Our Experience With Vagrant and Puppet
In this post I will try and outline the pros and cons and lessons learnt over the past week while pairing with Ginette and setting up our own VM managed with Vagrant.
Our project uses Rails with Postgres and Nginx/Passenger. We also use RVM to manage Ruby versions.
The reasons to use Vagrant were primarily these:
- Use one VM per project without cluttering your main machine with the sundry dependencies (MySQL, Postgres, Redis, Apache, Nginx, Passenger, et al) that projects invariably have.
- Have a 5 minute setup for any new developer who joins our team to
get up and running. This should be as easy as running a single
vagrant up. No installation, no troubleshooting. It should “just work” ™
- Have a consistent environment across all developers machines.
- Have a simple command line based workflow:
- Create reusable configuration scripts using Puppet that could be used on Test/Staging and Production environments as well. (Ambitious)
vagrant up“just works” ™ ! Excellent for hiding details and letting people focus on GTD. This also improves productivity by avoiding “it doesn’t work on my machine” delays! Because everything “just works” ™ on the VM.
- Setting up a Vagrant base box is not hard. If you setup a minimal base box with a operating system (sans GUI) and SSH, you can reuse it on every project and do the provisioning through Puppet/Chef.
- We recommend using a pre-existing base box, installing dependencies with a package manager (apt-get, etc.) and then hosting it somewhere on your local network for fellow devs to use as a pre-packaged VM. This is a quick and easy way to get setup.
- Our biggest roadblocks were with getting Puppet to do our bidding. We tried to reuse existing Puppet modules for installing dependencies. We managed to get RVM and Postgres installed through Puppet. But we just couldn’t seem to get Nginx/Passenger to work.
- Puppet modules tended to be either poorly documented / not maintained or just did not work! Maybe we were doing something wrong, but it was really hard to tell!
- Base boxes for your choice of operating system might not be readily available! (This is not a major disadvantage as it is quite easy to create your own base box, but it is time consuming)
We started by watching a short but informative
presentation by Cyril
Rohr about using Vagrant with Puppet. It was enough to convince us to
pursue the magical
How to setup Vagrant
Just follow these simple
instructions on the Vagrant site. You can try it out using the
lucid32 base box that the creators of Vagrant have provided. If the
setup was easy enough, you should already be convinced that Vagrant is
a useful tool ;-)
Customizing the VM for your project
Setting up Vagrant was easy. The more complicated part was setting up a Virtual machine for use on the project.
There were two strategies that we thought of:
- Create a fully loaded base box with all the dependencies pre-installed. This base box can be hosted on a URL that can be
shared with the team and added to the Vagrantfile as well. The first
time someone does a
vagrant up, the base box will be downloaded to their machine and stored locally.
- Create or reuse a minimal base box with only the OS and SSH installed. Provision the dependencies for your project through Puppet/Chef. Ideally reuse existing Puppet modules/Chef recipes.
Option 1 is relatively easier and faster for these reasons.
- There are quite a few base boxes that people have already created and shared on the Vagrant Boxes site.
- It is straightforward to use a package manager (apt-get, et .al) to install whatever you need on the base box
- Once your installation is done, you can freeze it on the VM and no one else needs to do it. So less troubleshooting, etc. is likely.
Option 2 was more interesting because we wanted to experiment with Puppet! So thats what we went with. Here is our experiences:
- Puppet uses the concept of Resource Abstraction Layer to abstract away OS level details for resources like: Files, Packages, Services, Users, Executables and so on.
- Puppet provides a language to specify configuration and one can maintain a “manifest” file with per-project requirements.
- Puppet modules are a way to reuse other people’s code to avoid the nitty-gritty of setting things up. There is a PuppetLabs module repository as well as a number of modules available on Github. However, we were not too happy with the documentation and support for most modules that we used.
- We ended up using Puppet modules for RVM and Postgresql
installation. In order to tell Puppet to install RVM, all we needed
to do was:
include 'rvm'in our Puppet manifest. The same was true for Postgresql.
- In conclusion, we realized that Puppet modules are nothing but collections of OS specific installation instructions for various packages and services. It is not too difficult to write your own module that delegates to the OS you are using. And that is probably a better way to go than rely on some obscure module that might not be properly tested/maintained.
Using Vagrant to manage VM’s is a great idea.
However, until I become a Puppet/Chef expert or I there is one on the team, I would go for a fully-loaded base box option it is less time consuming.