Elegant deployment with Fabric and Puppet.

Loom does the stuff Puppet doesn't do well or at all: bootstrapping machines, giving them roles, deploying Puppet code and installing reusable Puppet modules. It's useful for both serverless and master/agent Puppet installations.

$ sudo pip install loom

First of all, you create fabfile.py and define your hosts:

from fabric.api import *
from loom import puppet
from loom.tasks import *

env.user = 'root'
env.environment = 'prod'
env.roledefs = {
    'app': ['prod-app-1.example.com', 'prod-app-2.example.com'],
    'db': ['prod-db-1.example.com'],
}

You can then define any third-party Puppet modules you want in a file called Puppetfile:

forge "http://forge.puppetlabs.com"
mod "puppetlabs/nodejs"
mod "puppetlabs/mysql"

(This is for librarian-puppet, a tool for installing reusable Puppet modules. It can also install from Git, etc.)

Your own modules are put in a directory called modules/ in the same directory as fabfile.py. Roles are defined in a magic module called roles which contains manifests for each role. (If you've used Puppet before, this is a replacement for node definitions.)

For example, modules/roles/manifests/db.pp defines what the db role is:

class roles::db {
  include mysql
  # ... etc
}

And that's it!

Let's set up a database server. First, bootstrap the host (in this example, the single db host you defined in env.roledefs):

$ fab -R db puppet.install

Then install third party Puppet modules, upload your local modules, and apply them:

$ fab -R db puppet.update puppet.apply

Every time you make a change to your modules, you can run that command to apply them. Because this is just Fabric, you can write a task in fabfile.py to do it too:

@task
def deploy_puppet():
    execute(puppet.update)
    execute(puppet.apply)

Then you could use the included "all" task to update Puppet on all your hosts:

$ fab all deploy_puppet

It's only been tested on Ubuntu 12.04. I would like to support more things. Send patches!

Look at the source for now. It's all Fabric tasks, and they're pretty easy to read. (Sorry.)

• Ben Firshman
• Andreas Jansson
• Steffen L. Norgren