Issue with MPLS VPN default routes, using EIGRP between CE-PE

[dtangent]

A customer wants to do a "hub and spoke" overlay on MPLS VPN. Currently, they have two hub routers, each passing a default route (using EIGRP summary-address command) to the spoke.

After switching over to MPLS VPN, they are experiencing problems in that when hub1 goes down, all traffic flows to hub2, but when hub1 comes back up, traffic will still contintue to go to hub2.

Topology:

hub1 and remote router both connected to PE1, hub2 is connected to PE2.

When everything is up, show ip route on the PE1 sees the default route learned via EIGRP. However, show ip bgp vpnv4 shows only the locally redistributed default route, and NOT the default route learned from PE2 via IBGP. I would expect to see that route in the BGP table and not used, but it doesn't show up at all.

When you take the link to hub1 down, PE1 installs the default route learned from IBGP and puts it in the route table. That's strange because the route never showed up in the BGP table to begin with.

When link to hub1 is restored, PE1 still insists on using the default route learned via IBGP, when the default route learned via EIGRP should have kicked it off with its lower AD.

[Harold Ritter]

The reason why you didn't see the default route coming from the PE2 is that it is probably redistributing the default PE1 is sending into EIGRP.

Make sure that the default route is not redistributed in EIGRP.

Hope this helps,

[richard.harvey]

I agree that 0/0 should not be redistributed back in to EIGRP. However, to achieve failover/failback functionality between the sites you need to redistribute the default route into BGP with a different metric at each PE.

This will always give priority to Site1 over Site2 when they are both up.

[dtangent]

I am not sure I follow what you guys are saying. On PE1, it will get two default routes: one coming from hub1 via EIGRP, and one coming from PE2 via IBGP. PE1 will install the EIGRP default route because of better AD. The default route will then be redistributed into BGP and send to PE2. The default route PE1 receives from PE2 via IBGP will probably be in the BGP table (I have both cases), but since locally generated IBGP default route on PE1 (from the EIGRP->BGP redistribution) has a better weight anyway, the default route sent by PE2 will not be installed in the VRF IP route table, and thus not be redistributed back into EIGRP toward hub CE. Besides, the EIGRP default route has better AD so the IBGP route will never make it to the VRF IP route table, regardless of how favorable the metric is.