Doug Fawley

The gRFC was updated: https://github.com/grpc/proposal/pull/452

Yes. We implemented the gRFC as documented before, and now the behavior is changed.

> For the channel-level option, my inclination is that the xDS config should take precedence over the channel-level default, especially since there will be a knob in the bootstrap file...

I'm not sure about slog yet - it may be more complex than we need, and it might require that we change grpc over to that same interface, since the...

Looks like we should be accepting a `*slog.Logger` and not the `Handler` and not our own interface.

Filed #8276 for the flake. And we can ignore the vet-proto error - we need to regenerate our proto messages, but not in this PR.

> Client closes connection: Why is the application responding to a cancellation by the client differently from connection loss? Can you give reasonable examples of why it would want to...

Note that connection loss can already be detected using the stats handler's HandleConn getting a ConnEnd message, so it's possible that surfacing these connection termination events on your dashboards might...

> The question arises about what other cases there are, apart from that "typical" scenario. In any case, the connection to the RST frame is very subtle. The way RPCs...

> This seems to come from the assumption that the client is also implemented in Go. That paragraph was written with a Go client in mind, but the concept is...