In the event of a detected network partition, the distributed bean manager's Infinispan cache behaves in one of 3 ways:

1. Disallow both reads and writes from/to a SFSB in all partitions
2. Allow reads from a SFSB in all partitions, but only allow writes to a SFSB if the partition contains all data owners.
3. Allow both reads and writes to/from a SFSB in all partitions

For customers who opt for #3 (i.e. maximum availability), in the event that a SFSB is updated in both partitions, they need be able to provide a mechanism to resolve conflicts when the partitions eventually merge.

Something like:

public interface BeanInstanceMergePolicy<T> {
    T merge(T primaryPartitionBean, Iterable<T> minorityPartitionBeans); 
}