Asynchronous Requests

[lkingland]

Currently Knative will scale down a Function instance which is working on a long-running asynchronous task.

For example, a long-lived function which return immediately to a client (such as with a "201" accepted), and then performs follow-up actions, needs some way to signal to Serving that it is busy with work and should not be scaled down.

This could be implemented using an additional endpint ala "health/liveness", "health/readiness", "health/progress"

This endpoint, at minimum, would return either true or false, indicating that progress is being made towards asynchronous tasks, with the nomenclature leaving open the possibility to return structured data enumerating the tasks in progress by name and additional metadata about said processes, with an empty set indicating idleness (and availability to be scaled down). Importantly this latter would be an optional upgrade to enable more rich integration with monitoring etc.