I am sure this will be a simple question but I want to make sure I use the correct terminology for a meeting explaining my new API.
So if I have the following URIs for my API:
https://testapi.org/api/registration - POST defined
https://testapi.org/api/token - GET defined
https://testapi.org/api/pricedata - GET and POST defined
So these are 3 controllers in the same Web API. In conversation is this considered 1 endpoint or 3 endpoints?
In conversation is this considered 1 endpoint or 3 endpoints?
The REST answer is "0 endpoints".
There is no such thing as a REST endpoint -- Fielding, 2018
What you show here are three resource identifiers, and by implication three resources.
The fact that those three resources each have a separate "controller" is an accident of your implementation. REST really doesn't care.
In a context like OpenAPI 3, you will also see this described as four "operations" (path + method).
(You'll also see that the OpenAPI 3 documentation seems to consider "resource" and "endpoints" to be equivalent.)
Given Hohpe's description of the MessageEndpoint pattern, I'm inclined to think that resource isn't a particularly good match for "endpoint".
You might be able to argue that there is a message-endpoint in your web server process, which is handing off messages to your controllers? In which case, that would be one endpoint.
Are you saying I should have combined up some of the verbs into less controllers?
No, I'm not saying that. REST says that we need a client-stateless-server protocol with uniformly understood self descriptive messages. The HTTP specifications define self descriptive request messages and self descriptive response messages.
But the magic black box that decides what response to send to a request is yours. The number of controllers you decide to use is an implementation detail that is hidden from clients behind the HTTP facade.
There's nothing in REST, or HTTP, that requires/forbids that the server implementation use MVC.
Well, the responsibility of each one are different, it doesn't matter that they are acting in the same resource, so, for practice, all of that are different endpoint, you could put the 1 and 2 in the same one, but the 3 need to be a different product.
Related
In a Hypermedia-Driven RESTful solution, should I include relevant HATEOAS links in Error responses?
For instance, let's imagine we have a Microservices architecture.
We have FooSvc and BarSvc, where Bar resources are associated in a 1:1 relation with Foo resources, but managed independently (e.g. we have Projects that we can create just to present them, and then we can decide to Manage a project in our platform).
So, we can try to fetch the Bar resource associated with Foo with id 1 using GET /foo/1/bar (not caring how the request is routed downstream to the BarSvc).
Since there is no such Bar instance created yet, the service will retrieve a 404 response. Should that response include hypermedia links to point out how this can be created (e.g. a HAL _links field with a link to POST /foo/1/bar)?
Definitely, in fact, RFC7807 - 'Problem Details for HTTP APIs' can help with this, and the most popular frameworks should provide support for it (as for example the Spring HATEOAS library here)
Also, there is a proposed vnd.error mime type for this (further explained in this post), but it seems it has been superseded by RFC7807 mentioned above, at least Spring deprecated this in favor of the Problem Details approach (here)
I know the 404 vs 204 debate has been beaten to death, and I understand the argument for using 404 when there is no record in the table corresponding to a REST endpoint request, but it feels like there should be some way of differentiating between "This endpoint is malformed" and "there is no record in the table." For example if I have an endpoint like this:
https://mycloudfront.cloudfront.com/api/my-table/{userId}
Is there a recommended way of configuring error handling on the backend to differentiate between "no resource found because there is no entry for userId" and "no resource found because there is no table named my-table" or "no resource found because there is no cloudfront distribution named mycloudfront"?
I ask, because it would be nice on the frontend to inform the end user whether or not their request did not produce the desired result because they have no data in the table (in which case I would display a message encouraging them to take an action that would generate data) or because something went wrong (in which case I would display an error message).
it would be nice on the frontend to inform the end user whether or not their request did not produce the desired result because they have no data in the table
That's what the response body is for.
Except when responding to a HEAD request, the server SHOULD send a representation containing an explanation of the error situation, and whether it is a temporary or permanent condition. RFC 9110.
Status codes are metadata in the transfer-of-documents-over-a-network domain (Webber, 2011) - the information indicates to general purpose web components (browsers, proxies, caches, spiders....) the semantics (meaning) of the fields and response body (ex: does the message include a representation of a resource or a representation of an error?)
Bespoke HTTP message handlers (and human operators) are expected to look for information in the body (ex: a 404 for a web page returns a picture of a fail whale and a bunch of links to different resources that might clarify what's gone wrong).
