[c-nsp] Best Practices for quickly removing routes when BGP peer drops

[Saku Ytti]

You need to explain more about what is happening.

When this issue happens, do you see the interface go down?

If you see the interface go down, you can trivially converge in
subsecond. If you don't see the interface go down, it's much trickier.

You need to solve two problems.
1. how to detect that something is broken
2. how to move to a next route


To Gert's point

a) fast-external-fallover is default, you can break it by doing
multipath (e.g. some configuration related to blackholing next-hop
manipulation or GTSM may break it)
b) you may not need 'add-path' to have backup, if you have full-mesh
'advertise-best-external' if enough. If you have RR you need both.



You definitely should target a solution where the backup route is
already in RIB at least, maybe even in FIB. =>
add-path/advertise-best-external/PIC edge.
Then the remaining problem is, how to learn that the best route is no
longer valid. With PIC Edge this doesn't matter much, as you do the
repair on the 'bad' edge during convergence. But if you don't want
that, then you might want to consider if you want to expose PE-CE in
IGP, which would be invalidated during interface down, which would
then immediately invalidate the best route across network, allowing
the RIB to swap to the next route, provided by
add-path/advertise-best-external. But PIC edge is much more elegant,
even faster and you don't need to pollute your IGP with PE-CE routes.

On Wed, 10 Dec 2025 at 16:32, Shawn L via cisco-nsp
<cisco-nsp at puck.nether.net> wrote:
>
> We currently have a couple of Internet feeds (full tables) and peering
> sessions to most of the major streaming networks, etc.  When our peering
> sessions go down for whatever reason (lately it's been fiber cuts) it takes
> a couple of minutes for our routers to process all of the routing changes
> and for streaming sessions to start up again.
>
> I'm wondering what others are doing to try to reduce this time.  In the
> past I was always told that reducing the BGP timers was a bad idea.  Is
> that still the case?  BFD?
>
> Thanks
>
> Shawn
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/



-- 
  ++ytti