Toss your Cookies: Maintaining State on the Client with REST

REST quiz of the day: which is more important when following REST: updating the resource state or updating state on the client? Most developers are likely to say that REST focuses on updating resource state on the server. After all, POST, PUT, and DELETE all do so. GET is the only HTTP verb that fetches resource state without changing it.

If you’ve been reading ZapThink’s discussion on REST, you probably realize that those developers are incorrect. Updating resource state is clearly an important part of REST, but updating state on the client is even more important. Why? Because client state is the foundation for distributed hypermedia applications. And after all, such applications are the point of REST.

In fact, the RESTful world distinguishes between resource state and application state, which is the state information the client maintains. And since hypermedia are the engine of application state, it makes sense that application state is more important to REST than resource state. After all, REST is representational state transfer. The representations are what the resources send to the clients, including application state information that belongs on the client.

HATEOAS and the Cloud

The topic of application state came up in an interesting discussion during our recent Cloud Computing for Architects course in San Diego. I was explaining how it’s important to maintain a stateless application tier in the Cloud, because we want to put the components on that tier (Web servers, application servers, ESBs, etc.) onto virtual machine (VM) instances. In order to achieve the elasticity and resilience of the Cloud, then, we can’t afford to maintain state information on such instances. We always have the option of persisting such state information, making it a part of the resource state, but relying too heavily on our resource state limits our scalability. The clear alternative is to transfer state information to the client, and let hypermedia be the engine of application state (otherwise known as the dreaded HATEOAS REST constraint).

One of the attendees in the class was confused by this discussion. He pointed out that if we use the client (say, a browser) to maintain state in a stateful application like an eCommerce shopping cart, then such state information must either go into hidden form fields or into the URL, so that the server can pass it along to the browser from one request to the next. But if the user is clicking a link rather than submitting a form, then they are executing a GET, and with a GET, the only place to put state information from the client to the server is in the URL. And we all know that URLs have a maximum length. What do we do, he asked, if we have too much state information for the URL? For example, we might have dozens of items in our cart. Was I suggesting passing the entire contents of each cart – product descriptions, prices, etc. – in the URL?

The answer, of course, is no. I say “of course” because such a question would be silly for anyone who truly understands REST. But I must admit, it took me a while to think through my response. The problem was that I have a background as a Web developer, and my student may have also built his share of Web sites back in the day as well. And back in the 1990s, before REST, in the early days of HTML and JavaScript, maintaining state in a browser was problematic. All we had were cookies and the aforementioned hidden form fields and URL query strings. And since people can turn their cookies off, we never wanted to rely on them. So yes, back in the day, if the user is clicking a link (i.e., performing a GET), the URL was all you had to work with.

With REST, however, we’re working with an abstracted client. It need not be a browser, and in fact, it need not have a user interface at all. A RESTful client may serve as an intermediary, for example. Even when the client has a UI, it could be any type of application. For example, most mobile apps are written natively to the mobile environment (iPhone or Android, for the most part), and will continue to be at least until HTML5 is fully baked.

Even when the client is a browser, however, we have numerous ways of maintaining application state. Each approach, as you might expect, has its strengths and weaknesses:

  • Cookies – long a part of the HTTP protocol, cookies are universally supported and almost as universally reviled. We love them for enabling the “remember me” feature on Web sites with logins we visit frequently, and we hate them for enabling advertisers to track our browsing habits. Few app developers would rely on them for much else.

  • Hidden form fields – every Web developer’s secret sauce. You can put whatever you want into such fields, and as long as the user submits the corresponding form, the contents of hidden fields go along for the ride. The problem is that hidden form fields only work with POST. In REST, POST is only for initializing a subsidiary resource, so if that’s not what you’re doing, then you don’t want to POST. The other downside to hidden form fields is that application state information must always make a round trip to the server and back again, whether the server needs to do anything with it or not.

  • Frames – Frames made their debut in 1996 in Netscape Navigator 2.0, the same browser that introduced JavaScript to the world. Iframes came soon after. Both types of frames allowed new pages in the frame while maintaining state information in JavaScript variables in the enclosing page. In either case, however, updating the content of a frame doesn’t change the URL in the browser’s location field. As a result, reloading or following a bookmarked page takes you back to square one, deleting all state information.

  • JavaScript-only updates – More recent advances to the Document Object Model (DOM), in particular the innerHTML property, allow JavaScript to update any <div> element after the page is loaded. And since JavaScript can also perform all manner of RESTful calls, it’s possible to make it appear that any or all of a page is changing, even though the page itself isn’t actually reloading. As with frames, JavaScript variables can store state information, but also similar to frames, the URL the user sees doesn’t change when your script interacts with the server.

  • Signed scripts – What about simply writing arbitrary content to the user’s hard drive? Browser publishers have been monkeying with this capability since signed scripts debuted in the 1990s. The problem with giving the browser write privileges, of course, is security. One flaw and hackers can easily take over your computer. Needless to say, this capability never took off, although plugins like Adobe Flash can allow the ability to write to the users’ hard drive.

  • DOM storage – Today we have DOM storage. Think cookies on steroids. Every modern browser can essentially store a JSON object that persists as the browser loads different pages. You have the option of maintaining such information for a browser session (you lose it when you quit the browser) or persisting it across browser sessions (where the browser writes the data to a file). This capability also enables developers to write browser apps that can function properly when the user is offline. The downside to DOM storage is that it only works in newer browsers – although Firefox, Internet Explorer, Chrome, and Safari all support it.

  • URL query string – finally, let’s discuss storing state information in the URL. Yes, there’s a character limit – the HTTP spec recommends sticking to 255 characters or less, although most browsers and Web servers support much longer URLs. So, how much can you cram into 255 characters? More than you might expect, if you use a tool like RISON for compacting JSON to squeeze more of it into each URL. Don’t like RISON? There are alternatives available or you can create your own approach. Even with such techniques, however, there is still a size limit, and all such information must make the round trip to server and back.

The ZapThink Take

Based on the discussion above, you should have no more concerns about storing application state on the client. There are always tradeoffs, but one of the scenarios above should handle virtually every application state issue you’re likely to come up with. Feel free to transfer application state to the client and rest assured you’re following REST.

That is, of course, if you really are following REST, which means that you’re building a hypermedia application. And while POST, PUT, and DELETE update resource state for hypermedia applications, every representation from resource back to client updates client state. Even a GET, which never changes resource state, still changes the application state. In other words, clicking a link or submitting a form loads a new page. Of course REST behaves that way.

While this article focused more on maintaining state on the client, therefore, REST is more concerned with updating state on the client. The real point here is that we have the luxury of choosing to maintain the state information we require while running an application whose state is supposed to change. Either way, hypermedia are the engine of application state.

Image source: bloggyboulga