Spit-brain handling may be performed by the following element of the Shutdown Facility:
The Shutdown Facility internal algorithm
This method uses the node weight calculation to determine which sub-cluster is of greater importance. The total node weight is equal to the value of the defined Shutdown Facility node weight added to the total application weight of the Online applications for this node as calculated within RMS.
SF internal algorithm
When the SF is selected as the split-brain resolution manager, the SF uses the node weight internally.
The SF on each cluster node identifies which cluster nodes are outside its sub-cluster and adds each one of them to an internal shutdown list. This shutdown list, along with the local nodes node weight, is advertised to the SF instances running on all other cluster nodes (both in the local sub-cluster and outside the local sub-cluster) via the admIP network defined in the SF configuration file. After the SFs on each cluster node receive the advertisements, they each calculate the heaviest sub-cluster. The heaviest sub-cluster shuts down all lower weight sub-clusters.
In addition to handling well-coordinated shutdown activities defined by the contents of the advertisements, the SF internal algorithm will also resolve split-brain if the advertisements fail to be received. If the advertisements are not received then the split-brain will still be resolved, but it may take a bit more time as some amount of delay will have to be incurred.
The split-brain resolution done by the SF in situations where advertisements have failed depends on a variable delay based on the inverse of the percentage of the available cluster weight the local sub-cluster contains. The more weight it contains the less it delays. After the delay expires (assuming the sub-cluster has not been shut down by a higher-weight sub-cluster) the SF in the sub-cluster begins shutting down all other nodes in all other sub-clusters.
If a sub-cluster contains greater than 50 percent of the available cluster weight, then the SF in that sub-cluster will immediately start shutting down all other nodes in all other sub-clusters.
