FriendFeed's new realtime API

Wednesday, October 22, 2008 by Dave Winer.

Yesterday I got an email from Bret Taylor at FriendFeed saying that their new "realtime" API had been deployed publicly, so I quickly took a look, and found that, as with all other elements of their API, it's simple and easy to support. I immediately wrote glue to connect it to the OPML Editor. 

FriendFeed: Real-time API (beta). 

To understand how it works, consider a hypothetical web app. I live about 1/2 block from a bus that goes to the BART station. Suppose there's a web page that updates when the bus passes a stop about a mile from my house. I'd point my web browser to the page, but it wouldn't refresh right away, but when the bus approached, the page would upload, and flash some kind of message saying "Get out there Dave!" 

That's how FriendFeed's realtime API works. Ask a question and wait for a response. You might wait a long time, minutes, even hours, and that's a good thing. When the event you asked about happens, you can act on it instantaneously. And instead of making thousands of calls asking "Is it done yet?" you make just one. 

Now it may sound like a silver bullet, but like all things in computer design, there's a tradeoff. You have to keep a process running waiting for the answer, and over on the FriendFeed side, they have to keep a process running too. But it's probably a good tradeoff. And the performance is stunning. I tested it with my own FriendFeed account and the script running on my machine detects updates the instant they happen; unlike polling apps I have running against Twitter that sometimes take 2 or 3 minutes to detect a change.  

I have a project that's been waiting for just this functionality, I hope to get to it after the election.  

One more comment -- FriendFeed does a good job with the APIs. When asked, I recommend that other developers just do it the way FriendFeed does. I was able to get something running within a couple of minutes. Once I got that far, I'm pretty well hooked, but only had to put in another hour to complete the project. Considering that the goal of an API is to get developers to hook into your service, this feature, and the performance of the service which is also excellent, is all that it takes to get a chance at uptake. That's why I'll generally put aside other work when there's a new feature in the FriendFeed API.