Towards efficient measuring of web services API coverage

Abstract

We address the problem of interface-based test coverage for Web services. We suggest an approach to analyze the Application Programming Interface (API) of Web services, calculate the number of possible input combinations and compare it to the number of actual historical invocations. Such API coverage metrics are an indicator to which extent the service has been used. Measuring API coverage is a key concern for assessing the significance of Verification and Validation (V&V) techniques; on the other hand, API coverage metrics can also yield interesting usage reports for a service-based system in production use. The coverage metrics rely on the exact specification of service interfaces, and we provide a mechanism to specify restrictions for data types in the Java Web services framework (JAX-WS). As full enumeration of all possible inputs is often infeasible, we allow the definition of custom coverage metrics by means of domain partitioning: the user divides domain ranges into subsets, and a coverage of 100% means that the logged invocations contain at least one sample for each subset. Based on a prototype implementation, we evaluate different aspects of our approach.