Polynomial time queries over inconsistent databases

Abstract

This paper investigates the problem of repairing and querying relational databases which may be inconsistent with respect to functional dependencies and foreign key constraints. Specifically, particular sets of functional dependencies, called canonical, are considered. We present a repairing strategy whereby only tuple updates and insertions are allowed in order to restore consistency: when foreign key constraints are violated, new tuples (possibly containing null values) are inserted into the database; when functional dependency violations occur, tuple updates (possibly introducing unknown values) are performed. We propose a semantics of constraint satisfaction for databases containing null and unknown values, since the repairing process can lead to such databases. The proposed approach allows us to obtain a unique repaired database which can be computed in polynomial time. The result of the repairing technique is an incomplete database (in particular, an OR-database). The consistent query answers over an inconsistent database are the certain answers on the repaired database. Relying on the results in [17] on the complexity of query processing in OR-databases, we can identify conjunctive queries which can be evaluated in polynomial time. © 2008 Springer-Verlag.