programming apis software
I recently published a blog post for the Ad Hoc Blog titled What makes an API simple?. In that post I explored the tradeoffs around some API design decisions, and what it means for an API to be "simple". Here is that post :)
When designing an API, you should always strive for simplicity. Unchecked complexity can make your system more difficult to use, change, and debug. These compounding effects can put you and your team in a bind when it comes to improvements, and it can make the system painful or impossible to use for those who depend on it.
While you typically use an API to facilitate machine-to-machine communication, it’s almost always going to be a human who will spend the time building the thing that integrates with your API. Taking the time to ensure that it’s as easy as possible to get your API to do what the customer needs it to do will make your API a better and more effective tool.
But what exactly does "keeping it simple" look like in the context of API design? Does it mean keeping the request input as minimal as possible? Or keeping endpoints focused on a single purpose? Straight-forward and easy-to-use documentation? Or better yet, making it so simple that the required documentation is minimal?
Where your implementation should fall on these various spectrums must always come back to what is best for your users. In other words, there is no silver bullet! What’s important is taking the time to think through the implications of your approach and how it will affect your users.
Today, we’ll be exploring the tradeoffs you make when deciding how minimal to keep your API’s request input data. It’s easy to assume that the best way to keep your API simple is to have fewer request inputs, but what are we sacrificing by going that route?
Let's make a decision
Imagine you have an API endpoint that helps calculate the premium for an
individual on an insurance plan. Let’s assume that one of the variables going
into this calculation is whether the person likes to skydive or not. In short,
we're trying to arrive on whether goes_skydiving
is true or false.
On the surface, it seems like the simplest approach here is to allow the user
to provide the goes_skydiving
field as either true or false. We would be
giving the consumer a single field with an obvious purpose. Easy-peasy. This
raises the question though, are there cases in which this is the wrong
approach?
One assumption we’re making is that this boolean value implies the person actively goes skydiving at this point in time. But what if there are scenarios where we want to know if a person was an active skydiver at a different point in time or under different circumstances?
A changing context in how you’re making this calculation means that
goes_skydiving
might not always simply mean "are they currently an active
skydiver." This makes it a derived value that potentially requires several
pieces of contextual data and the knowledge of how to put that contextual data
together to arrive on the simple true/false value.
Knowing this, we have two options.
- Provide the single
goes_skydiving
boolean, keeping the API request data simple. As a side effect, the consumer is now responsible for knowing the business rules and required data around determining the value of that single field. - Expand the allowed request input, requiring all other contextual data
needed to derive the
goes_skydiving
field. The consumer now has to provide additional data in their request, but the burden of implementing the more complex business logic is now shifted onto the API.
Is it better to have a simple design, or be simple to use?
Having a simple design means that the inputs to the API are limited and easy to understand. It means looking at the expected request to an API endpoint doesn’t make your eyes gloss over. But does this actually mean the API is simple to use?
Option 1: Fewer inputs, simpler interface
Going with option one means we assume that the API user knows enough about the
business logic required to arrive at the goes_skydiving
value. Furthermore,
potentially changing over time.
If we go this route, it’s possible for the user to submit an incorrect value for that single field. The API would likely not have enough contextual data to know whether they applied the current business rules correctly or not. The system assumes the user sent the right value, and as a result, returns the wrong result, potentially in a silent fashion.
Option 2: More inputs, fewer assumptions
Option two requires that the API user gather more individual pieces of
information to pass to the API to get to the derived goes_skydiving
value,
which can be a burden. What if some of those specific pieces of information are
not readily available? What if there is so much required contextual data that
it requires frequent trips to the documentation to get right?
The question is whether this extra data burden is enough of a negative for the
user to offset the possibility that they will calculate the goes_skydiving
incorrectly. Sure, the API implementation will take a bit more effort, but
there is a greater assurance that the result coming from the API will be
correct according to the current state of the policy rules.
Did something change in how goes_skydiving
is calculated over the last year?
Do we now need to look at three years of skydiving history instead of the
previous two? As long as the API implements this change and the user provides
the data to make the decision, then all API users will automatically be using
the updated version of these rules, and we get a stronger guarantee for
compliance. In the government space where policies and rules are often in flux,
this is critical.
Being "simple to use"
When deciding between these two choices, it's important to consider your tool’s purpose. Why are people using your API? What do they need from it? Making your API simple to use means understanding what people are using it for, and optimizing in a way that makes their ability to use the API as simple as possible.
If the purpose of your API is to encapsulate business logic, ensuring calculations and determinations are done accurately according to some particular set of policies which could change over time, then "keeping it simple" could mean placing the burden of that complexity on the API.
By making extra contextual data part of the request input, you can use validation errors to guide the user to ensure they provide the right data. While the time and effort required to integrate with the API may increase, this approach helps ensure that users are integrating properly, and are receiving useful feedback from the API along the way.
Documentation
Whichever approach you choose, you still need good documentation. If you choose a more simple API input interface, the documentation for that single field should contain all of the information necessary for the consumer to arrive on a value for that field. This means packing in a lot of business logic, and hoping the person happens to read it and use the field correctly, and that they stay up to date with any changes in the business logic!
If you choose to require all contextual data and derive the value ourselves instead, you have to be transparent about how the calculation is being performed. Make sure you document each of the contextual data points in a way that shows how they’re used in overall determinations. Educating the users of our tool is always a worthy endeavor, and can only lead to better understanding and use.
Work with your users
Whenever you’re trying to make something simple, it's important to think of the people who are using your tool. When the design of the API reduces the possibility of errors for the user, it becomes a more valuable tool. Making your API easy to use also means people spend less time working to integrate with it, meaning they get more value in less time, a win-win for everyone. An API’s simplicity or complexity is always measured relative to its human consumers, not in isolation.
Sometimes it will be simpler to have a design with composable elements, and sometimes it will be simpler to encapsulate lots of nuanced logic so you guide them along a happy path. As we mentioned from the start, there is no fixed answer. Just make sure you’re considering all the potential tradeoffs of whatever decision you make.
Oh, and don't forget the documentation!