[c-nsp] (BGP identifier wrong) error on majority of ebgp peers

[Andy Davidson]

Hi,

I have a Cisco 6509 with SUP720-3BXL.  It has over a hundred bgp peers
configured, two full tables, 4 ibgp, several peerings at an IXP.

Seemingly without a config change, there are some sessions which refuse to
establish, because of a bgp notification :

%BGP-3-NOTIFICATION: received from neighbor XXX 2/3 (BGP identifier wrong)
4 bytes XXX

The router-id has not been changed - it was using the address from Loopback
0.  The router-id *is* unique, and the remote side of the peering is using
a different address.

We tried modifying the local router-id to match an address used on a
point-to-point link and clearing all sessions - same error.

This router was loaded with SXF3, moved to SXF16, moved to SXI3 - same
error seen with all three.

We removed the router bgp xx config and pasted it back in - when the
sessions established the same error was noticed.

There appears to be no commonality (eg. linecard) between the sessions
which establish, and those which don't - for example, some over the shared
ixp link establish fine, and some do not due to this router-id clash
notification.

It is always the same sessions which either come up, or refuse to come up
each time.  The ibgp sessions always establish (fortunately).

Has anyone seen this before ?

Thanks
andy