Also the control VM of the ULDR is not available since it has an IP from only one POD.
Check out the Control VM Failure & Recovery section on page 132 of the VMware® NSX for vSphere Network Virtualization Design Guide ver 3.0
At a high level, floating static routes on the ESGs for the subnets behind the DLR will allow ingress traffic to keep flowing even after the routing adjacencies are lost.
We eventually create an VPN beteen sites to achieve L2. So we could go on with a streched cluster between the sites.
In case of a datacenter failure the gateway of this segment only needs to be changed.