We have message bus working with ALBs, in fact meta is is hosted on AWS.
My guess is that somewhere in your stack you have buffering on requests enabled which is causing them to be held for long periods of time.
It is most likely proxy_buffering in NGINX being set to on instead of off