Ansible: One Role to Rule them All

I am a long time Ansible user and contributor (since 2012) and I have been struggling with a decent setup for a multi-environment case. I have been designing and re-designing a lot, until I came up with this design. And what a coincidence, a customer wanted a setup that was exactly this. So this concept is a real world setup, working in a production environment.

Did I get your attention? Read after the break, but take your time. it is a long read.

The setup is (really) easy. At least, that is what I think.

It is, in fact, just one repository in Git that has control over all Ansible roles used in the complete environment. This role is called setup.

The setup repository contains a couple of things:

  • The Ansible configuration
  • The complete inventory split into the different environments
  • All needed variables per environment, functionality or per host
  • Small playbooks to run per environment
  • Task lists that contain all tasks needed to get things done
  • The roles.yml, containing all the roles we build ourself
  • All (small) scripts to make things tick

The inventory

When looking at the inventory directory a little closer, there are a couple of files per environment. For instance, the dev environment contains three files mysql, wiki and zz_groups.

The mysql file contains the dev_mysql group, which contains all the hosts that are responsible for running MySQL in the development environment.


And similar the wiki file contains all machines that run the Wiki software in the development environment.


But the biggest “trick” in the dev environment are the files called zz_groups. These contain the dev group, with all the child groups.


The reason this file is called zz_groups is because of the parsing order of Ansible. When the Ansible inventory is a directory, Ansible collects all files in the directory and processes them in alphabetical order. But: A group can only be used as a child if it is already defined. So the definition of the dev group, containing all children should come last, hence the name zz_groups.

In the top-level of the inventory there is a zz_groups file as well, containing



This way you can run Ansible for the group wiki, mysql or the subgroups dev or dev_mysql, giving very fine-grained control over the set of hosts involved in the run.

Inventory layout

The variables

For the variables the standard setup is used. Variables are defined per group, but we use a directory per group, instead of a file. All files in such a group directory are merged together by Ansible, ensuring all variables are available. The only reason to split it into separate files is for maintainability, every item has its file, e.g. inventory/group_vars/dev/mysql_users:

# Users for geerlingguy.mysql role

  - name: "{{ nagios_mysql_user }}"
    host: ''
    password: "{{ nagios_mysql_password }}"
    priv: "mysql.*:SELECT"
  - name: localweb-admin
    host: '192.168.0.%'
    password: !vault |
    priv: '*.*:SELECT/*.*:CREATE VIEW/*.*:INSERT'

Of course all defaults are in the all group directory, because all hosts in the inventory are automatically member of the all group.

Group variables

The roles

When the inventory and the variables are in place, all the involved roles are need to be setup. Every role should have a branch for each environment, so a dev branch for the dev environment and so on. And all needed roles should be defined in the roles.yml file, like so:

- src:
  scm: git
  version: master
  name: apache

The version: tag should be present, but the value is ignored, it will be replaced by the refresh script with the branch that is currently being checked out.

Putting it all together

Once all the hosts, variables, roles and roles-files are in place, the next thing to do is to roll it onto the Ansible control-node.

The order is:

$ ssh
# cd /etc
# mv ansible
# git clone ansible
# cd ansible
# ./refresh -f

Of course, after the initial checkout, the only command needed is the refresh command. This will ensure all environments are filled with the correct information.

Now run Ansible through ansible_run for the selected environment and watch the magic happen.

All together

Where to get it

You can get the framework from the Github “One Role” repositories. Do read the docs in the docs repository to get more familiar with the concept.

I presented this concept in the Ansible Room at CfgMgmtCamp on February 5th in Ghent, Belgium. The slide deck is available at SpeakerDeck.

See also