[c-nsp] BGP VPNv4 scanner wedgy

[Tim Durack]

On Thu, Jul 1, 2010 at 3:11 PM, Tim Durack <tdurack at gmail.com> wrote:
> Anyone run into BGP VPNv4 scanner wedgies? I appear to have 4 boxes,
> SUP720-3C running 12.2(33)SXI3 with this problem:
>
> RTR-1#sh bgp vpnv4 unicast all pending-prefixes
> ...
> 1050 pending network entries using 143850 bytes of memory
> These nets will be cleaned up by the BGP Scanner once all
> update-groups have been converged.
>
> Not sure what triggered it.  Any way to resolve this without a full reboot?

For the record, TAC said it smells like:

CSCsr62529 Bug Details
pending prefixes are not being deleted from sho ip bgp pending

Conditions:

Withdraws are not sent and updates are stuck when peers go down in large scale
scenario. Has been seen when there is a neighbor configured as admin down
(shutdown), or in idle state, from which the route was previously learned.

Symptoms:
bgp updates are not sent to peers, although most show commands signal that the
route is getting sent. An example scenario is that neighbor
default-originate does not send a default route if the 0.0.0.0/0
route is
stuck in the show ip bgp pending-prefix, although it
should ALWAYS
send a default route, in theory, to the configured neighbor.

Workarounds:
Deleting "idle" neighbors has been shown to clear up some issues. Routes may
clear from "pending" state by deleting neighbors that are in a shutdown state.


Affected routers had both an idle and a shutdown BGP peer. Removing
both has resolved the problem.

-- 
Tim:>