Testing Adaptation Policies for Software Components

Self-adaptive systems have to implement adaptation policies described by sets of rules that express how the components are reconfigured within the system, the priority of a given reconfiguration to happen, when a given (sequence of) event(s) occurs, and when specific conditions on the system state are satisfied. However, when this priority is given by a fuzzy value (e.g., high, medium, low) depending on external and internal events, it has to be implemented inside the software with particular implementation choices made. This paper is dedicated to the validation of adaptation policies, using a model-based testing approach and a verdict establishment that is based on both the runtime verification of temporal properties, and the detection of inconsistencies between the adaptation policy and the reconfigurations implemented in the self-adaptive system. We propose means to establish a test verdict based on the respect of the adaptation policy by the implementation, along with coverage measures of the rules. This provides interesting feedback on the adaptation policy rules, allowing to detect reconfigurations that should not have occurred, high-priority reconfigurations that are never triggered, or low-priority reconfigurations that are too frequently executed, potential inconsistencies in the rules, or wrong interpretation of priorities. The test verdict is made based on the analysis of the execution traces of the system, which is stimulated using a usage model that describes the probabilities of external events to occur. An experiment, performed on a vehicular ad-hoc network of autonomous vehicles, illustrates the interest of the approach.