Re: chef + vlad/capistrano?

[Adam Jacob <adam@opscode.com>]

On Mon, Mar 9, 2009 at 5:53 AM, Michal Frackowiak <michalf@wikidot.com> wrote:
> 1. New instances (nodes) register to chef-server first and get their
> configuration from there, depending on their type (app, web, worker, db
> etc). Role is passed as an attribute. We can launch new instances by
> Capistrano tasks.
> 2. Capistrano pulls list of :app, :web, :db servers dynamically from
> chef-server (or another registration server, iClassify?) through the role
> defer option.

Something like this should work (this is totally untested)

require 'chef'
require 'chef/node'
require 'chef/rest'

# A registration for your API access
api_user = "monkey"
# A password for the API access
api_pass = "pants"
# A validation token, if you use them
validation_token = nil
# The chef server URL
chef_server_url = "http://chefserver.example.com";
# The chef sever openid URL
chef_server_openid = "http://chefserver.example.com:4001";

##
# This should move up into Chef::REST, see CHEF-168
# When it does, you'll do rest.search(...)
##
def search(type, query, attributes=[], &block)
  Chef::Log.debug("Searching #{type} index with #{query}")
  r = Chef::REST.new(chef_server_url)
  results = r.get_rest("search/#{type}?q=#{query}&a=#{attributes.join(',')}")
  Chef::Log.debug("Searching #{type} index with #{query} returned
#{results.length} entries")
  if block
    results.each do |sr|
      block.call(sr)
    end
  else
    results
  end
end

rest = Chef::REST.new('http://chefserver.example.com')

rest.register(api_user, api_pass, validation_token)
rest.authenticate(api_user, api_pass)

search(:node, 'environment:production') do |node|
  role node[:role].to_sym, node[:fqdn]
end

This assumes each system has one role - if you want to get crazier, go
ahead and check to see if role is an array.

> 3. We can run cap tasks on the cluster.

Yep.

> Obviously we want to add :app, :web or :worker on-the-fly, possibly with one
> or two commands, and later move to auto-scaling solutions.

This is how we've done things in the past.

> I am not sure if this is an optimal design, and I am sure you already have
> an idea how Chef and Capistrano would eventually work. Any advice, ideo or a
> hint would be highly appreciated.

I think you are on the right track - let us know how it works!   I
would love to see this sort of thing broken in to a new project on
Github, at least as a loadable Chef library.  If we can provide any
more assistance with using the Chef REST API to make this work, let us
know.

Adam

-- 
Opscode, Inc.
Adam Jacob, CTO
T: (206) 508-4759 E: adam@opscode.com