NVGRE - How do publish servers internally when they are behind an NVGRE Router?

I'm looking to span 2 server rooms across my network. I get how outgoing traffic from the Hyper-V hits the GW and gets nat'd from the CA to PA.

What I don't get is how does external machines access the virtual machines when they initiate the conversation?

in my scenario I want one NVGRE router at each server room with a site-to-site vpn to span the network. (I'm assuming)

Lets say I span 10.20.0.0/24 across the L3 network so servers at each site can use the network. Now lets say I put an application on the network that network with a 10.20.0.10/24 CA IP.

How do my clients on the know how to reach this application server and how do they now which server room to connect to?