Rate limiting

[tzurbaev]

Rate limiting

[tzurbaev]

Actually, there're more questions than answers about rate limiting. How exactly should we handle this?

1. Throw RateLimitExceededException? If yes, then the request somehow should be repeated after we're allowed to make requests to the API. How this should be done? Save failed requests to cache/queue and repeat after 30 seconds? I don't think that caching/queueing mechanism should be added to SDK, so this must be controlled in the target application.

2. Or should we ignore failed requests and require SDK's users to repeat them manually?

3. Or even put the script to sleep($secondsLeftBeforeNextWindow); (that's the ugliest solution as for me and probably this won't work at all)?

At this moment, the only way I see is to throw exception on HTTP 429 responses (actually, Guzzle already throws it but we can rename it to more informational one) and let users handle it in their preferred way. What do you think?

[acurrieclark]

I have been thinking about this, and I feel that it might be worth offering an option. For example, setting an optional useRateLimiting flag on the $forge instance to which a closure is passed. The closure provides the rate limiting code, and could pause as required.
I imagine most users will only need to rate limit across one instance of an application, but this would also allow people to code an implementation which rate limited across multiple instances.
I will try and look at how this can be achieved over the next week or so and offer something up.