[c-nsp] Unusual EIGRP flap on 3750 stack...

[Jeff Kell]

Have a very unusual issue I'm at a loss to explain (though it is
reproducible at this point)...

3750 stack (4x48-ports used as top-of-racks) attached to a ring/star
layer-2 mesh of other cluster nodes and central core.

Running VRF-lite, about 8 VRFs active plus global, connected to
ring/star via 802.1Q trunk, one vlan per VRF backbone.

VRFs are running EIGRP, there are some redundant L3 links for
load-balancing by bandwidth, standard hello/hold-time (5/15 secs).

Somewhere in the process of adding a vlan and creating an SVI (w/ip vrf
forwarding ...), the 3750 stack "flapped" with respect to several EIGRP
neighbors, for several VRFs, the neighbors logging a hold-time exceeded,
followed shortly thereafter by a new neighbor established.  The 3750
itself logged hold-time exceeded for several neighbors, followed by new
neighbor established.

The added vlan was not allowed on any trunk, especially not the backbone
ring connection.

We checked spanning tree, but doesn't appear to be involved (and
downtime was shorter than 30 seconds, assuming the 15 second holdtime is
what triggered the route flap).

Just for "grins" trying to tie the flap to the added vlan, we did a "no
vlan xxx" to remove it.  Another flap, on multiple VRFs.

We add vlans (local to the routed clusters) quite often, and I don't
recall this ever happening before, here or any other node.

Ring any bells?  Things seem to be stable after the flap, so I can live
with it, until the next new vlan request at least.

Jeff