The topology is as follows:
Beijing (as1) Tianjin (as2)
J Corporation M160----C corporation 12416
BGP is operated on four routers and they belong to one AS. ISIS is enabled in internal AS. M160 and 12416 act as RR and they configure next-hop-local. NE80E and NE80 are clients of JM160 and 12416. The destination of the topology is on NE80. The route to Tianjin passes 12416 and the route to Beijing passes NE80E. On NE80E the route to Tianjin passes NE80 and the route to Beijing passes M160. The topology changes cost of IGP to realize the aim. In fact it can be realized on NE80, but routes to Beijing and Tianjin on NE80E pass M160.
Change M160 configuration. When it distributes the route to the client, it does not change next-hop of the route learnt from another RR. The problem is solved.
1. Check BGP routing table on NE80E. Analyze the route to Tianjin. It is learnt from M160 and 12416, but the route sent by M160 is chosen first. The next-hop is loopback address of M160. Check two learnt routes and it is found that the next-hop learnt from M160 changes and points to loopback address of M160.
2. Check BGP routing table on NE80 and analyze the route to Beijing. It is learnt from M160 and 12416, but the route sent by NE80E is chosen first. Check two learnt routes and it is found that the next-hop learnt from 12416 does not change and it still points toloopback address of M160. 12416 does not change next-hop when it receives the route from another router and reflects the client.
3. It is confirmed that M160 can change next-hop with command when it receives the route from another RR and sends it to the client.