If you're using Puppet you already have a list of the hosts and types of hosts under your control. So why duplicate this information in your capistrano configuration? Every time you provision or destroy a new machine you have to remember to update your deployment configuration and that's just an error waiting to happen.

This requires you to be running my web-puppet service to get at the relevant host and tag information. https://github.com/garethr/web-puppet

A sample capistrano configuration might look like the following. The only important parts are the require, creating a new server instance and the calls to get_servers to return a list of hosts.

require 'capistrano-puppet'

web_puppet = CapistranoPuppet::Server.new('http://username:password@localhost:9295')

set :application, "test"
set :repository, "."
set :scm, :none
set :deploy_via, :copy

default_run_options[:pty] = true
set :deploy_to, '/server/path/'
set :user, 'deploy'
set :use_sudo, false

role :web do
  web_puppet.get_servers('webserver')
end

role :app do
  web_puppet.get_servers('appserver')
end

Note that this is the first release as is really just a proof of concept. Running web-puppet in environments with lots of hosts will likely be very slow.

The code was inspired by capistrano-cobbler, which uses Cobbler rather than web-puppet to get the host information. https://github.com/proffalken/capistrano-cobbler