[c-nsp] eBGP multihop, link failure, and multi-path IGP (OSPF)

[Rick Ernst]

We have some eBGP neighbors that have their peering session reset in the
case of link failure (root-cause analysis and problem resolution as a
separate subject).  The peers are connected via loopback interfaces and
multi-path OSPF.

bgp fast-external-failover is supposed to be used for directly connected
eBGP peers, but it seems like a link failure on a pair of redundant
(layer-3) links is also causing the peer to go down:
Nov  1 11:33:12 10.56.205.1 %OSPF-5-ADJCHG: Process 1, Nbr a.b.c.d on
FastEthernet8/0/0 from EXSTART to DOWN, Neighbor Down: Interface down or
detached
Nov  1 11:33:12 10.56.205.1 %BGP-5-ADJCHANGE: neighbor w.x.y.z Down
Interface flap

The destination to the peer is still in the FIB, and the peer comes back up
almost immediately (in this case, about 15 seconds).

I'm considering disabling fast-external-failover, but want to better
understand the event.  The eBGP peer is not "directly connected" on the
interface. It is reachable via a loopback peering IP with multi-path OSPF.
Is this expected behavior (any link with a route to the destination going
down will cause the session to go down)?


Any gotchas with disabling fast-failover?

Thanks,