This will be useful for stateful workloads, for instance a game engine. The administration / gateway server (as described in Hosting game servers on demand (when another service asks for it) - #4 by kurt) can listen to any changes in the status of machine and pro-actively register new instances if needed (without waiting for a client connection to come, and then figuring out if the machine is healthy or not)
Also, is it ok to query machine status every 5 or 10 seconds for let’s say 10 machines running in an app? Is that a cheap call, or could we expect to run into some rate limits.