[c-nsp] BGP Cease notifications with Graceful Restart
Phil Mayers
p.mayers at imperial.ac.uk
Tue May 21 12:27:21 EDT 2013
On 21/05/13 17:15, John Neiberger wrote:
> I guess that's what I get for skimming too quickly. :) thanks! I did
> find it interesting that during our testing we saw two different log
> messages on the receiving router:
>
> %ROUTING-BGP-5-ADJCHANGE : neighbor a.b.c.d Down - Peer closing down the
> session (CEASE notification sent - administrative shutdown) (VRF: default)
>
> %ROUTING-BGP-5-ADJCHANGE : neighbor a.b.c.d Down - Peer closing down the
> session (VRF: default)
>
> Sometimes we would see the first, other times we would see the second. I
> assume that in the first case we are, in fact, receiving a CEASE
> notification. In the second case, I'm thinking maybe the TCP session is
> simply being torn down but we have not received a CEASE message.
>
> Either way, it seems that it would be correct behavior to immediately
> remove that neighbor's routes from the routing table and BGP table,
No, tearing down the routes on a "bare" TCP session close is
unquestionably wrong IMO.
Think about a peer that has rebooted and is non-stop forwarding, but not
yet ready to re-open a BGP connection. If we send a keepalive (or any
TCP packet) in the window before the peer re-opens a new session to us,
we'll get a RST, and dropping the routes in that situation is wrong.
More information about the cisco-nsp
mailing list