Networking:Two S9312 uplink with NE40 via two links independently.The two S9312 interconnect with each other and enable VRRP. VRRP configuration only track uplink ports.
The S9312-1 is master and the S9312-2 is standby,downlinked with an array load balance device.This device has two network cards which link with S9312-1 and S9312-2 respectively and the configurations are as below: (i)The VRRP virtual address is the gateway address.（ii）The interface ip address is 10.96.170.252（iv）The virtual ip address is 10.96.170.251 which is used by the service.
Symptom:When the fiber link between the S9312-1 and the load balance device breaks down,the service switchover to the standby link.But when the fiber link is repaired,the service cannot restore to the master link in time.We find:(i)Ping 10.96.170.252,the reply is normal;(ii)Ping 10.96.170.251,it will take 90 seconds to get the reply and during this time, the service of load balance device is interrupted.
1. Switchover again to repeat the trouble phenomena.At S9312-2,process the port mirroring and capture the bidirectional packets at the port which links with the S9312-1.The result indicates the VRRP doesn’t swichover,however, the interaction packets of VRRP is normal.It means the VRRP is normal.
2. Switchover one more time.Process the port mirroring and capture the packets at both S9312-1 and S9312-2.At S9312-1,capture the packets which link with load balance device and S9312-2,naming it as ”master”;At S9312-2, capture the packets which link with load balance device and S9312-1,naming it as ”standby”.Analyze the result,we find that:(i) When the fiber link between the S9312-1 and the load balance device breaks down,the S9312-1 can get the arp information of 10.96.170.251 and 10.96.170.252 via the port which links with S9312-2(ii)When the fiber link is repaired,S9312-1 can get the arp information of 10.96.170.252 through the port which links with load balance device.However,the S9312-1 get the arp information of 10.96.170.251 through the port which links with S9312-2
3. Replace the S9312 with Cisco and H3C switch,the trouble is cleared.
4. Comparing the arp learning mechanism among these three kinds of switch,we find:(i)For Cisco or H3C switch,the arp table can be refreshed based on the “flow”.As soon as the flow passes through the device,the arp table will be refreshed.(ii)For S9312,only when the arp interacting information exist can S9312 refresh the arp table.There is one hop (10.96.170.252)between SS9312 and 10.96.170.251.The S9312 cannot refresh the arp table if 10.96.170.252 doesn’t send out arp information.At this time, the load balance service use 10.96.170.252 and the master port status is up whereas the standby port doesn’t process the packets.The arp itme of 10.96.170.251 failure result in the service interruption.
5. The virtual ip of load balance device doesn’t initiate arp request packets,however,replies the arp request.In addition,the arp-aging mechanism of S9312 is different with other device.These two result in the trouble .There is a command configured at global mode: l2-topology detect enable.It can enable S9312 L2 topology detection function to resolve this trouble.
1. The VRRP of S9312 is abnormal
2. The array load balance device is abnormal
3. S9312 arp is abnormal
The S9312 cannot refresh the arp table automatically when the arp item exists already.Thus,we can use “l2-topology detect enable” to refresh the arp table in time when the layer 2 interface change the status from up to down.This command can refresh the arp table of the vlan assigned for the l2 interface.Warm tips:It will take a long time to relearn the arp if the vlan of l2 interface has a large arp table.