Key learning:

Routing black hole
Issue recap:

 

Inside customer experienced network slowness issue after sw02 shut down for maintenance. Before that, two c3745 routers are deployed adhering to site backbone switches which cause the EIGRP routing re-calculation. From WAN router perspective, two next hops were added on. When routing go through the exterior switches, the routing can move into inside network as normal, but if the routing go through new landing routers, the routing will stuck there. Theoretically speaking, there will be two routes for c3745 to inside network. One is from vlan 900, a static routing, the other is from vlan 2/300, an EIGRP routing. Due to the routing AD, the routing from vlan 900 is selected into route table. But there is no down link to inside network at c3745, a routing black hole appears. Overall, half traffic will be dropped, that’s the reason the inside customer meet slowness issue.
Another thing we should take care is SVI, switching virtual interface. The specific vlan SVI can be active either there is active client in this vlan, or this vlan is in a active trunk. In this case, vlan 900 was active since in a active vlan, which make the routing available with next hop vlan 900.
 
Question:
The trigger of this issue is shutting down sw02, that’s the weird thing can not explain. :(