CDP data collected from a Cisco Switch Cluster commander, the stack members will appear as having the same IP address as the commander (which they in effect do). ipdevpoll will, however, identify the neighbor as the box itself based on that IP, and ignores the cdpCacheDeviceID field.
This results in adjacency candidate records that indicate self-loops. Self-loops seem to cause the topology analyzer to crash. The topology analyzer should probably have a more reasonable fail-safe for a situation like this, but the core problem can be fixed in the CDP cache parsing:
If a CDP neighbor record indicates the same IP as the source netbox, we should treat it the same as if there was no IP address there. This will cause the algorithm to move on to cdpCacheDeviceID for neighbor identification. A stack member will normally have a member index suffixed to its sysname, which makes it unidentifiable by NAV, and will cause an uncrecognized_neighbor entry to be generated instead. This is acceptable as long as we have no NAV support for switch clustering.
attempted fix here: https:/