If I/O fencing is started in disabled mode, and if you change the /etc/vxfenmode file later to scsi3, the vxfen driver throws the following spurious error when you stop vxfen:
However, the actual I/O fencing stop operation is successful. [1301698, 1504943]
Workaround: This error message may be safely ignored.
The fencing driver in 5.0 uses Veritas DMP to handle SCSI commands to the disk driver if fencing is configured in dmp mode. This allows fencing to use Veritas DMP for access to the coordinator disks. With certain disk arrays, when paths are failed over due to a path failure, the SCSI-3 persistent reservation keys for the previously active paths are not removed. If the nodes in a cluster are all rebooted at the same time, then the cluster will not start due to a Preexisting split brain
message. [609407]
Workaround: Use the vxfenclearpre
script to remove the keys from the coordinator disks as well as from the data disks.
Trying to stop the vxfen driver when the fencing module is being configured results in the following error.
VCS FEN vxfenconfig ERROR V-11-2-1013 Unable to unconfigure vxfen
VCS FEN vxfenconfig ERROR V-11-2-1022 Active cluster is currently fencing.
Workaround: This message may be safely ignored.
The vxfenconfig -c
command fails if any of the following commands are running on other nodes in the cluster:
Some options of the vxfenadm utility do not work well with DMP paths such as /dev/vx/rdmp/sdt3.
Workaround: Use the -a option to register keys instead of -m option for DMP paths.