Oct 2010 revised Nov 2012
Unicorn is an interesting Unix Ruby HTTP server which makes great use of Unix:
Unicorn is an HTTP server for Rack applications designed to only serve fast clients on low-latency, high-bandwidth connections and take advantage of features in Unix/Unix-like kernels.
In this post I’ll describe Unicorn’s design then walk you through setting it up.
Unicorn follows the Unix philosophy:
Do one thing and do it right.
For instance, load balancing in Unicorn is done by the OS kernel and Unicorn’s processes are controlled by Unix signals.
Unicorn’s design is officially described here. I will list some of the things which I consider core for why Unicorn is an interesting alternative.
Load balancing between worker processes is done by the OS kernel. All workers share a common set of listener sockets and does non-blocking accept() on them. The kernel will decide which worker process to give a socket to and workers will sleep if there is nothing to accept().
Load balancers conventionally reverse proxy the request to the worker that is most likely to be ready. This assumption is usually based purely on whenever that worker last served a request. This suffers from two evident disadvantages:
- Some requests take longer to complete (e.g. heavy I/O, slow client)
- Software fails and times out
The common load balancer does not account for this, queueing clients at workers behind slow requests.
Unicorn solves this problem with a pull-model rather than a push-model. All
requests are initially queued at the master on a Unix socket, workers
accept(2) (pull) requests off the queue (shared Unix socket) when they are
ready. Thus requests are always handled by a worker which can handle request
immediately. This solves the problems mentioned above.
Slow clients slow down everything. Twitter has shed some light on this issue in their blog post on why they moved to Unicorn:
Every server has a fixed number of workers that handle incoming requests. During peak hours, we may get more simultaneous requests than available workers. We respond by putting those requests in a queue.
Welcome to Unicorn’s world of evented I/O:
This is unnoticeable to users when the queue is short and we handle requests quickly, but large systems have outliers. Every so often a request will take unusually long, and everyone waiting behind that request suffers. Worse, if an individual worker’s line gets too long, we have to drop requests. You may be presented with an adorable whale just because you landed in the wrong queue at the wrong time.
And then they continue to talk about supermarket queues, read the whole thing.
In the conventional web server using the busyness heuristic to determine where to push the request, you have many short queues at each worker. Easily, a lot of fast requests can end up behind slow requests, because they are distributed essentially randomly, which means your request can timeout simply because you were unlucky enough to end up behind a slow request.
Because of Unicorn’s long queue model, this will not happen. Instead, you will be taken off the long queue quickly and slow requests will fail in isolation.
With Unicorn one can deploy with zero downtime. This is rad stuff:
You can upgrade Unicorn, your entire application, libraries and even your Ruby interpreter without dropping clients.
The Unicorn master and worker processes respond to Unix signals. Here’s what Github does:
First we send the existing Unicorn master a
USR2 SIGNAL. This tells it to begin starting a new master process, reloading all our app code. When the new master is fully loaded it forks all the workers it needs. The first worker forked notices there is still an old master and sends it a QUIT signal.
When the old master receives the QUIT, it starts gracefully shutting down its workers. Once all the workers have finished serving requests, it dies. We now have a fresh version of our app, fully loaded and ready to receive requests, without any downtime: the old and new workers all share the Unix Domain Socket so nginx doesn’t have to even care about the transition.
We can also use this process to upgrade Unicorn itself.
Unicorn’s signal handling is described here. Github has shared their init for Unicorn, which sends the appropriate signals according to the spec for various actions. This makes 100% uptime possible, without any significant speed drop since children are slowly restarted.
Rails on Unicorns
We’re going to set up nginx in front of Unicorn.
Start by installing nginx via your favorite package manager. Afterwards
we need to configure it for Unicorn. We’ll grab the
configuration shipped with Unicorn, the nginx configuration file is
usually located at
/etc/nginx/nginx.conf, so place it there, and tweak it to
your likings, read the comments–they’re quite good.
nginx.conf you may have stumbled upon this line:
While this works, it’s generally adviced to run as a seperate user (which we have more control over than nobody) for security reasons and increased control. We’ll create an nginx user and a web group.
Configure your static path in
/var/www, and change the owner
of that directory to the web group:
Add yourself to the web group to be able to modify the contents of
Now we have nginx running. Install the Unicorn gem:
You should now have Unicorn installed:
unicorn (for non-Rails rack
unicorn_rails (for Rails applications version >= 1.2) should
be in your path.
Time to take it for a spin! (You may wish to re-login with
su - USERNAME if
you haven’t already, this ensures your permission tokens are set, otherwise you
will not have write permission to
There we go, we now have our Unicorn Rails test app in
/var/www! Let’s fetch a
Unicorn config file. We’ll set our starting point in the example configuration
that ships with the Unicorn source:
You will want to tweak a few things to set the right paths:
Then Unicorn is configured!
Start the nginx deamon, this depends on your OS. Then start Unicorn:
-D deamonizes it.
-c specifies the configuration file. In production you
will probably want to pass
-E production as well, to run the app in the
production Rack environment.
That’s it! Visiting localhost should take you to the Rails default page.
You should follow me on Twitter here.