Cuoco provides plumbing necessary to run Chef Solo through Capistrano with zero configuration.

• reuse as much server information from Capistrano as possible (including roles);
• comply to Chef directory structure as much as possible to simplify transition;
• bootstrap bare machines with minimal footprint.

Capistrano and Chef Solo, sitting in a tree...

Capistrano is a framework that allows you to run commands in parallel on multiple remote machines. Its primary use is application deployment, but Capistrano can automate any task you can do over SSH. A big advantage of Capistrano is that it is wildly popular among web developers, and thus well supported and rich with plugins.

Another common task you have to do on the remote machines is server provisioning/management, and Capistrano has no facilities for that. It's definitely possible to manage machines using Capistrano, but you have to bring your own scripts.

Chef Solo is a tool that uses well-structured, data-driven Ruby scripts to describe and run configuration management routines. It manages the machine it's running from. So before running Chef Solo on a remote machine, you have to install it there and upload your scripts to that machine. You can do that by hand, you can use one of the many Chef Solo bootstrap scripts, or you can use some tool like littlechef, which needs to be configured to know about your servers.

But wait - not only Capistrano knows where your servers are, it can already run commands on them, and in parallel. It's exactly what Chef Solo is lacking.

Cuoco provides a set of Capistrano tasks to tie in Chef Solo into Capistrano.

gem install cuoco

If you're using bundler you already know the drill.

To make Cuoco tasks available in Capistrano, require this file in your Capfile or config/deploy.rb:

require 'cuoco/capistrano'

Cuoco can turn a completely bare machine into a live server with a single command. You buy servers, you declare them in Capistrano, you run cap deploy:setup deploy. Done! [*]

To do so, require this file in your Capfile or config/deploy.rb.

require 'cuoco/capistrano/automatic'

It hooks up cuoco:update_configuration before deploy:setup. This built-in task, quote, "Prepares one or more servers for deployment.". Which is what Chef does. If you update your server configuration you run cap deploy:setup again, it is non-destructive.

My personal opinion is that running Chef on each deploy is wasteful and forces sudo rights as a requirement for all deployers. If you really want to do that, you'll have to figure out the (extremely tricky) Capistrano directive yourself.

[*] Naturally, you have to write the provisioning scripts first.

There are separate Cuoco tasks defined for more granular control.

cuoco:bootstrap # Installs chef, but does not run anything
cuoco:run       # Runs chef assuming it's already installed

If application deployment doesn't apply in your scenario, you can just run

cap cuoco:update_configuration

directly.

As of September 2, 2012 the pg gem fails to compile on Chef installed with the default bootstrap script.

You can replace the Omnibus installation script with any other, like this one I wrote for Ubuntu 12.04, by setting the :chef_install_command variable:

set :chef_install_command, 'wget -q -O - https://raw.github.com/gist/3601146/install_chef_ubuntu_precise.sh | sudo bash'

Here are the minimal requirements for running Cuoco:

• the server is running one of the operating systems supported by Omnibus (most flavors of Linux will do);
• the server has sudo, bash and either curl or wget installed (most Linux servers do);
• Capistrano can connect to the server;
• the Capistrano user has sudo rights on the server. See my article on setting up user accounts for the suggested approach.

Cuoco works with a traditional Chef directory structure. By default it looks for it in config/chef. You can change that:

# set the root Chef directory 
set :chef_path, 'chef'

Cuoco assigns Chef roles from Capistrano roles.

This means that every server that you will run cuoco:update_configuration on will get its own run list based on the roles it has: if it has :app and :db roles, Chef's run list for that server will be role[app], role[db].

The chef roles directory must contain a role file for every Capistrano role you are using.

You can pass data into chef by declaring a :chef_data variable

set :chef_data, {
  :myapp => {
    :path => deploy_to
  }
}

The hash gets passed to Chef as if it was specified on the command line with chef-solo -j chef_data.json.

This approach is very simplistic and not very DRY, so I'm planning to expand on variables in further releases.

TODO Environments are not supported by Chef Solo. It would be nice to provide the functionality based on Capistrano stages.

"Cuoco" means "cook" in Italian. The original Capistrano is a city in Italy. So a chef working solo in Capistrano would be called un cuoco... get it now?

(If you've been expecting to see Kaley Cuoco here, I'll save you a trip to Google:)

2012 Leonid Shevtsov