You can also leverage ideas like web linking (RFC 8288), if you want to describe relationships between the error and other resources.
Problem Details (RFC 7807) describes a standardized JSON schema for communicating error information, if you want a JSON representation but prefer not to do all of the schema design yourself.
First and foremost, REST has no endpoints but resources.
there should be some way of differentiating between "This endpoint is malformed" and "there is no record in the table."
By "This endpoint is malformed" I guess you probably mean the request issued to the server doesn't conform to the HTTP specification. As voice already mentioned, HTTP status code are coordination metadata for the outcome of the transportation and not necessarily the outcome of your business logic. Of course you need to come up with a mapping for problems you noticed while applying your business logic to the HTTP transport domain.
Unfortunately, REST is polluted with false assumption and believes. Plenty of people seem to think of it as HTTP based CRUD mostly done with JSON payloads. But this is just a very tiny fraction of what REST really is. At its heart it is a technique used in distributed computing to help decouple clients from server to allow the latter to evolve freely in future. Clients on the other hand are build with the inherent design decision of a possible change in mind and therefore get much more robust towards change in the end.
So, how does REST help to decouple clients from servers?
First, the spelling of a URI is not of importance. The URI needs to be a valid one but that's it basically. Clients shouldn't parse the URI or try to extract some knowledge off the URI nor does a URI pattern like /api/user/1 and /api/user/1/stuff mean that both of those URIs are somehow related. That's what link-relations are there for.
Next, in order to teach a client what an URI returned by the server is good for URIs should come with one or multiple link-relation names, which should either be based on registered ones or at least follow the Web Linking extension mechanism, which basically is just a further URI that does not necessarily need to point to a valid resource. Treat it like a predicate in a (SemWeb) ontology.
Use forms similar to HTML, like HAL-forms, JsonForms or Ion, if your server needs further input from clients. Forms also teach clients on what HTTP methods to use, which URI to send the request to, what media-type to encode the request in and of course a description of the properties the resource has and/or the server expects input for. This information is enough to let a client send valid HTTP requests in terms of the transport domain to the server. Note that this does not mean that there won't be any issues then. Requests still might fail to reach the server due to internet outage on whatever end, the request being routed badly and exceed the maximum number of allowed hops and so on but depending on the HTTP method used for sending the request a client might automatically reissue a request once it hit its timeout threshold.
In order to increase interoperability of any peer in a REST ecosystem REST has a strong focus on media types. Think of it as the binding contract between a client and a server which should be negotiated between both of them. This guarantees that both are capable of exchanging "messages" both understand and are able to process. One of the difference to regular RPC services here though is that RPC services are usually restricted to one payload mechanism while REST supports more or less an unlimited amount of payloads, depending on its support for various media-types. Media types are a human-readable description on how payload should be encoded and processed and also contains information, besides the syntax description of allowed elements, a semantic description on the purpose of the respective elements. A payload issued for plain application/json doesn't teach a client really what the properties of the respective JSON objects used in the payload mean nor does it really support URIs in first place. Note however that issuing a plain JSON request to the server is fine if the client was "instructed" that way using a form the client was acting upon. The server here just expects that kind of payload then. Just look at how a typical HTML document is build up and read up on some of the tag definitions that are used within the HTML document and you might get the gist of this paragraph.
Especially about the latter two points Fielding himself was quite vocal about in his famous rant:
A REST API should spend almost all of its descriptive effort in defining the media type(s) used for representing resources and driving application state, or in defining extended relation names and/or hypertext-enabled mark-up for existing standard media types ...
So, back to the actual question at hand. Is "there is no record in the table" really a business logic error? You could also design it to return what's currently available there and return an empty list. This at least spare you the hazzle of mapping that business error onto the transportation domain in that case.
If you want or need to express a business logic failure to the client you should, as voice also recommended before, look into application/problem+json (or its XML alternative application/problem+xml) which define properties such as type of failure, general title, status and details among others. The respective type the response is issued for may define further properties specific to that type that are part of the payload. I.e. you may define an extension type of http://acme.com/problem/validation and this extension type defines that the payload needs to contain a target-ref property to identify the element that failed the validation check as well as a property for the actual error message.
