Runit services for heavy-duty Rails hosting
===========================================
 
Here's a bunch of scripts for deploying a full production Rails app
that runs under runit.
 
This scheme is currently used to run several large production
Rails apps, including http://www.looktothestars.org/, and it
has workd very nicely for a couple of years now.  More reliable
than system-wide init.d scripts, and fewer permission headaches.
 
 
How to use it
=============
 
Create an unprivileged user for your Rails app, then set up a
system-wide Runit service to manage user-specific services for
that user -- see the following page for details:
 
  http://www.sanityinc.com/articles/init-scripts-considered-harmful
 
All further steps assume that you are working as the unprivileged user.
 
Now, clone this project into the user's home directory:
 
 cd
 git clone git://github.com/purcell/rails-runit.git
 
Symlink your rails app's root directory to this directory with the
name 'app':
 
 ln -s ~appuser/releases/current ~/rails-runit/app
 
Now you can create some services:
 
 cd ~/rails-runit
 ./add-mongrel 4001
 ./add-mongrel 4002
 ./add-mongrel 4003
 ./add-haproxy 4000
 ./add-nginx 4020 4000
 
This creates (potential) runit services under ~/rails-runit/service.
Symlink them to ~/service, and runit will start them for you.
 
Once started, you can set your system-wide web server to proxy requests
for your Rails app to port 4000 (haproxy => mongrel * 3), or to port 4020
(nginx => haproxy => mongrel * 3).
 
Browse to port 3998 (= haproxy port - 2) to see the haproxy status
page.
 
You can then easily restart your app (e.g. as part of a capistrano
deployment) like this:
 
  sv restart ~/service/mongrel-*
 
Further "add-*" and "*-run" scripts are provided for varnish and
ferret_server.
 
 
Status
======
 
The haproxy service assumes that the mongrels will be found on
the 10 ports higher than itself.
 
The use of the 'production' RAILS_ENV is hard-coded, so look out
for that if you try it on your own machine.
 
The entire scheme is subject to change -- see below.
 
 
Future plans
============
 
I'd like to rewrite all this in ruby so we can have templates for the
config files in ERB, and to make the port mapping a bit more flexible.
The 'run' scripts would then look up their config templates under
RAILS_ROOT/config first, and fall back to ~/rails-runit/ if necessary.
 
It'd be nice to automatically generate monitrc fragments that could
be included by a system-wide monit instance for the purposes of
monitoring the app.