EDIT 2007-06-05: An updated version of this is up here…
After my hiatus on posting it seems appropriate to get back into the meaty stuff…
Ever since I started using Debian (instead of FreeBSD) I’ve been having weird problems with my rails dispatch.fcgi processes multiplying in the night. Nothing shows up in the logs, but I spawn 2 externally of Lighttpd, and in the morning i’ve got 6 of the little blighters. Of course, at 20-30 meg a pop, the poor little Xen VPS isn’t too happy about that, so I have a nightly job that kills them all and restarts them. And then, 12 hours later, 4 more than I asked for are there. It seems to be a load issue, but I digress.
So ever since I noticed that I’ve been wanting to use Mongrel to run my Rails apps. For those not familiar with mongrel, it’s a Tomcat-style application host for rails apps that avoids (huzzah) the FCGI palaver that we ordinarily have to deal with. The problem was that according to mongrel, a USR2 signal should fully restart the daemon, and it does, but it’s a tiny little bit funny in a way that makes it totally painful to use with Capistrano for automated deployment.
You see, on restart, it doesn’t re-evaluate what the ./current symlink is pointing to. So when it’s restarted over a ./current that points to ./releases/2006xxxxxxxxx1 that’s fine, but on a redeployment through Cap when this symlink is repointed to ./releases/2006xxxxxxxxx2, the mongrel instance still points to ./releases/2006xxxxxxxxx1
So that’s not so good. Btw, Zed, you’re an awesome coder and I in no way mean you disrespect here. I’m just telling the problem I faced.
Along comes mongrel_cluster. Totally fixed the issue for me, and here’s how.
1. Install mongrel_cluster, and then (in ./current) run “mongrel_rails cluster::configure”
2. open the new ./config/mongrel_cluster.yml and edit the line that starts with “cwd:” like so:
- change “cwd: /path/to/app/releases/2006xxxxxxxxx1” to “cwd: /path/to/app/current”
- change the port to the first one you want for this cluster, and select how many you want (defauts to 2 which should be ‘enough’ for most cases)
- change from “development” to “production”
3. Add the following tasks to your capistrano deploy.rb file
desc “The spinner task is used by :cold_deploy to start the application up”
task :spinner, :roles => :app do
send(run_method, “cd #{deploy_to}/#{current_dir} && mongrel_rails cluster::start”)
end
desc “Restart the mongrel cluster”
task :restart, :roles => :app do
send(run_method, “cd #{deploy_to}/#{current_dir} && mongrel_rails cluster::restart”)
end
4. Rejoice!
5. Before you do anything else, add to your crontab an @restart task to start the mongrel instances when your server comes up! Very important!
Now “rake remote:cold_deploy” and “rake remote:deploy” work for the mongrel cluster! I used the instructions on the mongrel site on how to integrate with lighttpd at http://mongrel.rubyforge.org/docs/lighttpd.html but there are equally good Apache 2.2 docs out there (why Apache 2.2 isn’t available for Debian is anoher question entirely…)
Much kudos and thanks to Zed Shaw for the excellent mongrel server and to Bradley Taylor for mongrel_cluster. Oh, and Jamis Buck for Capistrano!