BGP over VPC on NEXUS

Question

here is my network topology:

i have a customer which is connected to my top of rack switch and my top of rack switch has 2x 10g LACP towards my both Cisco Nexus 3k and both of my Nexus 3k has VPC PEER link, here is customer router interface vlan config :

interface vlan542
IP address 192.168.20.1/30

then my customer will set 192.168.20.2/30 in own router and now we want to establish bgp from customer router towards n3k, but if we do this then bgp flap occur, because sometimes bgp request send to first n3k and sometimes send to second n3k, what is the solution for this? also i know one of the solutions is to connect my customer router directly to both of my n3k then establish 2 different bgp or form LACP or such things, but i want to know if we can find any solution in this senario. Thank you.

Answer 1

The only circumstances under which you'd have the same address on two different boxes are various flavors of anycast and HSRP/VRRP. This definitely does not include routing protocols. In your scenario the customer router would need to have sessions to both 3K's and, of course, the 3K's would need to have distinct address. This would likely push you to expand the local subnet from /30 to /29. This is also necessary if you want to run HSRP, as both 3K's would still need to have unique IP's as well as the shared address.

Keep in mind that VPC's purpose is to provide L2 resilience. It's absolutely possible to pass routing protocols over VPC, but in certain situations it doesn't make much sense. In this instance, for example, if the purpose is just to establish BGP connections then the utility of VPC is sort of questionable, as BGP will already be handling ECMP, fail-over, etc.

VPC makes a lot more sense when the downstream device has a static route - at which point VPC and HSRP provide a lot of value (...much of said value being no need for setting up routing protocols on the customer router). You'd likely be better off either standardizing on /29's for the VPC use-case or deploying individual p-t-p /31's on a trunk to the user routers and then turning up routing protocols.