In the end some general recommendations in terms of REST are:
Design the interactions of client and servers first as if you'd interact with a typical human-focused Web page and then translate the interaction steps onto the application domain. REST in the end is nothing more than a generalized approach for how we humans interact on the Web for decades. REST is basically Web surfing for applications rather than humans. As we humans follow an outlined state machine of i.e. Amazon.com to order some books, computers can do the same. Therefore design the whole interaction between client and server as state machine that clients just follow along and may exit at certain points
Allow servers to teach clients what they need to know using various form-support and use link-relations to set given URIs in context to the current resource
I have an API that will delete a resource (DELETE /resources/{resourceId})
THE above API can only tell us to delete the resource. Now I want to extend the API for other use cases like taking a backup of that resource before deleting or delete other dependant resources of this resource etc.
I want to extend the delete API to this (DELETE /resources/{resourceId}?backupBeforeDelete=true...)
Is the above-mentioned extension API good/recommended?
According to the HTTP Specification, any HTTP message can bear an optional body and/or header part, which means, that you can control in your back-end - what to do (e.g. see what your server receives and conventionally perform your operation), in case of any HTTP Method; however, if you're talking about RESTful API design, DELETE, or any other operation should refer to REST API endpoint resource, which is mapped to controller's DELETE method, and server should then perform the operation, based on the logic in your method.
DELETE /resources/{resourceId} HTTP/1.1
should be OK.
Is the above-mentioned extension API good/recommended?
Probably not.
HTTP is (among other things) an agreement about message semantics: a uniform agreement about what the messages mean.
The basic goal is that, since everybody has the same understanding about what messages mean, we can use a lot of general purpose components (browsers, reverse proxies, etc).
When we start trying to finesse the messages in non standard ways, we lose the benefits of the common interface.
As far as DELETE is concerned, your use case runs into a problem, which is that HTTP does not define a parameterized DELETE.
The usual place to put parameters in an HTTP message is within the message body. Unfortunately...
A payload within a DELETE request message has no defined semantics; sending a payload body on a DELETE request might cause some existing implementations to reject the request
In other words, you can't count on general purpose components doing the right thing here, because the request body is out of bounds.
On the other hand
DELETE /resources/{resourceId}?backupBeforeDelete=true
This has the problem that general purpose components will not recognize that /resources/{resourceId}?backupBeforeDelete=true is the same resource as /resources/{resourceId}. The identifiers for the two are different, and messages sent to one are not understood to affect the other.
The right answer, for your use case, is to change your method token; the correct standard method for what you are trying to do here is POST
POST serves many useful purposes in HTTP, including the general purpose of “this action isn’t worth standardizing.” -- Fielding, 2009
You should use the "real" URI for the resource (the same one that is used in a GET request), and stick any parameters that you need into the payload.
POST /resources/{resourceId}
backupBeforeDelete=true
Assuming you are using POST for other "not worth standardizing" actions, there will need to be enough context in the request that the server can distinguish the different use cases. On the web, we would normally collect the parameters via an HTML form, the usual answer is to include a request token in the body
POST /resources/{resourceId}
action=delete&backupBeforeDelete=true
On the other hand, if you think you are working on an action that is worth standardizing, then the right thing to do is set to defining a new method token with the semantics that you want, and pushing for adoption
MAGIC_NEW_DELETE /resources/{resourceId}
backupBeforeDelete=true
This is, after all, where PATCH comes from; Dusseault et al recognized that patch semantics could be useful for all resources, created a document that described the semantics that they wanted, and shepherded that document through the standardization process.
Closed. This question needs to be more focused. It is not currently accepting answers.
Want to improve this question? Update the question so it focuses on one problem only by editing this post.
Closed 2 years ago.
Improve this question
Can we say Like Rest API we have SOAP, XML RPC, JSON RPC, GRAPH QL I mean Rest is just a type of API?
Is Rest is just a mechanism to share data between applications by using HTTP methods?
REST can share data between applications only with HTTP there is no alternative?
What is the relation between REST & CRUD exactly? we are saying HTTP: POST is CRUD: CREATE my question is HTTP: POST will just post the data to server and it's business logic's headache to CREATE A NEW RESOURCE in the server side but how we all are saying HTTP: POST is nothing but CRUD: CREATE here HTTP: POST is just helping to share the data only right then how it is related to CRUD: CREATE? If that is the case we can create a new resource with HTTP: GET by writing business logic right then why so many sites are saying REST is a mechanism to do CRUD operations... But it should be REST is a mechanism to help in Sharing the data between applications right?!!! (I have doubt similarly with HTTP: DELETE, GET, PUT aswell)
Last but not least... what exactly Representational state transfer mean? could you please answer this with a very low level general answer instead of definition.
