I’ve been thinking a fair amount about service design recently, and one of the things that I’ve found difficult is deciding how to implement and version a service in such a way that makes it pretty trivial to support (or not) older versions of that service. Turns out that advice which was meant to be for writing code works for writing services too viz. Program to an Interface.

Program to an Interface

Borrowing from a lot of blogs and books, I’ve decided that viewing a service interface in the same way I’d view a programming interface will help me. Interfaces can be versioned, they isolate the client code from the implemention of the interface, and versions should be immutable once published.

It is desireable that a service hide implementation details so that if, for example, a database table changes inside an application which is providing a service, the clients of that service should not have to change.

Much like interfaces in a programming language, by specifying a well known interface to the service we free ourselves from having to worry about how clients of our service interact with the service, so whenever we need to change the implementation of our service we can do so without worry, and vice-versa; clients of our service do not need to worry about the service implementation changing as long as the interface remains consistent.

Of course, interfaces may have to change as we try to support new functionality, and when they do change we want to be confident that we’re using the correct version of the interface; just because v2 has been released, that doesn’t mean our clients automatically support it. We want to continue using v1 until we get a chance to update our a client code. Interfaces should be versioned to give us that choice.

When we think of a service we generally think of a web service and in these heady ReSTful days the interface is generally though to be the combination of URIs that we interact with. That’s not strictly correct. Supporting an interface does not just mean that your URIs are manageg between service versions, it also means that the content returned from service calls (ie HTTP requests).

Of course, a web service is only one kind of service; it’s quite possible to have services that do not have a web interface, most usually in situations where processing not suitable for synchronous request-response style messaging that is usually provided by a web services, situations such as order processing or inventory handling, where we may take several seconds to confirm an order is valid and paid for, that there is stock and that the order meets some anti-fraud criterea which we have defined. The interface to these services should also be versioned for the same reasons as as web service should be.

The version of the interface we are conforming to should be detectable with each message passed or sent, no matter the transport or service type used; we should know that we’re dealing with version 3 of an API and not just that we’re talking what might be some version of the API and trying to guess which one to use.

Handily, versioning interfaces for these types of services is pretty much the ideal use for a mime-type, and most message transports support setting custom mime-types:

Mime types have a space reserved for vendor specific types, application/vnd, inside which we are free to define our own types, although thre are a few conventions which are useful to make sure we do not have name collisions. Name your mime types such that they to include your organisation name, a very short - one or two words - description of what they’re representing, a version number, and a base format.

For example, say you worked at the Acme Toy Company, and when your web service accepted an order via its ReSTful interface it would put on a message queue a message with four fields, customer_id, purchase_order_id, amount and description, in a JSON representation:

  "customer_id": 123,
  "purchase_order_id": "ASLA-001-2031",
  "amount": 1000,
  "description": "100 x Acme Toy Dynamite"

We could coin a mime-type, application/vnd.acme.order-v1+json, and publish an interface specification that says any message claiming to be application/vnd.acme.order-v1+json would have these four fields. Then in a consumer of the orders message queue we could use the Selective Consumer pattern to subscribe to the queue asking for just messages of this mime-type. Inside the consumer we can now be fairly confident that we will deal with only orders which we know the format of and can process. We could have parters POST messages with this mime type in the Content-Type header so we know we’re talking about the same thing all the way through the system.

A few months pass, several partners are using the order API, but we want to automate our stock inventory systems, so we decide that instead of a plain text description we want item ids. We don’t want to force this change on our partners however, because their development cycle is slow and they’re sending us loads of orders - we like their cash.

We can now publish a second version of our order interface, application/vnd.acme.order-v2+json, which defines messages which look like this:

  "customer_id": 123,
  "purchase_order_id": "ASLA-001-2031",
  "amount": 1000,
  "items": [
    { "item_id": 1032, "quantity": 100 }

It’s now pretty trivial to add a second Selective Consumer which consumes only application/vnd.acme.order-v2+json messages and updates our inventory accordingly. There’s no reason that the v1 conumer can’t keep running and working with v1 orders, and there’s a smooth and unhurried migration path for our clients from version 1 to version 2 of our API. We can easily support both versions or reject older versions as we choose. We could even use some combination of Splitter, Translator and Enricher to route the v2 message into the consumer for the v1 messages whie splitting off the inventory management messages and routing them to a tiny consumer elsewhere. None of this matters to our partner however, because they know that they’re working to the interface we’ve defined.

We may eventually decide that the ReSTful order service is not appropriate for v3 of our service; perhaps we bought over by web 2.0 hip kids and want to use WebSockets because hey, those are cool, right? When we receive an order which claims to be v1 or v2 on the ReSTful service we can still happily accept it, if we receive anything else we can return an HTTP 406 to tell the client that we can’t accept orders that way for v3.

In contrast, if we had used application/json we’d have to do some sort of guessing based on the message fields we were passed to know which version of the service the client wanted to us. This would be just about practical with the trivial example above, but it’s messy, and once there are several versions of the service interface it becomes much harder.

written by
Disagree? Found a typo? Got a question?
If you'd like to have a conversation about this post, email craig@barkingiguana.com. I don't bite.
You can verify that I've written this post by following the verification instructions:
curl -LO http://barkingiguana.com/2011/12/05/principles-of-service-design-program-to-an-interface.md.orig
curl -LO http://barkingiguana.com/2011/12/05/principles-of-service-design-program-to-an-interface.md.orig.asc
gpg --verify principles-of-service-design-program-to-an-interface.md.orig{.asc,}