This could be a beginner question so I apologise.

It was not clear to me in the documentation how HTTP2 load balancing works.

As new HTTP2 requests come in from a single client on a HTTP2 connection to a envoty route does each client request continue to communicate with the original backend instance (say 7 of 9) or does each request within a client connection to Envoy get load balanced to all 9 backend instances of a cluster.

Is this also configurable as I could imagine many architecture scenarios where different affinity makes sense