Maybe we can add throttling to the API

[jllorencetti]

Now that #18 is done, maybe it'd be a good idea to add throttling to the API, here.

For now I don't think that we really need some great values, probably 1/minute will be fine.

[cuducos]

I'm not really sure about what it really means:

Before running the main body of the view each throttle in the list is checked. If any throttle check fails an exceptions.Throttled exception will be raised, and the main body of the view will not run.

Does it really block access to data (responding the request with a 4XX HTTP status?) or it just returns cached data?

As we have only GET entry points to the API so far (I mean, people do not send data to Pets), I would limit access (i.e. save resources) by recurring to a cache, but not blocking access to data:

• Once there is a request, the response is saved in Redis or something like that;
• For the next few minutes (hours?) every request to the same entry point respond with the cached response, not really accessing the view, the database and so on…
• The first request after a few minutes (hours?) generate a new cached version, valid for the next interval and so on.

I use this approach with the news in Cunha já caiu? (in the backend there is requests, requests-cache and Redis).

[jllorencetti]

Hey @cuducos thanks again!

The idea behind the throttling suggestion is not about saving resources, to be honest.
It's more about helping others be better internet citizens, and choosing a reasonable refresh rate.

I mean, at the moment the only website using the code is really small, we usually get a new pet once per week or even less. So there isn't much point refreshing, let's say, every second.
With this in mind I'd probably prefer to put some kind of "road block" and deny access to the data in cases of "abuse".

This way I hope that if someone come across the API, decide to create an app for it, using a production website to test it, they will probably hit the rate and hopefully change the rate refresh. But sure, it's just speculation from my side.

Hopefully we'll face a time that we''ll need a cache system in place, but I don't think that this is case here.

What do you think about it?

[cuducos]

Ok… I get it. I think my mind just plan on needs & requirements, so that's why I misunderstood you. But I do get your point and I'm willing to experiment with this feature from Django REST Framework — so this might be my next commit. Can't promise anything soon, unfortunately.

[jllorencetti]

@cuducos no problem!
Take your time, the help is much appreciated. :)