A question we commonly get at EnterpriseWeb is whether our platform follows REST or not. Representational State Transfer (REST) is an architectural style for distributed hypermedia systems such as the World Wide Web, and is perhaps best known for providing a lightweight, uniform Web-style application programming interface (API) to server-based resources. On the one hand, EnterpriseWeb can both consume and expose any type of interface, including tightly coupled APIs, Web Services, as well as RESTful APIs, and the platform has no requirement that customers must build distributed hypermedia systems. It would be easy to conclude, therefore, that while EnterpriseWeb supports REST, it is not truly RESTful.
Such a conclusion, however, would neglect the broader architectural context for EnterpriseWeb. The platform builds on top of and extends REST as the foundation for the dynamic, enterprise-class architectural style we call Agent-Oriented Architecture (AOA). EnterpriseWeb’s intelligent agent, SmartAlex, leverages RESTful constraints as part of the core functionality of the EnterpriseWeb platform. The resulting AOA pattern essentially reinvents application functionality and enterprise integration, heralding a new paradigm for distributed computing.
The Limitations of REST
One of the primary challenges to the successful application of REST is understanding how to extend REST to distributed hypermedia systems in general, beyond the straightforward interactions between browsers and Web servers. To help clarify this point, Figure 1 below illustrates a simple RESTful architecture. In this example, the client is a browser, and it sends GETs and PUTs or other RESTful queries to URIs that resolve to resources on a server, which responds by sending the appropriate representation back to the client. In addition, REST allows for a cache intermediating between client and server that might resolve queries on behalf of the server for scalability purposes.
As an architectural style, however, the point of REST isn’t the uniform interface that the HTTP verbs enable. REST is really about hypermedia, where hypermedia are the engine of application state – the HATEOAS constraint essential to building hypermedia systems. In figure 1, we’re representing HATEOAS by the interactions between human users and their browsers as people click links on Web pages, thus advancing the application state. The RESTful client (in other words, the browser) maintains application state for each user by showing them the Web page (or other representation) they requested when they followed a given hyperlink.
However, software clients that do not necessarily have user interfaces may be problematic for REST, but they are a familiar part of the Service-Oriented Architecture (SOA) architectural style, where we call such clients Service consumers. Combining REST and SOA into the combined architectural style we call REST-Based SOA introduces the notion of an intermediary that presents a Service endpoint and resolves interactions with that endpoint into underlying interactions with various legacy systems. The SOA intermediary in this case exposes RESTful endpoints as URIs that accept GETs, PUTs, etc. from Service consumers, which can be any software client. See figure 2 below for an illustration of the REST-Based SOA pattern.
Note that adding SOA to REST augments the role of the intermediary. Pure REST allows for simple caching and proxy behavior, while SOA calls for policy-based routing and transformation operations that provide the Service abstraction. SOA also reinforces the notion that the Service consumer can be any piece of software, regardless of whether it has a user interface.
Even with REST-based SOA, however, we still have problems with implementing HATEOAS: coding our clients so that they are able to gather the metadata they need by following hyperlinks. In other words, how do we apply REST to any hypermedia system, where instead of a browser we have any piece of software as a client? How do we code the software client to know how to follow hyperlinks, where it doesn’t know what the hyperlinks are ahead of time or what representations they’re supposed to interact with? Humans simply click hyperlinks until they get the representation they want, even if they don’t know beforehand how to find it. How do we teach software to automate this process and gather all the metadata it needs by following a sequence of hyperlinks?
Introducing Agent-Oriented Architecture
The answer to these questions is to cast an intelligent agent in the role of SOA intermediary in the REST-Based SOA pattern in Figure 2. Intelligent software agents (or simply intelligent agents when we know we’re talking about software) are autonomous programs that have the authority to determine what action is appropriate based upon the requests made of them. In this new, Agent-Oriented architectural pattern, the agent interacts with any resource as a RESTful client, where the agent must be able to automatically follow hyperlinks to gather all the information it requires in order to respond appropriately to any request from the client.
In other words, when following this newly coined AOA architectural style, software clients do not have to comply with HATEOAS (they may, but such compliance is optional). Instead, the agent alone must follow the HATEOAS constraint as it interacts with resources. To achieve this behavior, we must underspecify the intelligent agent. In other words, the agent can’t know ahead of time what it’s supposed to do to respond to any particular request. Instead, it must be able to process any request on demand by fetching related resources that provide the appropriate metadata, data, or code it needs to properly respond to that request with a custom response, for each interaction in real time. Figure 3 below illustrates the basic AOA pattern.
For each request from any client, regardless of whether it has a user interface, the agent constructs a custom response based on latest and most relevant information available. In fact, requests to the agent can come from anywhere (i.e., they follow an event-driven pattern). The agent’s underspecification means that it doesn’t know ahead of time what behavior it must exhibit, but it does know how to find the information it needs in order to determine that behavior – and it does that by following hyperlinks, as per HATEOAS. In other words, the goal-oriented agent resolves URIs recursively in order to gather and execute the information it needs – a particularly concise example of fully automated HATEOAS in action.
The Benefits of AOA
An earlier Loosely-Coupled newsletter explained that if you follow REST, you’re unable to accept out-of-band metadata or business context outside of the hypermedia. Agent-Oriented Architecture, however, solves these problems, because the agent is free to fetch whatever it needs to complete the request, since it treats all entities – metadata, data, code, etc. – as resources. In other words, the agent serves as a RESTful client, even when the software client does not. What was out-of-band for REST isn’t out-of-band for AOA. Everything is on the table.
The true power of AOA, though, lies in how it resolves the fundamental challenge of static APIs. Whether they be Web Services, RESTful APIs, or some other type of loosely-coupled interface, every approach to software integration today suffers from the fact that interactions tend to break when API contract metadata change.
By adding an intelligent agent to the mix, we’re able to resolve differences in interaction context between disparate software endpoints dynamically and in real time. Far more than a traditional broker, which must rely on static transformation logic to resolve endpoint differences, the agent must be able to interpret metadata, as well as policies, rules, and the underlying data themselves to create real time interactions that maintain the business context – an example of dynamic coupling, a central principle to AOA.
Dynamic coupling, therefore, represents a paradigm shift in how to build and utilize APIs. Up to this point in time, the focus of both SOA and REST has been on building loosely-coupled interfaces: static, contracted interfaces specified by WSDL and various policy metadata when those interfaces are Web Services, or Internet Media Types and related metadata for RESTful interactions. Neither approach deals well with change. AOA, in contrast, relies upon dynamic coupling that responds automatically to change, since the agent interprets current metadata for every interaction in real time.
Icons by http://dryicons.com