Feature request for no_traffic_edge_interval

Description:

We have a setup where we have two envoys pointing to the same cluster. At the moment we have traffic routing to only one of the envoy instances. What is problematic is that we have no_traffic_interval set to 100ms. Which is great when we only have one envoy instance, since when a cluster is first up, we want to check to ensure it is healthy before any requests goes upstream. But when we only have traffic going to one envoy instance, the other envoy will be spamming the cluster every 100ms thinking that there has been no traffic.

We could increase the no_traffic_interval to something reasonable like 1 second to reduce the spam, however if once there is traffic coming over, from what I understand, envoy will hold onto the request before letting it go upstream which leads to slower latency?

Proposed solution

Add a no_traffic_edge_interval that specifies the interval to run the health check after a cluster is marked healthy during the no_traffic_interval stage