[c-nsp] BGP Path Selection and next-hop reachability (IGP vs BGP)

[Phil Mayers]

On 12/02/2012 11:51 AM, Gert Doering wrote:
> Hi,
>
> On Sat, Dec 01, 2012 at 10:37:36PM +0000, Oliver Boehmer (oboehmer) wrote:
>> Ack.. BGP in IOS will use whatever valid route to resolve the next-hop,
>> including a BGP route (which could be perfectly valid in some specific
>> scenarios).
>> You need to configure selective address tracking to avoid this to happen.
>> With an appropriate "bgp nexthop route-map .." config, for example only
>> considering loopbacks or only OSPF or connected routes, BGP would consider
>> 5.5.5.5 unreachable, and ignore this path for the best-path calculation.
>
> Mmmmh.  This sounds like interesting stuff, but unfortunately, my Google-fu
> is failing me today - all I turn up is the normal "set ip next-hop..."
> stuff inside "neighbour 1.2.3.4 route-map foo".
>
> Oli, can you point us to some documentation and/or examples for this?

This is handy:

http://stack.nil.com/ipcorner/DesigningBGPNetworks/#chapter4

Basically it's:

neighbor x.x.x.x fall-over route-map <somthing matching loopbacks>

FYI it was absent from 6500 IOS at one point; don't know if it's made it 
in yet. Cisco gave me an absolutely absurd "3 years" time-to-implement 
when I asked them about it!

Another option is to use EEM to track the /32s and issue a "clear ip bgp 
nei". This is, in some ways, superior, because you can use a track 
object with delay to give it a bit of wiggle room. Downside is the extra 
config, plus EEM (yuck).