I want to deploy a single codebase to two different Heroku apps.  I'd like
one of the apps to launch an API service, which is a small Rack app defined
in the same codebase of our much larger web app that uses Rails.

Ideally, I'd like to set something on the api app to use a different app
root (and hence different Procfile) for launching the  API.  This way I can
maintain one codebase, but a separate, lightweight API service that still
has access to things like models etc...

Is this possible?  I can't find anything in the docs.

My only other thought was to define a config variable that is the service
to actually launch, then use that in the Procfile<https://gist.github.com/1752849>,
but this seems a bit messy and possibly error prone.  It would be much
nicer if I could specify which Procfile to use or change the root directory
of the app to launch on a per-app basis.

I know the Foremen gem allows you to specify different Procfiles and app
roots <http://ddollar.github.com/foreman/#OPTIONS>, but does Heroku?

From what I know of Procfiles, it's routine and reliable to have environment variables available to your Procfile, so I'm not sure that the gist approach you link to is as messy and error-prone as you might think.

If you were to abstract it one layer down so that your rack processes pickup which application they're meant to be running, you're still essentially doing the same thing, but in a different place.

To be honest, there's probably a billion ways of doing this, all of which have some sort of pain.  The approach I've taken in the past is either to have two deployable branches which go to different applications on Heroku, or do abstract the common code out and share it with two deployable applications.

However, saying this, stipulating the web command in your environment seems a fairly reasonable way of tackling this.

-Neil

So potential errors aren't the only concern.  If my API uses different
workers/process etc. from the App it becomes a bit onerous to define all
these different process types using config vars...

Definitely don't want to split my code out (yet) if I can get away with it.
 I wonder if anyone at Heroku can comment on this?  Presumably they have an
internal mechanism for finding the Procfile, it would be great if this was
configurable.

Sure, but remember that Heroku won't scale any of the other processes unless you ask it to.  You can easily declare all the other processes in the Procfile and scale them as needs be for each deployment of the application (so the names would need to be unique), only your web process would need to be configurable.

For instance

web:  bundle exec $APP
worker_app1: bundle exec blah
worker_app2: bundle exec blah blah

etc

If there is it isn't something that they mention in the docs ("Process types are declared via a file named Procfile placed in the root of your app.") nor is it something that you really have control with with custom build packs (unless you want to get /really/ fruity).

-Neil

ya I pretty much guessed all of that, just thought I'd see if I could get a
definitive answer from Heroku.

Thanks for the help!

I'm not sure about the different app roots. But the other approach
definitely works. My (working) example:

$ cat config.ru
$LOAD_PATH.unshift File.dirname(__FILE__)

if ENV['APP_NAME'] == 'web'
  require 'web'
  run Foo::Web
else
  require 'api'
  run Foo::API
end

$ cat Procfile
web: bundle exec rackup -p $PORT
worker: bundle exec rake queue:work VERBOSE=1 QUEUE=*

--
Jason Dixon
DixonGroup Consulting
http://www.dixongroup.net/