A dependency mechanism and service supervision are quite useful; the various cluster stacks such as Pacemaker implement this at a multi-node level, and are sometimes (not incorrectly) described as a "cluster-wide init".
A move to an event-based system is very desirable.