Automatic generation of predictive monitors from scenario-based specifications.

Abstract Context Unpredictability and uncertainty about future evolutions of both the system and its environment may easily compromise the behavior of the system. The subsequent software failures can have serious consequences. When dealing with open environments, run-time monitoring is one of the most promising techniques to detect software failures. Several monitoring approaches have been proposed in the last years; however, they suffer from two main limitations. First, they provide limited information to be exploited at run-time for early detecting and managing situations that most probably will lead to failures. Second, they mainly rely on logic-based specifications, whose intrinsic complexity may hamper the use of these monitoring approaches in industrial contexts. Objective In order to address these two limitations, this paper proposes a novel approach, called PREDIMO ( PREDI ctive MO nitoring). The approach starts from scenario-based specifications, automatically generates predictive monitors called MA s (Multi-valued Automata), which take into account the actual status and also the possible evolution of both system and environment in the near future, and enables the definition of precise strategies to prevent failures. More specifically, the generated monitors evaluate the specified properties and return one of the seven different values representing the degree of controllability of the system and the distance of the potential incoming failure. The translation from scenario-based specifications to MA s preserves the semantics of the starting specification. Method We use the design and creation research methodology to design an innovative approach that fills highlighted gaps of state-of-the-art approaches. The validation of the approach is performed through a large experimentation with OSGi (Open Service Gateway Initiative) applications. Results We present a novel language to specify the properties to be monitored. Then, we present a novel approach to automatically generate predictive monitors from the specified properties. Conclusion The overall approach is tool supported and a large experimentation demonstrates its feasibility and usability.