Best practices for new Rails deployments on Linux?

Question

I\'ve used straight Mongrel, I\'ve used Mongrel clusters behind Apache, I\'ve looked at Thin, and I\'m becoming very intrigued by Passenger. I\'ve looked at Nginx, too. I\'ve lo

Answer 1

We've been using the old standard nginx -> mongrel stack for the last 18 months, and although it was not trivial to set up the first time around, it's proven flexible, and has dealt with some very high traffic sites for us. Nginx in particular has been absolutely rock solid and fast, and if you can get your app page-caching you can deal with a lot of requests.

Stuck mongrels have been an issue, so we use monit to kill them when they misbehave. Again, it was not totally trivial to set up, but we've used the same process on many many sites at this point.

We haven't played with passenger yet, so perhaps it's easier and more stable, I'll defer to the other responders on that one, all I can say is that there's no reason at all you can't build a solid stack with nginx and mongrel.

Answer 2

This morning, DHH talks about this very topic on his own blog:

But somehow the message of Passenger has been a little slow to sink in. There's already a ton of big sites running off it. Including Shopify, MTV, Geni, Yammer, and we'll be moving over first Ta-da List shortly, then hopefully the rest of the 37signals suite quickly thereafter.

So while there are still reasons to run your own custom multi-tier setup of manually configured pieces, just like there are people shying away from mod_php for their particulars, I think we've finally settled on a default answer. Something that doesn't require you to really think about the first deployment of your Rails application. Something that just works out of the box. Even if that box is a shared host!

http://www.loudthinking.com/posts/30-myth-1-rails-is-hard-to-deploy

Tobias Lütke on the topic of switching Shopify (million requests/day) to Passenger:

All this means that the total amount of memory that is used by Shopify during normal operations went from average of 9GB to an average of 5GB. We evenly distributed the savings amongst more Shopify processes and more memcached space which moved our average response time from 210ms to 130ms while traffic grew 30% in the last few months.

In conclusion: I cannot see any reason to choose a different deployment strategy at this point. Its simple, complete, fast and well documented.

http://blog.leetsoft.com/2008/11/15/passenger

Answer 3

We have switched fron NginX+Mongrel to Passenger.

I fully believe that Passenger is going to be the new standard for rails, despite NginX and Mongrel cluster being endorsed by some very smart people. Recent advances in Passenger have really propelled it forward.

Our current configuration is something like this:

Web servers

• Ubuntu 8.04 LTS
• Phusion Passenger on Apache2
• MRI Ruby 1.8.6 and friends (form apt)
• Ruby Gems 1.3.0 (Installed from source)

Database servers

• Centos 5
• MySQL Cluster (we just switched to this, but it is promising)

Having standardized on the exact linux distro we've been able to write Capitrano recipes to help deployment (slight variations in configuration have been the source of MANY service outages) and otherwise simplify our lives.