Can we say Like Rest API we have SOAP, XML RPC, JSON RPC, GRAPH QL I mean Rest is just a type of API?
REST stands for representation state transfer and is just an architectural style, not a technology or protocol. According to Robert C. "Uncle Bob" Martin an architecture is about intent and the intention behind REST is the decoupling of servers and clients in a distributed system.
REST basically just defines a set of constraints that when followed correctly allow servers to change at any time without breaking clients as clients will just depend on the data given by the server and not on any external data or documentation. REST can be regarded as Web surfing for applications. The main premise should always be that a server teaches a client on how certain things can be achieved.
On the Web a server can i.e. teach a client on the supported properties of a resource through the help of HTML forms. Not only does a client learn that way what a server expects as input, it also learns what HTTP operation should be used to send the data to the server, the endpoint URI to send the request to as well as, usually implicitly given, also the media type to convert the input to, which is application/x-www-form-urlencoded usually by default, which transforms an input like the HTML example below for first and lastname to something like this:
fName=Roman&lName=Vottner
Is Rest is just a mechanism to share data between applications by using HTTP methods?
REST itself is protocol agnostic meaning that it is not tide to HTTP itself and could just work on other transport protocols as well. Though the common perception many developers have is that it is based on HTTP. After all, as Jim Webber put it, HTTP is just a transport layer whose domain is the transfer of files or data over the Web. All HTTP does is to send one document from one machine to the next and any business rules we conclude from sending/receiving a request are just a side effect of the actual document management. It is therefore always better to think of a request as a whole document and the HTTP operation define how the document should be stored on the current machine, especially when such a document is already available, instead of thinking of a service method invocations. The latter one is a typical RPC view.
REST can share data between applications only with HTTP there is no alternative?
HTTP is just a transport layer used in a REST architecture. The architecture cares more on the interaction model between client and servers than on the technical nuances of HTTP. As REST itself is transport protocol agnostic it could be used with other, maybe proprietary protocols as well.
What is the relation between REST & CRUD exactly? we are saying HTTP: POST is CRUD: CREATE my question is HTTP: POST will just post the data to server and it's business logic's headache to CREATE A NEW RESOURCE in the server side but how we all are saying HTTP: POST is nothing but CRUD: CREATE here HTTP: POST is just helping to share the data only right then how it is related to CRUD: CREATE? If that is the case we can create a new resource with HTTP: GET by writing business logic right then why so many sites are saying REST is a mechanism to do CRUD operations... But it should be REST is a mechanism to help in Sharing the data between applications right?!!! (I have doubt similarly with HTTP: DELETE, GET, PUT aswell)
Fielding's thesis on the REST architecture style does not contain the term CRUD at all. The term REST nowadays is heavily misused as people probably didn't bother to actually read the thesis, which admittingly is a bit abstract, and just follow what some people thought may be REST but was actually RPC. Nowadays, if a typical stakeholder talks about REST they usually think of a JSON-based HTTP CRUD-API whose supported endpoints are defined in some Web documentation (Swagger, OpenAPI, ...) and where the HTTP operations for POST (= Create), GET (= Read), PUT (= Update), Delete (= Delete) are supported by default. However, this is unfortunately far from the truth. Though people are just to accustomed with their (wrong) definition and don't see or don't care about the actual problem in their misusage. They don't care about a long-lasting service as in 2-5 years the next-gen technology is here that allows to reduce the number of lines of codes even more and if a new "version" of a service is needed, this usually goes hand in hand with a technology switch also, to justify the "cost of change" somehow.
Last but not least... what exactly Representational state transfer mean? could you please answer this with a very low level general answer instead of definition.
Probably the easiest way to grasp how the interaction in a REST architecture should be is by analyzing typical interaction on the Web, the big cousin of REST. You, as a user, usually start by opening your browser (= client) and typing in some URL in the search bar. Next a Web page is rendered on your screen. Behind the curtain a couple of things happened. Besides the whole connection management and any eventual TLS handshake your browser sent at least one GET request to the target server. On sending the GET request, the client included information on his capabilities, i.e. through the Accept HTTP header. This header is used on the server side to decide which representation format to generate and send to the user. On the Web this might be something like text/html or application/xhtml+xml or if some report is generated might be something like application/pdf or application/vnd.ms-excel or the like, depending which representation format fit the data best.
