I am seeing strange behavior from a 3850 stack running EIGRP to a couple of 4507's. Let me start out by saying that over the past couple of weeks, we have replaced 5 stacks using the same configuration template, code version, etc. with no issues. Before we execute the change to install these switches, we always run failover testing to prove routing. Last night, on the 6th switch stack, failover testing did not succeed. Maybe you can help me figure out why.
Here is what happened. I have a loopback configured on the stack. We connect the stack to each of the 4507's at the distribution layer and bring up an eigrp adjacency. I start a continuous ping from another site to the loopback. Then we pull the first link and the ping continues successfully. Plug it back in, bring up the adjacency. Then we pull the second uplink and the ping begins failing (TTL Lost in transit). The route at the source of the ping was lost so it was using the default route which led to nowhere. When I check the route against the table on the 4507's the subnet is not in the table but the adjacency is up.
I have attached a topology drawing of the relevant devices (as I see it). Again, we never experienced this issue with the 5 previous stacks which are all connected to the 4507's the same way.
