Automatic Learning of Repair Strategies for Web Services

The process of repairing Web Service failures may be connected to the nature of the fault that caused the error generating the failure. The selection strategy for composed services repair may be drawn from an analysis on temporal behavior of the fault, assessing if fault is transient, intermittent or permanent. The repair process strictly depends on the permanence type of faults, as substitution is applied with permanent faults, while retry is chosen with transient faults and the retry period is to be determined. In this paper we propose a methodology and a tool for learning the repair strategies of Web Services to automatically select repair actions. This methodology is able to incrementally learn its knowledge of repairs, as faults are repaired. Thus, it is at runtime possible to achieve adaptability according to the current fault features and to the history of the previously performed repair actions. This learning technique and the strategy selection are based on a Bayesian classification of faults in permanent, intermittent and transient, followed by a comparative analysis between current fault features and previously classified faults features which suggests which repair strategy has to be applied. Therefore, this methodology includes the ability to learn autonomously both model parameters, which are useful to determine the fault type, and repair strategies which are successful and proper for a particular fault.