The representation format itself is now a concrete instance of a document following a certain media type specification. I.e. the HTML forms specification defines the supported elements within a <form>...</form> tag pair as well as describes the semantics of each of the elements. The concrete instance may now define a form as such:
<form action="/action_page.php" method="get">
<label for="fname">First name:</label>
<input type="text" id="fname" name="fname"><br><br>
<label for="lname">Last name:</label>
<input type="text" id="lname" name="lname"><br><br>
<input type="submit" value="Submit">
</form>
which should only use elements and attributes defined within the specification else a receiver of that document might not be able to process it correctly.
This process of telling the server which document types a client supports and allowing the server to chose a fitting representation is called content type negotiation and allows the exchange of arbitrary, type-less documents. Of course, both parties need to support and understand at least one common media type to be able to interact with each other. This is similar to a Frenchmen who does not understand a word of Chinese and a Chinese one that does not understand any word of French who need to communicate (for whatever reason), if both speak English they will be able to communicate.
There are loads of different media types already available that all server different purposes. Depending on your needs an all-purpose one, such as text/html, might be sufficient, others such as application/json or even application/hal+json might though lack support for certain needed elements. Existing media types might not support needed elements. In such a case extending such media types and registering those is probably easier than creating a whole media type from scratch.
REST assumes that a resource (i.e. a remote document) contains some internal data, its state. This state can be represented in many different ways. Think of some monthly sails figures. You might ship the data either in a HTML table, as CSV data, as Excel file, as PDF or yet a different representation format. Regardless of the chosen media type to marshal the data in, the actual data at least should express the same. Instead of questioning which media type you want to support, you should better ask how many different ones you want to support as this just increases the likelihood that other clients may interact with your server also.
edit:
I got all the points except 4th point... Could you please elaborate a bit.. So is it just a mechanism that helps in sharing the data between applications by using HTTP methods? we can say like that?
CRUD is a typical term in the context of persistence, especially with databases. REST or more formally the REST architecture itself treats persistence as internal detail. A typical user usually does not care whether some data is persisted into a DB, a local file system or is kept just in memory. All s/he cares is that the server can process it or for storage services also return the same data that was sent to the server.
In regards to the mapping of CRUD to the HTTP operations, if you take a look at HTML you might see that it only supports POST and GET operations. So anything related to C, U or D are performed with POST which is defined to process the enclosed representation according to the resource's own semantic. With POST you are basically allowed to do anything, even retrieving data if you like.
However, HTTP defines certain properties for the respective operations:
safe
idempotent
cacheable
The first property is a promise to clients that a well-behaved server should not alter the state of a resource upon requesting. The second one is a promise in regards to automatic retry attempts caused by i.e. temporary connection issues. And the latter one allows clients to store responses locally and reuse these instead of requesting the same resource again, if the cached content is "fresh-enough".
GET and HEAD are both safe, idempotent and cacheable, meaning that a client can request resources with such an operation without being hold accountable for any eventual changes. Think of a Web spider that is invoking arbitrary URIs all the time to learn new pages over time. If a GET request on a URI would trigger an order of a Pizza or the like, it is basically the server's fault and not the clients one if a crawler would order Pizzas every time such URIs are called.
PUT and DELETE are only idempotent, which basically allows a client to automatically resend a request in case of a network issue as the outcome of the operation leads to the same result regardless whether the request was processed once or multiple times in a row. Note that this property does not consider changes done by other clients to that resource between a resend. Such data would of course be overwritten.
The remaining operations (POST, PATCH, CONNECT, TRACE) are neither of these.
While technically it is sufficient to only use POST for each request, the above mentioned properties should trigger an inner intention to use them, when appropriate. However, as before mentioned, not the client should chose which operation to perform but the server should tell a client which HTTP operation it should use.
In regards to POST vs. PUT, both operations should behave similar on creating a resource. Both need to add a Location header within the response that teaches a client about the location of the new resource. PUT however, in contrast to POST, replaces the current representation of the requested resource with the one provided in the request body. So it already targets the respective resource while for a POST request the server defines where the resource is created. It is allowed to perform certain sanity checks and also to transform the representation to fit the representation format of the current one. It is also allowed to have side effects, i.e. think of Git where a commit creates a new entry on top of the current branch and moves the HEAD to the new commit.
