N8000 Cluster Abnormal When Deleting Coordinator Disks Without Disable Fencing

Publication Date:  2012-07-18 Views:  133 Downloads:  0
Issue Description
In N8000 system, you can’t use the console IP to log into the system and the cluster status is abnormal. Two nodes can not form one cluster.
Alarm Information
Handling Process
1)      Stop VCS and IO Fencing :
Cer_02:~ # hastop –local
Cer_02:~ # /etc/init.d/vxfen stop
Stopping VxFEN:
2)      Delete keys on each nodes
Cer_01:~ # cd /opt/VRTSvcs/vxfen/bin/
Cer_01:/opt/VRTSvcs/vxfen/bin # ./vxfenclearpre
3)      Modify vxfenmode configuration file
Cer_02:~ # cp  /etc/vxfen.d/vxfenmode_disabled  /etc/vxfenmode
4)      Modify main.cf , delete all “UseFence = SCSI3”
Root Cause
1) when N8000 system enable fencing function, the coordinator disks will be used to find which nodes form one cluster, it can keep the cluster normal when the brain-split happens.
2) In this case, coordinator disks are deleted without disabling fencing function, so the cluster can’t recognize which is the master node, and see the system as two separate ones.
1)      Do not delete coordinator disks when enabling the IO Fencing