08-17-2016 11:58 PM
In the event of a spilt-brain scenario, all the private network links are down and the cluster nodes are unable to communicate with each other. As a result, the nodes fail to know if the active node is still alive and whether or not the application is online.
In such a scenario, a single VCS cluster forms separate sub-clusters, and the nodes in each sub-cluster considers the other sub-cluster has faulted. Each sub-cluster tries to carry out recovery actions for the departed nodes. As a result, more than one node tries to import the same storage (VxVM LUNs) and may cause data corruption, have an IP address up in two places, or mistakenly run an application in two places at once.
Node membership arbitration guarantees against such spilt-brain conditions. The membership arbitration works differently on UNIX and Windows platforms.
For details about the platform-based VCS behavior during membership arbitration, refer to the infographic here: