I'm a little curious why in this case one wouldn't choose to do "RPC over HTTP". Assuming the following points from the post are correct: a) the perf problem

I'm a little curious why in this case one wouldn't choose to do "RPC over HTTP". Assuming the following points from the post are correct: a) the perf problem is YAML marshalling b) that can be solved via a go rewrite then perhaps an easy way to solve "I want to call the old code" is to access it over HTTP, using a cheaper marshalling method than YAML. I guess the tradeoff is whether you'd prefer an AMQP broker in there mediating the requests or an HTTP load balancer (or http client pool code in the go app). And whether you see other uses for that broker in your setup.