[c-nsp] BGP - Older routes?

[nachocheeze at gmail.com]

On Feb 1, 2008 9:22 AM, Vincent De Keyzer <vincent at autempspourmoi.be> wrote:
> Hello list,
>
> I have a BGP router with 2 eBGP peers (upstreams). This morning one of
> the two upstreams (say A) had a scheduled maintenance, and most of the
> outgoing traffic went to B. But when A came back up 4 hours longer,
> outgoing traffic mostly kept going through B, and traffic towards A did
> not reach back its previous level. As a consequence, there currently is
> now too much traffic going to B (given the current CDR).
>
> I have no reason to believe that A has made any configuration change; so
> I believe this behaviour is due to "older route first" criterium of the
> BGP route selection algorithm (as described in step 10 of
> http://www.cisco.com/warp/public/459/25.shtml).
>
> Does that sound like a reasonable?

Yep, very much so.  I've seen this before many times.  If both your
eBGP upstreams are of the same tier level, more than likely this is
it.

>How do I check this is indeed the
> cause?

show ip bgp nei <x.x.x.x>

Check the remote router-id of both peers.  Bet you a dollar the
preferred route has a lower RID (assuming it's two different upstream
provider AS's).

>If it is, would a "soft in" clear be enough to fix the problem?

I believe you have to hard clear to have the BGP RIB re-learn that
same prefix and rerun the path selection process.  Even so, it'll
still come back at the next peering flap, so that's not a good
solution.  You should probably set up policy for your learned routes
to avoid this in the future.