If the network links between the sites are disrupted, the application environments may continue to run in parallel, and this may lead to inconsistencies between the disk group configuration copies at the sites. If the parallel instances of an application issue writes to volumes, an unrecoverable data loss may occur and manual intervention is needed. To avoid data loss, it is recommended that you configure the VCS fencing mechanism to handle network split-brain situations. When connectivity between the sites is restored, a serial split-brain condition will be detected between the sites. One site must be chosen as having the preferred version of the data and the disk group configuration copies. The data from the chosen site is resynchronized to other the site. If new writes are issued to volumes after the network split, they are overwritten with the data from the chosen site. The configuration copies at the other sites are updated from the copies at the chosen site.

At the chosen site, use the following commands to reattach a site and recover the disk group:

# vxdg -g diskgroup -o overridessb reattachsite sitename
# vxrecover -g diskgroup

In the case that the host systems are configured at a single site with only storage at the remote sites, the usual resynchronization mechanism of VxVM is used to recover the remote plexes when the storage comes back on line.