Codd's rule of entity integrity stipulates that every table in a database has a primary key. Hence, the attributes that form the primary key carry no missing information and have unique value combinations. In practice, data records cannot always meet such requirements. Previous work has proposed the notion of a key set, which can identify more data records uniquely when information is missing. Apart from the proposal, key sets have not been investigated much further. We outline important database applications, and investigate computational limits and techniques to reason automatically about key sets. We establish a binary axiomatization for the implication problem of key sets, and prove its coNP-completeness. We show that perfect models do not always exist for key sets. Finally, we show that the implication problem for unary key sets by arbitrary key sets has better computational properties. The fragment enjoys a unary axiomatization, is decidable in time quadratic in the input, and perfect models can always be generated.
展开▼
