On Sun, Nov 02, 2008 at 08:14:43PM +0100, Falk Brockerhoff wrote: > Hi, > > I have to routers running 4.3 GENERIC#826 i386 and 4.2 GENERIC#476 i386. > On both routers I runs a BGP session to the same Juniper Router. Last > weekend there was a configuration change on my neighbor's side: it would > not accept any prefix more or equal (!) specific to /24. > > Except for the not really wanted not-accepting of /24 prefixes this > should not have any other consequences. BUT both of my routers stopped > announcing the rejected prefixes to _any_ of their neighbors. It seems > that the openbsd bgpd process takes the information notice, that the > prefixes are rejected by one neighbor, as a command. > > I have checked it twice, I have no community-based actions configured > which can cause this behavior. In fact, I have none communities or other > filters defined. Expect of these (standard) ones: >
This should not happen. There is now way a neighbor knows what the other side rejects (unless a bgp notify happens and the session is closed). Announced networks are always valid and the only way somthing like this can happen is that a other route is selected best for some strange reason. If you can reproduce the situation please include all the RIB information for the prefix: bgpctl show rib <address> bgpctl show rib detail <address> bgpctl show rib neighbor <peerip> out <address> bgpctl show rib neighbor <peerip> out detail <address> -- :wq Claudio
