Today we've been testing the service, and almost got to the approach we're looking to but we've some questions and we'd to have someone to shed some light on this one. We got the following scenario running: 
We got a few hosts, with a service deployed exposing a port on every host, we scaled it up and down and everything went perfect.
However, we're thinking how might we develop a Zero-Downtime deployment and taking advantage of compose we thought of the following: 
We have two versions (one being rolled out and current) of the same service, and once the new version is warmed up and running we'd love to switch on LB from Backend:3000 to Backend:4000. All this on an automated fashion, but we can take care of that.
Moreover we ideally we'd love to do path routing as follows 
We'd love to have multiple micro services on different ports, and path-based routing for each one of the services. And keep the scheme described above (two versions, swap ports at LB rules when deploying a new version).
We have been researching on Application Gateway however we don't find an easy way of doing this, and we couldn't find more docs on this. Ideally our VMSet for Agents would serve as backends, and each one of the micro services could be treated as backend rules
(at the app gateway level, path to port relationship). But I still don't clearly see, if this is the appropriated technology to use (App Gateway) or what other options do you suggest).