Dunno if this has happened to others before, but in case not:
I have a setup where 3 routers (ciscos, running 9.1.x if that's important)
are running IBGP.
A---------B-----------C
>
ext-net
Router-A peers with 'B' & 'C', Router-B peers with B & C and C with B&A
(a full IBGP mesh). Router A is an exit/border router. In addition to BGP,
the routers run IGRP. The BGP routes are kept separate from
the IGRP routes.
For whatever reason (not a configuration error), the BGP session between
A-B died but not between A-C, B-C.
The result was that C 'knew' via the BGP session with A, that to
get to external network 'ext-net', it should send traffic to A. It does
the IGRP lookup to get to A, and sends the traffic to 'B'. However,
since B did not have a BGP session with A, it did not know how to get
to the ext-net (in the real case, B was sending traffic back to C since
the default route was via C).
A potential disaster!! I realize that the problem was the unexpected
'lost' BGP session between A-B, but are we sure that in a more
complicated topology, something similar cannot happen ?
What really make me worried is that there doesn't seem to be way
to detect this kind of a problem easily.. how do others monitor
their BGP sessions ?
-vikas
vikas@jvnc.net