when a load-balancer or health check pings any transport connector endpoint to verify that the broker is listening on a port; using socket.open/close, any subsequent read failure is treated as an error and logged as a WARN.
This makes sense in general b/c it is indicative of a rogue client.
However when it is the norm, ie: from a health check, then the logs get filled with these worrying messages that are in fact expected.
For the somtp transport, where there is no protocol close method, we already suppress EOF and connection reset exceptions.
This improvement would make that the default behaviour for all tcp transports and allow it to be enabled when required via configuration: