This is nice and could be really useful for event-driven workers. Thanks for this great update! 
Since you referenced the intended behavior of current health checks that will route requests away from failing machines, I’ve actually posted a question a few hours ago about this not working as intended. Could it be related to this change?