Most Read This Week
Enterprise Cloud Computing
Bringing Intelligence to REST
The API Economy has nearly run its course.
By: Jason Bloomberg
Feb. 22, 2014 04:00 PM
Whether you're a Cloud Computing aficionado, an enterprise integration specialist, or an IT executive, it's hard to have a conversation today without mention of REST. Representational State Transfer, or REST to the cognoscenti, is an architectural style that treats distributed computing problems as though they were Web problems. On the Web you have browsers chatting via HTTP to Web servers, and those servers can work whatever magic they need to in order to serve up the full wealth we've all come to expect from the World Wide Web. Take those basic Web patterns, extend them to general distributed computing problems (including Cloud and legacy integration), and voila! You have REST.
However, while REST has achieved substantial success in simplifying software interfaces and thus facilitating many forms of integration, it is still inherently inflexible. What works well for humans using browsers often doesn't apply to arbitrary software clients. And most fundamentally, REST does not address the most difficult distributed computing challenge of all: how to deal with dynamic business context.
Freeing Ourselves from REST's Four Constraints
But there is more to REST than a uniform interface. In fact, REST isn't an API style at all. It's an architectural style. As an architectural style, REST consists of a set of constraints on software architecture. In other words, feel free to follow what architectural rules you like, but if you want to follow REST you must comply with the following RESTful constraints:
Let's take a quick tour of these constraints to put them in plain language. In so doing, we'll also why REST fails to adequately address the problem of dynamic business context.
The first constraint is essentially the encapsulation requirement. Resources are abstractions of capabilities on a server, while the representations are what the resources provide to the client. For example, a resource might be a php script running on a Web server, and the representation might be an HTML file it returns when a browser makes a GET request of it. But the browser never, ever gets the php itself; it only sees the HTML. The php is forever hidden from view.
The second constraint calls for the uniform interface. The only way that clients are able to interact with resources is by following hyperlinks in representations - in other words, making GET, POST, PUT, or DELETE requests to the URI of the resource, assuming we're using HTTP as our transport protocol, which we usually are.
The third, self-descriptive message constraint is actually quite straightforward: all the data as well as all the metadata the resource needs to process a request must be contained in that request, and correspondingly, the resource must send all necessary metadata in the representation response to the client that the client will need to understand the representation. In other words, REST requires that there be no out-of-band metadata: information pertinent to the interaction that doesn't actually appear in the interaction. Furthermore, the interaction must be stateless: the resource isn't expected to keep track of any information pertinent to any particular client.
The problem with this third constraint, of course, is that out-of-band metadata is very handy in many situations. Take security-related metadata, for example. REST calls for all such metadata to be in every request, which led to the development of the OAuth (Open Authorization) standard. Yes, OAuth is quite powerful and Web friendly. Yes, OAuth is making inroads into the enterprise. But do you really want to restrict the security protocols for all of your interactions to OAuth and nothing but OAuth? Probably not.
If you have a more complex interaction than a simple request, then the ban on out-of-band metadata becomes increasingly impractical. For example, let's say you're trying to support a complex business process by building a composite application. You're trying to follow REST so you're composing resources. But then you find you need to somehow deal with a range of policies, business rules, or other out-of-band metadata that impact the behavior of your composite application for certain users but not others. REST alone simply doesn't deal well with such complexities.
And then there's the fourth constraint: the dreaded HATEOAS. REST separates state information into two types: resource state and application state. Resource state is shared or persisted state information on the server, while application state is specific to the individual client, who negotiates the application (think abstracted Web site) by following hyperlinks. The HATEOAS constraint hammers home the fact that the point of REST is building distributed hypermedia systems, where the client is responsible for running hypermedia-based applications. In other words, the hypermedia contain the business context for the interactions between client and server.
Hypermedia drive the Web, of course, but once you start breaking down REST's notions of client and server, however, then the power of hypermedia starts to wane. After all, enterprises often want to build or leverage business applications that offer more than a simple Web site, especially when there's a shared business context across nodes, where those nodes are more than just clients and servers. Hypermedia - and REST - simply weren't built for such complex, dynamic situations.
The Devil in the Details
The challenges in the section above, however, go well beyond REST. The problems we're discussing have plagued software interfaces in general, from the earliest screen-scraping programs to object-oriented APIs to Web Services to today's RESTful APIs. The entire notion of a software interface is an agreement between the people building the software provider and consumer endpoints that the interface behaves a particular way. Loose coupling, after all, relies upon an interface contract that fixes the behavior of the API so that the parties involved can make various decisions about their software under the covers without breaking the interaction. But woe to those who dare to change the contract, or who want to consider metadata the contract knows nothing about!
Out-of-band metadata and business context outside hypermedia applications are by definition exterior to the contract, and thus aren't amenable to any distributed computing architectural style that relies too heavily on static APIs. Therein lies the essential challenge of the API. To those analysts trumpeting the API Economy I say: the API Economy has nearly run its course. We've solved as many problems as we're going to solve with contracted software interfaces. But the business stakeholders still aren't happy. After all, it's their context - the business context - that APIs (whether RESTful or not) are so woefully unable to deal with. It's time for another approach.
The EnterpriseWeb Take
What we really need is an entirely different kind of intermediary. A smarter intermediary that knows how to deal with all types of metadata and furthermore, can resolve the more difficult challenge of business context - in real time, where the business lives. In the next issue of Loosely-Coupled I'll discuss how such a smarter intermediary might actually work. And naturally, if you want to see one in action, drop us a line.
Image credit: lin440315
Subscribe to the World's Most Powerful Newsletters
Today's Top Reads