Some random things I’ve been thinking about this week:
The first thing that comes into my mind when I hear “category theory” is the basic stuff we learned about at university, such as the set union operation. It turns out that category theory is much more than that - we can also reason about the functions and types that make up a software application using category theory. I don’t really grasp it yet, but I will try to dive into it a bit more - and maybe pick up some Haskell along the way.
- Bartosz Milewski has a really nice article series called Category Theory for Programmers.
- For Clojure, there is the cats library.
- I heard about a book called Maybe Haskell that should clear up some basic concepts.
I recently had a scenario where an entity fetched from the database needed an additional field from a different storage backend, requiring an additional query to that storage medium for each row. Think of fetching a collection of books and for each book having to fetch its author separately. That’s a lot of queries! This is called the N+1 problem.
In my case, I solved the problem with some application-layer caching. Another common pattern
is to get the id’s that need to be fetched and do a batch query using e.g.
IN (1, 2, 3 ..),
which is one of the things that the neat muse library for Clojure provides. But keep in mind that using a list of id’s is not going to perform very
a large number of id’s.
Queues, back pressure
I have grown very fond of queues as a way of decoupling the different parts of a system. The Language of the System is one of those classic Rich Hickey talks and provides some context for why system communication is best done using data put on queues. Zach Tellman also did a great talk on the challenges of a queue-based system at Clojure/West 2015.
One of the terms that gets thrown around a lot when talking about queues is back pressure. This basically refers to some mechanism that the consumer of a queue has to let the producer know that it can’t handle any more work at the moment and may come in the form of an HTTP response code, a RejectedExecutionException or something other interesting. Another strategy would be to eliminate the need for back pressure by using e.g. a dropping buffer.
Proper API design and HATEOAS
I’ve finally gotten around the reading Leonard Richardson’s and Mike Amundsen’s book RESTful Web APIs. It’s making me realise all the mistakes I’ve made designing APIs in the past - I’m definitely guilty of forcing API consumers to read the documentation instead of following the HATEOAS pattern.
It’s also made me think again about the different data formats REST APIs normally produce. Everyone’s favorite, JSON, turns out to be a pretty bad choice since it isn’t a hypermedia format due to its lack of links. The Transit format does and also takes advantage of the fast, native JSON parsing that browsers provide. Other extensions to JSON include HAL and Collection+JSON.
Another observation would be that XML perhaps isn’t so bad at all, since it can support the HATEOAS pattern bettern than JSON.