ConcurrencyLimitLayer that rejects requests instead of queueing it

[lcmgh]

https://docs.rs/tower/latest/tower/limit/concurrency/struct.ConcurrencyLimitLayer.html

Does it delay responses or does it return an error in case the concurrency limit is met? And if yes which status code?

Edit: From my observation it delays the req/response processing. How would I reject the request when concurrency limit is met (to protect service from accumulating too many yet unprocessed requests)?

[jlizen]

The general approach would be to use a load shed layer outside it: https://docs.rs/tower/latest/tower/load_shed/struct.LoadShed.html

However this would also drop requests if any other inner service returned pending on poll_ready().

I think that there is probably a place for adding a knob to the concurrency layer, personally.