Deploying apps with Puppet

Use Puppet master/agent

I want to deploy a simple application that installs a file in /tmp on a single box.

Here is the puppet module definition:
/etc/puppetlabs/puppet/modules/myapp/manifests/init.pp:
class myapp {
file { 'testfile' : path => '/tmp/testfile-local', ensure => 'present', content => 'Test', mode => 0640}
}

Typically, the puppet master uses a site.pp file for node definitions:
/etc/puppet/manifests/site.pp:

node development {
include "myapp"
}

node staging {
include "myapp"
}

Start the puppet master:
puppet master --no-daemonize --verbose
notice: Starting puppet master version 2.6.4

This is telling puppet master that myapp needs to be installed on the development box. If I have a puppet agent running on the client (which in this case happens to be the same box), then it will apply the latest configuration from the server automatically when it polls the next time.

I could run the agent manually onetime by ssh'ing into the box and invoking the agent:
[root@learn ~]# puppet agent --no-daemonize --onetime --server puppet --verbose
info: Retrieving plugin
info: Caching catalog for puppet
info: Applying configuration version '1310559260'
notice: /Stage[main]/Myapp/File[testfile-local]/ensure: created
notice: Finished catalog run in 0.02 seconds

CONS:
The main problem here is that I am not sure how you to achieve this via puppet master itself .. have not checked the UI - but surely there must be a way to update specific nodes - i.e. dev nodes only ?

You can always automate this using a capistrano script:
set :user, "root"
task :deploy
role :app, ‘development’
run 'puppet agent --no-daemonize --onetime --server puppet --verbose'
end

The problem with using a script like this is that the environment information has to be maintained in two places – in the script and the site.pp file. One way to avoid this is to generate this script from the site.pp file.

PROS:
You are using Puppet to install the application, just like a sysadmin would to ensure that machines were setup correctly. Easy to sell to ops.

Server-less puppet

Use capistrano to run puppet “apply” on the relevant nodes. Here is a Capfile for the app. It assumes that you have manifests checked out on the clients.

set :user, "root"

task :development do
role :app, “development”
end

task :staging do
role :app, "staging"
end

task :deploy do
# TODO: checkout manifests to module path
run "puppet apply -e \"include myapp\""
end

To deploy to development environment, you would run cap for that environment:

rg6977:puppet Thoughtworks$ cap development deploy
* executing `development'
* executing `deploy'
* executing "puppet apply -e \"include myapp\""
servers: ["192.168.56.101"]
Password:
[192.168.56.101] executing command
command finished

PROS:
Node definitions are now in Capistrano and not in puppet. Puppet is used only to install the application in a given environment. Puppet does a poor job of managing environment specific information. See http://docs.puppetlabs.com/guides/environment.html.

CONS:

Node definitions are now in Capistrano and not in puppet. Puppet is used only to install the application in a given environment.
This is more code than the previous approach. Puppet “apply” will only apply manifests to the local machine. So, if your application must be installed on a web server and db server, your Capistrano script needs to invoke puppet apply on the appropriate manifest (db vs web).