ARP not learning properly while cluster failover

[Jerald Simon M.]

I have a 6500 serious switch that one port (portfast enabled) hold a  server .This server has sort of a Cluster configured for high availability .

During primary server failure the failover server acquires the cluster IP address (ie virtual IP).

Cluster IP 192.168.10.1

Primary server =192.168.10.2  Failover server =192.168.10.3

when failover happen , i am unable to learn arp. need to reboot the server for arp . After reboot i am getting different mac address(secondary) with cluster IP.

query:-normal clustering how mac address generating(means logicaly any mac address will be created or LAN card mac address)? 

Is ther any issue with portfast configuration(with out portfast configuration TCN will generate and max age timeout decrease 15 second from 300second) ?

how can avoid this issue ?

Note : we checked a scenario same ip address  with two system(ipconflit) and removed arp learned system from network and same thing happen we need to clear the arp manually in L3switch

[pille1234]

Hallo,

portfast should not be an issure, as a TCN will only be generated if the link actually goes down. The server may however fail without changing the link state. Even if a TCN was generated, this would only flush the MAC address tables and not the arp cache on the gateway.

How is the failover supposed to work? Is there a virtual MAC corresponding to the cluster IP or do the servers use their burnt-in MAC addresses? Either way, the network needs to learn that there is a new server taking over the cluster IP. This is done with a gratuitous arp sourced by the backup server usually.

Try to find out the theory behind this cluster configuration, then we can try to check why it is not working.

Regards

.

[Jerald Simon M.]

Thanks for your reply .When ever failover happen in cluster i am not getting ip with mac(ARP is not updating) so we can conclude that gratuitous arp not getting secondary server. How can check gratuitous arp coming from neighbor device (from secondary server) ?

in the second scenario(Note:we checked a scenario same ip address  with two system(ipconflit) and removed arp learned system from network and same things happened we need to clear the arp manually in 6500 switch) how the switch will react normal.

[pille1234]

I'm afraid the situation is too foggy to help you. We need to know the theory behind this failover construction. Find the product documentation and tell us what failover variations does it offer and how do they work.

How is the standby server supposed to realize that it should become active? What is it supposed to do then? Is there something like a heartbeat between the cluster members? Is it working with multicast (like Microsoft NLB)? There are too many variations out there...