A split-brain condition is one in which one or more cluster nodes have stopped receiving heartbeats from one or more other cluster nodes, yet those nodes have been determined to still be running. Each of these distinct sets of cluster nodes is called a sub-cluster, and when a split-brain condition occurs the Shutdown Facility has a choice to make as to which sub-cluster should remain running.
Only one of the sub-clusters in a split-brain condition can survive. The SF determines which sub-cluster is most important and allows only that sub-cluster to remain. SF determines the importance of each sub-cluster by calculating the total node weight and application weight of each sub-cluster. The sub-cluster with the greatest total weight survives.
Node weights are defined in the SF configuration file rcsd.cfg.
Application weights are defined in RMS. Each RMS userApplication object can have a ShutdownPriority defined for it. The value of the ShutdownPriority is that application's weight. RMS calculates the total application weight for a particular node by adding up the weights of all applications that are Online on that node. If an application is switched from one node to another, its weight will be transferred to the new node.
SF combines the values for the RMS ShutdownPriority attributes and the SF weight assignments to determine how to handle a split-brain condition.
RMS supports the ability to set application importance in the form of a ShutdownPriority value for each userApplication object defined within the RMS configuration. These values are combined for all userApplication objects that are Online on a given cluster node to represent the total application weight of that node. When a userApplication object is switched from one node to another, the value of that userApplication objects ShutdownPriority is transferred to the new node.
The higher the value of the ShutdownPriority attribute, the more important the application.
The Shutdown Facility supports the ability to define node importance in the form of a weight setting in the configuration file. This value represents a node weight for the cluster node.
The higher the node weight value, the more important the node.
Note
Although SF takes into consideration both SF node weights and RMS application weights while performing split-brain handling, it is recommended to use only one of the weights for simplicity and ease of use. When both weights are used, split-brain handling results are much more complex.
It is recommended that you follow the guidelines in "7.3.4 Configuration notes" for help you with the configuration.
