Dynamic invariant detection for relational databases

Despite the many automated techniques that benefit from dynamic invariant detection, to date, none are able to cap-ture and detect dynamic invariants at the interface of a pro-gram and its databases. This paper presents a dynamic invariant detection method for relational databases and for programs that use relational databases and an implemen-tation of the approach that leverages the Daikon dynamic-invariant engine. The method defines a mapping between re-lational database elements and Daikon’s notion of program points and variable observations, thus enabling row-level and column-level invariant detection. The paper also presents the results of two empirical evaluations on four fixed data sets and three subject programs. The first study shows that dynamically detecting and inferring invariants in a relational database is feasible and 55 % of the invariants produced for each subject are meaningful. The second study reveals that all of these meaningful invariants are schema-enforceable us-ing standards-compliant databases and many can be checked by databases with only limited schema constructs