Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

Table of Contents

Quick Answer

REST is a way of producing and consuming web APIs. It tends to be simpler and easier to work with than other API methodologies (such as SOAP). It seeks improved scalability, performance, and client-server independence.

Digging Deeper

REST is an acronym for Representational State Transfer. This Wikipedia article offers a good primer on the subject but for those who want to go right to the source material, read Chapter 5 of Dr. Roy Fielding's dissertation. There are numerous web articles and vlogs that discuss various REST topics. This article will focus on what REST means to the SRP HTTP Framework.

At a simple and high-level, REST is a way of producing and consuming web APIs. In this regard, it is no different from other web API methodologies (generally known as RPC or remote procedure call) such as SOAP or XML-RPC. They all communicate with web servers using URLs and HTTP. Contrary to conventional thinking, REST is not a standard . Rather, but it does point to other standards (e.g., HTTP). REST attempts to describe a philosophy of building web APIs using six different constraints:

...

Client-Server Architecture is assumed when working with web APIs, so we won't explore this. The Cacheability and Layered System constraints are generally handled through intermediary devices between the client and the server as a way of improving performance and reliability, so we won't explore these either. Finally, Code on Demand is an optional constraint that is only useful in environments where the client and the server are tightly controlled. Therefore we'll skip over this as well. This leaves us to explore Statelessness and Uniform Interface.

Statelessness

A stateless system is one where the server is unaware of the state of the client. That is, the server makes no assumptions about what data the client already has or what options are available to the client. In systems where the client state is managed (i.e., a stateful design), this is often handled through session managers, which are server-side systems that track the activity of each client. REST maintains that stateful designs will eventually become over burdened and will hinder scalability.

The SRP HTTP Framework does not enforce statelessness , but it also does not nor does it offer any tools to make the system stateful. Much of this is how the API developer implements the response to a particular requestStatelessness is really a matter of self-governance in the design of the API. Developers can move toward statelessness by avoiding, or minimizing, database locks, and returning the resource with meta-data that instructs . Resources can also be returned along with metadata (see Uniform Interface below) that instruct the client how it can request a state change. Stateless APIs are not difficult to implement unless they share resources (i.e., database tables) with a desktop OpenInsight environment. In these cases, OpenInsight clients typically maintain pessimistic locks on database rows. Stateless APIs typically use optimistic locks (or no locks at all), both of which require a way to resolve conflicts if an OpenInsight client is retaining a lock for an extended period of time.

Uniform Interface

REST, as it is argued, attempts to use HTTP more faithfully. This is the primary basis for a uniform interface. That is, by adhering to the published HTTP standards, API producers and consumers can better anticipate how to interface with each other. It also provides for greater decoupling, allowing independent evolution between the client and the server.

A One key element is that the URL is a reference to a resource on the server rather than a reference to function (or remote procedure) on the server. REST is known for embracing all of the defined documented HTTP methods so clients can convey a wide variety of intent with the resource. For instance:

...

HTTP provides a way of managing metadata through the use of request headers and response headers. Effective use of headers avoids the need to build proprietary messaging within the payload or with URL through query params. A good example of this is the Accept request header. Clients should use the Accept header to indicate their preferred data format (aka media type). Servers should look for the value specified in the Accept header and return the resource in this format if it is able.

HTTP also provides a set of response status codes that inform the client of how the request was handled. For example: 200 means OK, 201 means the a new resource was created, 404 means the URL indicates a non-existing resourceresource does not exist, 405 means the HTTP method is not supported, etc. Servers should send back the most appropriate applicable status code and clients should attempt to handle any valid status code appropriately.

The most significant, albeit the most misunderstood and underutilized, An important aspect of the Uniform Interface , constraint is described as Hypermedia As The Engine Of Application State (, or HATEOAS). We have a dedicated page to the question, "What is HATEOAS?"discuss the nature and value of HATEOAS in our Why is HATEOAS important? article, but we'll provide a simple explanation of it here. HATEOAS is a design feature where information about the state of a resource should be provided to the client through hyperlinks (aka hypermedia). This avoids the need for maintaining state and it avoids the need for the client to assume (or hardcode) how to request a state change to the server. Applying HATEOAS to API responses adds requests are made. Incorporating HATEOAS to the design of web APIs requires a fair amount of extra work and careful designplanning. This often leads to what drives many developers refer to opt out of HATEOAS. This is often referred to as practical REST which is contrasted with pragmatic REST whereas purist REST includes and advocates for HATEOAS. Regardless of the internal debating that continues to go on with this subject, Drongoing internal debates, Dr. Fielding himself has writtenexpressed his feelings rather clearly, "...if the engine of application state (and hence the API) is not being driven by hypertext, then it cannot be RESTful and cannot be a REST API" (emphasis added).

...