Your application will grow faster than you can say Jack Robinson. Scalability and extendability should be built into architecture from day one. Often people are taking shortcuts for sakes of simplicity, but many 'simpler' solutions get deeper and deeper into system's core and after some time it's quite difficult to get them out of there.
Modularity often helps to build systems with interchangable and disposable parts. Back in Java days it was cool to use those IoC (Inversion of Controls) and Dependency Injection things. In Ruby world everything is much easier. It's all solved by Duck Typing. Although, it's important to bring the modularity to the next level, and start separating not only low-level-component-related functionality, but also more higher-order concepts: functionality and features.
For example, if you use ElasticSearch for model indexing, it's good to have functionality related to indexing in a separate module, or sometimes, even in a complete new library. Same goes for denormalization. If you want keep your data denormalized in some common manner, it would be good to keep related functionality separate. If you do so, extending, replacing or even throwing these parts away becomes easier.
After some time, you start noticing that denormalizing data in your models touces hundreds of records, indexing takes a bit longer than initially expected, converting 250Mb video from avi to 3gp forces users to wait in front of screen and wonder why the heck page load takes so much time. Having functionality already separated, it's enough to move everything you had to a different app, launch callbacks whenever that functionality required, and handle processing offline, without keeping the user waiting.
Before I even start my long story, I'd recommend you to bookmark and add to your reading list 2 things: http://rubyamqp.info/ (which I'm proud to help maintaining to Michael Klishin) and a book RabbitMQ in Action by Alvaro Videla. You'll see here why I consider these things important. There's no better source of clear and elborate AMQP info than these two.
Let's start with a simplest case, that is often handled in a very limiting way. Handling file uploads. Most of people (including the ones I sometimes look up to, although in a product rather then technical sense) in Ruby world handle their uploads with one of the two very popular solutions: carierwave and paperclip. They both are popular, and maybe even OK to be used in a 5-minute prototype, but I think that it's completely unacceptable to let them handle your uploads and processing in production.
Just recently, while discussing a problem with Bundler requiring to fetch gem index from rubygems.org, someone said "Network transfers in ruby are not exactly known for their speed". Besides that, handling uploads in your main app occupies at least one of your web workers for the time of upload.
If people'd only handle uploads in Ruby, I'd probably keep silence at least for a little bit of time. But image resize, format changes and all other kinds of stuff are handled in a similar manner. That's a bit too much.
The solution I'm going to talk about here was considered to be sic. "a ridiculous overkill for most apps starting out", by Chad Pytel (one of creators of paperclip) here. Let's see the downsides of a paperclip solution, and see what are the upsides of following my advise.
So, downsides of a 'standard' solution (and upsides of an alternative approach):
What I suggest here, includes:
Most likely, you already use Nginx. If you do not, you should, but that's a story for another day. 30 minutes is not that much, so the only thing that's missing is a messaging system. Here I'll cover only RabbitMQ. It's easy to install, it's powerful and blazingly fast, and having more or less complex application will force you implement most of it's features on top of Resque (bad thing) sooner or later.
There's plenty information on the web about how to build nginx with nginx_upload_module, so let's skip that and jump right into the configuration.
Here's some sample nginx configuration:
After setting up Nginx, set up your rails application.
Controller is as small as you usually have it. Upload itself was already handled by Nginx, here you only create a model instance out of whatever it have sent you. It's also trivial to test it now.
We've been handling uploads separately from a main form post. There are many plugins for any front-end framework you may possibly use that help you to achieve that.
Now, moving forward to the most important part: the backend processing. There's a great quote from Alvaro Videla's book on AMQP, which covers exactly the same topic.
For example a web application may require that when a user uploads a new picture, the user own image gallery cache must be cleared and also he should be rewarded with some points for his action. We can have two queues bound to the 'upload-pictures' exchange, one with consumers clearing the cache and the other one for increasing user points. Also from this scenario we can see the advantage of using exchanges, bindings and queues over publishing messages directly to queues. What is such advantage? Let's say that the first requirement of the application was that after a picture was uploaded to the website then the user gallery cache was cleared. Easy enough we can have one queue to accomplish that. But what happens when the product owner comes to us with his super cool new feature of giving awards to users for their actions in the website? If we are sending messages directly to queues, then we have to modify the publisher's code to send message to the new points queue. If we have been using fanout exchanges the only thing that we have to do is to write the code for our new consumer and then declare and bind a new queue to the fanout exchange. As we said earlier, the publisher's code is completely decoupled from the consumer's code allowing us to increase our application functionality with ease.
Let's convert these words to some Ruby code. I'll be using ruby-amqp gem.
We'll have two parts of the application: producer and consumer. Producer will generate and "push" messages to the queue, notifying backend about the incoming upload, and the fact that it should start handling it. Consumer will receive the message, convert it, resize it, send email to the user etc.
For each action you want to be triggered on the event, you have to bind to fanout exchange, declare a queue, giving it a descriptive name, like "images.resize", "images.crop" etc. You will receive payload and metadata published by producer, and run an according action. Here, the performance of your service doesn't matter that much anymore. It's complicated to make it less performing than it was before. Use image-magick directly, use any gem that you feel comfortable with. that functionality is very easy to change afterwards, and you have a complete control over it, too.
So, the producer:
Here you see that we connect to our AMQP server, create a fanout exchange and pubish you message to that exchange, not much more. It's just as easy as it is!
There's nothing scary in AMQP. Installing broker is really easy, writing Quickstart code doesn't even take a minute. Mastering your AMQP-fu will certainly take some time, but you'll discover a feature-rich protocol and will see how easy it is to solve even the most copmlex task using it. The rest of things for solving this particular task are already at your disposal.
If you want some 'takeaway', I'd want it to be "write modular applications" and "do not couple your functionality". Both statements are to be older than I am, and you have already heard a lot about that. But apps become so much easier to develop, maintain and scale by just following these two. There are many ways of doing so, of course, but the concept remains the same: use right tools for right job, and if a little new concept will in the end simplify your solution and bring benefits, you'll have to step over the semblant complexity.
That's only the first article of series. So far I have enough material for about 3-4 publications, but it needs some systematization and, since a topic of 'where to process my shitz' is quite touchy and quite often is a point of discussion, material will be presented in chunks that are easier to discuss.
If you like my content, you might want to follow me on Twitter to subscribe for the future updates!