I have a working prototype for replacing the dynamically created response queues/exchanges with a single queue per caller process. It does fix the scaling problem in a non-mirroring RabbitMQ clustered environment. It uses a single thread for receiving responses (consume_in_thread()) and a "call_waiting" dictionary as Vish suggested for correlating the message IDs with the waiting greanthread. I need to do some more testing and cleanup, but I would like to know what the process to use for delivering this patch. Should I report it as a (performance) bug? Or should I open a blueprint?<br>
<br>Thanks,<br>Ray<br>