February 13th, 2008
Craig Ambrose writes about how he came up with a way to receive mail RESTfully in Rails. Basically, he makes the point that the method for processing email described in the Rails wiki is a little scary, as it can lead to the loading of the full Rails stack on every email you want to process. Craig speculates that this method could be problematic in terms of memory usage, and I tend to agree with him. He comes up with a pretty clever hack to basically redirect each incoming email at his actual running web application by grabbing it from postfix and posting it to his site via the use of Ruby’s net/http library. After that, it’s up to the Rails controller to handle the request.
While I like the cleverness of the solution, I can’t help but feel like it’s somehow asking for trouble. As some comments on the post point out, there doesn’t seem to be an elegant way to handle the mail if the web server happens to be down. Another concern is hammering the web server with too many requests in the case of multiple emails or text messages coming in, as each incoming email takes a up an http connection and occupies a mongrel server (or whatever) for however long it is needed (it’s no worse than submitting form data, but I think the concern is that there is potential for more volume).
I’ve been thinking about a way to handle incoming mail myself, and here’s the solution I’ve come up with. I have not had time to implement it, and maybe it’s been done before (please let me know if it has and has been documented).
- Postfix accepts incoming mail and stores the mail as a file somewhere on the server. A simple procmail recipe would probably be useful here to identify where to store incoming mail depending on it’s address, contents, day of the week, or whatever.
- A backgroundrb worker hangs around and processes the mail files on a periodic basis, by reading files in a specified directory one at a time, parsing each message with TMail, and doing any other Rails stuff you need it to do with each message. Remove each file after the message has been processed.
Backgroundrb has a great built-in scheduling system, so there’s no need to think cron in this setup. Backgroundrb will load the whole rails stack once. The workers sit in a separate process. Backgroundrb won’t run itself over: if you set the schedule to process mail every 2 minutes and the mail takes more than 2 minutes to process, backgroundrb will queue it’s jobs up. In this scenario, a worker is just looking in one directory, so once all the mail is processed, it gets deleted. If no new mail comes in by the next time the worker looks at the directory, it can see there’s nothing to do and go back to sleep. Knowing the expected volume of email coming in is important, but I’m willing to bet in most cases you could ask backgroundrb to run your mail-processing worker once a minute.
This method doesn’t get to claim the label RESTful, but to be honest I don’t think regurgitating incoming mail out to your web server is really in the spirit of “REST” either. If you stay true to the skinny controller, fat model mantra, handling email via model methods (all accessible via backgroundrb) or even module/other non-model class methods in a backgroundrb worker should be as easy as (if not easier than) doing so from within a controller.