Multiple-View Self-Maintenance in Data Warehousing Environments

Abstract

A data warehouse is a collection of materialized views derived from relations that may not reside at the warehouse. Using these stored views, user queries can often be ev aluated much more cheaply than using the base relations. Keeping the views consistent with updates to the base relations, however, can be expensive, since it may involve querying external sources where the base relations reside. T o reduce maintenance costs, we try to maintain the views, in response to a base update, using information that is strictly local to the warehouse: the view definitions and the view contents. Only when we fail to do so do we resort to accessing a subset of the base relations. However, there may be situations where, under a specific base update and given a specific state of the views and of the subset of base relations used, there is no way to maintain a view unambiguously. Thus, the two critical questions we address are to determine, in a given situation, whether a view can be maintained unambiguously, and how to maintain it. We provide algorithms that answer these questions for a general class of views. For an important subclass, we provide algorithms that generate SQL queries whose answer determines if a view can be maintained in a given situation. We also generate SQL updates that maintain the view. We improve significantly on previous work by solving the maintenance problem in the presence of multiple views, with partial access to base data, and under arbitrary mixes of insertions and deletions. We provide better insight into the view self-maintenance problem by showing that view self-maintainability can be reduced to the problem of deciding query containment.