GitLab

When #8171 (closed) is done, make the facilitator dynamically adjust its check-back-in times, separately for IPv4 and IPv6, in order to keep a desired proxy capacity.

Let's say that a desired proxy capacity is that an arriving client gets proxy service with expected waiting time 2 seconds, given the current client arrival rate. Make a reasonable model of proxy arrival (e.g. a Poisson process) and tune it so the mean interarrival time is 2 seconds. Put a minimum of 10 seconds on it so if there are few proxies we don't ask them to check back in 0 seconds.

Quantize, maybe to one-minute or perhaps geometrically increasing intervals, so that you can't exactly calculate the current number of proxies from the polling interval.