PUT is probably considered as update operation as the replacement of the document more or less has the effect of an update. If no representation was yet available this just has the effect of the creation (including the location header). In the past, unfortunately, many developers used PUT incorrectly by performing a partial update instead of really replacing the whole document. While the spec states that a partial update could be achieved by overlapping resources (i.e. share parts of the same data in multiple resources), the usage of PATCH, which also is used incorrectly most of the time, may be better from a performance standpoint on larger resources.
Due to POSTs definition, one can do anything with it, though historically a document upload in HTML was triggered through this operation that is basically a resource creation on the server side. That POST is used for many other things as well though is not that important for that CRUD paradigm.
In regards to your concerns about the right terminology, most people, according to my experience, simply do not care. They just want to get the job done ASAP and move on. As roughly 90% of the users seem to understand a pretty similar concept when talking about REST (even though this view is flawed) which usually resolves around JSON, HTTP, CRUD, Swagger/OpenAPI, ... they usually only look for quick-win-solutions and more or less agreement on their thought process.
As HTTP (0.9-1.1) is a plain text protocol sending a GET request is not much different from a POST or PUT request, so technically you can create resources with GET request or support payloads on GET requests (semantics of the payload is undefined according to the spec). That's why I mentioned well-behaved client/servers above. In such a case, however, due to the safe property of GET, if you as a server maintainer violate the HTTP protocol you are the one to blame in case something "unexpected" is happening (crawler is ordering 500 Pizzas).
Just started my first WCF rest project and would like some help on what are the best practices for using REST.
I have seen a number of tutorials and there seems to be a number of ways to do things...for example if doing a POST, I have seen some tutorials which are setting HttpStatusCodes (OK/Errors etc), and other tutorials where they are just returning strings which contain result of the operation.
At the end of the day, there are 4 operations and surely there must be a guide that says if you are doing a GET, do it this way, etc and with a POST, do this...
Any help would be appreciated.
JD
UPDDATE
Use ASP.NET Web API.
OK I left the comment REST best practices: dont use WCF REST. Just avoid it like a plague and I feel like I have to explain it.
One of the fundamental flaws of the WCF is that it is concerned only with the Payload. For example Foo and Bar are the payloads here.
[OperationContract]
public Foo Do(Bar bar)
{
...
}
This is one of the tenants of WCF so that no matter what the transport is, we get the payload over to you.
But what it ignore is the context/envelope of the call which in many cases transport specific - so a lot of the context get's lost. In fact, HTTP's power lies in its context not payload and back in the earlier versions of WCF, there was no way to get the client's IP Address in netTcpBinding and WCF team were adamant that they cannot provide it. I cannot find the page now but remember reading the comments and the MS guys just said this is not supported.
Using WCF REST, you lose the flexibility of HTTP in expressing yourself clearly (and they had to budge it later) in terms of:
HTTP Status code
HTTP media types
ETag, ...
The new Web API, Glenn Block is working addresses this issue by encapsulating the payload in the context:
public HttpResponse<Foo> Do(HttpRequest<Bar> bar) // PSEUDOCODE
{
...
}
But to my test this is not perfect and I personally prefer to use frameworks such as Nancy or even plain ASP NET MVC to expose web API.
There are some basic rules when using the different HTTP verbs that come from the HTTP specification
GET: This is a pure read operation. Invocation must not cause state change in the service. The response to a GET may be delivered from cache (local, proxy, etc) depending on caching headers
DELETE: Used to delete a resource
There is sometimes some confusion around PUT and POST - which should be used when? To answer that you have to consider idempotency - whether the operation can be repeated without affecting service state - so for example setting a customer's name to a value can be repeated multiple times without further state change; however, if I am incrementing a customer's bank balance this cannot be safely be repeated without further state change on the service. The first is said to be idempotent the second is not
PUT: Non-delete state changes that are idempotent
POST: Non-delete state changes that are not idempotent
REST embraces HTTP - therefore failures should be communicated using HTTP status codes. 200 for success, 201 for creation and the service should return a URI for the new resource using the HTTP location header, 4xx are failures due to the nature of the client request (so can be fixed by the client changing what they are doing), 5xx are server errors that can only be resolved server side
There's something missing here that needs to be said.
WCF Rest may not be able to provide all functionality of REST protocol, but it is able to facilitate REST protocol for existing WCF services. So if you decide to provide some sort of REST support on top of the current SOAP/Named pipe protocol, it's the way to go if the ROI is low.
Hand rolling full blown REST protocol maybe ideal, but not always economical. In 90% of my projects, REST api is an afterthought. Wcf comes in quite handy in that regard.