Hacker News Reader

Comment by pen2l | Hacker News Reader

There’s also that when you’re kind of big, there’s a brand to be maintained and that rarely involves presenting the dissected view of your stack. In fact at some it’s readily avoided by some. I recently discovered, for example, that Django-REST is quite often used for big sites. Like Robinhood and Eventbrite.

View on HN

I'm assuming you're referring to Django Rest Framework? It's one of the nicest API frameworks I've had an excuse to use professionally. It's easy to see why.

Top notch documentation. Carefully thought out patterns like authentication pipelines. Simple syntax one you learn it. There's lots to like.

I feel like DRF's documentation is a bit lacking, especially when compared to Django's documentation.

Agreed, while the answers are usually in there (somewhere, which is better than most documentation), it can sometimes be quite difficult to find "that page you saw in here once".

I experienced this first hand when I first started doing timed coding interviews and didn't prepare my DRF environment. Rookie mistake for sure, but it really emphasized to me that things can be difficult to find in there unless you know exactly what you're looking for.

It has so many built in footguns, however. For instance,`SerializerMethodField`, in conjunction with a list endpoint, can generate a huge number of queries if one doesn't prefetch appropriately, and this issue is very difficult to catch with a linter. Granular test coverage is difficult because the testable unit is the ViewSet or the Serializer. I think DRF is great for rapid prototyping and early stage development.

In the long term, with a lot of complexity and a large organization, a more mature architecture that relies on components that wrap tables, composed together to form components that wrap business logic, which are composed into endpoints. Something like that is much easier to test and more scalable for disparate teams to step into and understand.

Can you point out a better generic framework for this? And starlette doesn’t count as Tom Christie is the author of both.

I haven't found one, unfortunately -- just setting team-wide architecture policies, and then using linters extensively to make sure those policies are adhered to.

That's my experience as well. There isn't a nicer one than DRF / Starlette, but for a specific use case, you can get away with a custom one no problem.

Boring tech is not blog or social media worthy I guess!

It's interesting though. The container orchestration at work has kind of moved past the initial scary phase, and kinda past the first honeymoon phase. In the beginning, it was like "Oh we can use all kinds of technology in there, how amazing!" - by now we're much rather becoming very, very efficient a smacking problems with the 3 boring old sticks dev has. Something like deploying and monitoring low-volume spring boot applications based on postgres has become extremely streamlined by now.

Reply on news.ycombinator.com