The main issue is that CloseMessage is written to the pipe and then the app task exits, then connection cleanup starts which will complete the transport pipes, this results in a race where the CloseMessage might not get sent if there was any backpressure or thread scheduling was unfavorable for the pipe dispatches.

Additionally, there are scenarios where we would like to specify the reason for closing the connection and it is non-trivial to do that, so we should make it easier.

Potential solutions:
Wait a little after sending the close message to give it time to send
Somehow detect that the close message is sent before completing the app task (with timeout of course)