There is no standard model at the service layer. However, fault management to distributed services and applications needs to construct and utilize complex models of the participating objects and their interdependencies, Thus, model-based fault management tools can predict the correct behavior of diagnosed systems and use the resulting predictions to identify faults. When used on-line in real systems, diagnostic tools based on such models should be able to provide prompt response and accurate, comprehensive explanations of the root causes of faults, In this paper we propose to address these requirements: modeling, proactive diagnosis, and explanation. We apply a recent extension to the constraint satisfaction paradigm, called composite constraint satisfaction, to facilitate modeling of complex systems, and we use constraint propagation techniques to support proactive diagnosis and explanation. We demonstrate the applicability of our approach on an example of a basic groupware service, namely, distributed database replication.