, Libvirt...
• Vagrant can run multiples machines and provision them. Alternatives: Test-Kitchen, Beaker, shell script.
• Ansible. Alternatives: Puppet, Chef, Salt...
• Testinfra to run tests. Alternatives: Serverspec, shell script.
• Tox to setup the virtualenv and run vagrant and testinfra. Alternatives: Make, shell script.
• Github as git repository and pull request review. Alternatives: Gitlab, Gerrit.
• Travis as CI server to run tests on pull requests. Alternatives: Jenkins

Installation

You will need a recent (>= 1.7) vagrant that's supports docker: https://www.vagrantup.com/downloads

To install docker see https://docs.docker.com/installation/

Then install tox and requirements to build the virtualenv:

$ sudo apt-get install python-tox python-dev

Infrastructure code

The playbook is a simple Nginx installation that setup an website with a "Hello world" page.

Target images

Two docker images (based on ubuntu:trusty) are configured in the vagrant config, A default image that is not provisioned and a production image that is provisioned at the same state than your production servers (eg: master branch).

The default image is build using this Dockerfile:

$ docker build -t philpep/test-driven-infrastructure-example:default .

The production image is the default image that is provisioned:

$ vagrant up --no-provision --provider=docker default
$ vagrant provision default
==> default: Running provisioner: ansible...

PLAY [all] ********************************************************************

[... ansible stuff ...]
PLAY RECAP ********************************************************************
default                    : ok=6    changed=6    unreachable=0    failed=0


$ docker ps
CONTAINER ID        IMAGE                                              [...]
89ab9d4c3e52        philpep/test-driven-infrastructure-example:default [...]

$ docker commit 89ab9d4c3e52 philpep/test-driven-infrastructure-example:production
$ docker push philpep/test-driven-infrastructure-example:production

The tests

The tests are written using Testinfra.

• test_nginx.py simple test that validate nginx is working.
• test_same_state.py a test that check the website root directory is the same after full (default) or half (production) provisioning.

Before running the tests, we need to start the two containers:

$ vagrant up --no-provision --provider=docker

Then look at the tox config and run:

$ tox

Tox will:

• Setup a virtualenv and install dependencies from requirements.txt
• Provision with Ansible as specified in the vagrant config
• Run Testinfra against the two docker containers

Let's break the tests

There are three pull requests in this repository.

At a first look, all the patch seems corrects, but in fact they are not.

• #2 will break if you deploy a new server, but works on an already provisioned one.
• #4 will break an already provisioned server but works on a new one.
• #3 will result to a different state between your old and new servers.

Now think about your experience with infrastructure code, this is some of the common error patterns that you have or will encounter.

CI Server

See the travis config used to test pull requests.

This repository has also two Jenkins jobs:

• https://jenkins.philpep.org/job/test-driven-infrastructure-example/ Test the master branch
• https://jenkins.philpep.org/job/test-driven-infrastructure-example-pr/ Test the pull requests using Github pull request builder plugin

Workflow

A normal workflow can be applied:

$ git checkout -b awesome-feature origin/master

# code, test, fix code, test...

$ git push

# Make a pull request

Then when the pull request is merged and the new state applied to production servers, rebuild the production image and push it:

$ vagrant up --no-provision --provider=docker production
$ vagrant provision production
$ docker ps
CONTAINER ID        IMAGE                                              [...]
0164b99d5a3f        philpep/test-driven-infrastructure-example:production [...]
$ docker commit 0164b99d5a3f philpep/test-driven-infrastructure-example:production
$ docker push philpep/test-driven-infrastructure-example:production

You could also automate this build with Jenkins or Travis when changes are merged in the master branch.