[nsp] BGP route withdrawal bug?

[sam_ml at spacething.org]

Hi,

Was wondering if anyone had ever seen a C6500 (Sup2/MFSC2) behave like 
this before. First warning we had was when all of a sudden, without 
warning, the amount of routes it was advertising via iBGP dropped 
dramically.

Looked at the router in question and it appeared to not be having any
problems. However, when I did a 'sh ip bgp nei x.x.x.x routes' on a full
transit feed it only showed ~500 prefixes, but 'sh ip bgp sum | inc
x.x.x.x' implied it was still receiving a full table (~138,000 routes)

Doing a 'clear ip bpg x.x.x.x soft in' only took the 500 odd routes off
the local prefix counters, so when the routes started arriving again it
hit the prefix limit.

Only solution was to tear down the BGP sessions and re-establish them. 
Unfortunately I didn't check the smaller peering sessions with only a few 
routes (that might had been a good datapoint).

The router had at least 200MB of RAM free (~150MB in a single block), and 
only does BGP.

This is on 12.1(20)E

Anyone got any ideas what might had caused this?

Sam