This type of clustering is not very well covered in the literature, but it has been in use in clustering technologies for the last twenty years. The key element is to divide the resources protected by a cluster into independent groupings called hierarchies. Each hierarchy should be capable of operating independently without needing any other resources than those which it already contains. When an event occurs causing the cluster to re-establish, each node calculates, for each hierarchy, based on the then available communications information whether it is the current master (i.e. it has lost contact with all nodes whose priority is higher for that hierarchy). If the node is the current hierarchy master, it immediately begins a recovery. In order to prevent contention, each hierarchy must contain one own-able resource (usually disk resources), ownership of which must be acquired by the node before hierarchy recovery may begin. In the event of an incomplete or disrupted communications channel, the nodes may race for ownership, but only one node will win out and recover the hierarchy.
The essential operational feature is that no notion of a ``complete'' cluster need be maintained. At recovery time, all a node needs to know is who is who is preferred over it for mastering a given hierarchy. Operation of a resource driven cluster doesn't require complete communications (or even any communication at all) since the ownership of the own-able resources is the ultimate arbiter of every hierarchy.