Understanding the implications of view update policies

Abstract

Database views are traditionally described as unmaterialized queries, which may be coincidentally updatable according to some fixed criteria. One of the problems in updating through views lies in determining whether a given view modification can be correctly translated by the system. To define an updatable view, a view designer must be aware of how an update request in the view will be mapped into updates of the underlying relations. Furthermore, because of side effects, the view designer must also be made aware of the effects of underlying updates back into the view. To address this problem, we present a general algorithm that predicts the effects of arbitrary mapping policies. Given an update policy, this algorithm indicates whether a desired update will, in fact, occur in the view and describes all possible side effects it may have, documenting the conditions under which they occur. The algorithm subsumes the results obtained by other view design tools, and generalizes their use to encompass a larger class of views.