Comet/HTTP push with nginx

Oct 1, 2009

One of the most cumbersome problems of implementing some kind of HTTP push a.k.a. Comet functionality is that the client (website) need to be served from the same host and on the same port as the actual push (long-polling or multipart response) mechanism. Now, as we need to maintain a high number of concurrent client connections we can not use traditional server-side applications like PHP or Ruby on Rails. Using PHP for instance would require one PHP process per client connection – the main memory would quickly become saturated and we’ll most likely hit some scary limit of fds and processes the kernel handles without being a sad little kernel.

For a little project I’m involved in I needed to have the best of two worlds and to a cheap price…

Cheap? Yes, as in not expensive. I refrain from running two or more separate hosts exploiting the “iframe domain relaxness” (for instance like Facebook do) because it’s expensive. Hey, I don’t want to spare a cute little IPv4 address just for this thing.

Nginx is a russian workhorse (uhm, no actually it’s a really neat web server by Igor Sysoev) which have become a popular choice for running large websites with high load. And I really like it, both as a user and as a code-pornographer (it’s written in a very pretty C).

Now, let’s not eat the cream off of the cake. First I’d like to tell you what happened before I settled with nginx. The most apparent solution is to…not solve it. Not by yourself I mean. Let someone else do the job – find ready-made software. After a night of Googling and evaluation I’ve read about and tried a few of the most popular choices: Orbited, Meteor, CometD, GlassFish, ErlyComet, and numerous others (see end of this article for a few more links). Too complex and all of them basically needs to be running on their own host in an expensive >2-host setup (exploiting the previously mentioned “iframe domain relaxness”). I also refuse to run Java – call me old and grumpy, I just don’t trust a Java app to stay under 1 GB RAM at 10,000 connections.

It’s fun to write software, so what the hell – I wrote my own little comet (pub/sub on HTTP channels, basically) using my fave async I/O library libevent. I called it cometps. But… it basically suffered from the first problem mentioned with the “big ones” – it wasn’t a real battle-tested web server at the same time, which is a requirement for the little project I’m working on. At least I discovered how neat it is to use YAML for configuring C apps :).

Let’s take a reading break and look at some pretty ASCII art:

      channel {$push_id} 
      ||      ||      ||
      \/      ||      \/
   listener   ||   listener

Then yesterday I stumbled upon NginxHttpPushModule by Leo Ponomarev. This module is basically a chat server and I need a broadcast model, where all subscribers to a “channel” gets all messages published in that channel. So I forked it, rewrote some parts and added broadcasting functionality – voilá! A publish-subscribe server… oh wait, it’s Nginx at the same time! Awesome!

And it works really well.

Never write software without a real-life context. To honour that sentence (I just made up) I wrote a simple Cocoa (OS X) application which grabs an image from my laptops camera, POSTs it to a PHP script which saves it to disk and then the Cocoa app publishes the newly created image’s URL to a NginxHttpPushModule channel. In the other end there is a ridiculously simple HTML client (jQuery and XHR) which listens to the same channel and when receiving (JSON) from the camera app, updates an image. Here’s a video of the thing, running on a live remote server.


I thew in old dirty Internet Explorer 6 just for kicks. Note that 95% of the delay is caused by uploading the photo over my slow connection at home – the publishing and propagation is faster then both Batman and the Phantom.