Staggered Maintenance of Multiple Views

A framework for supporting data integration using the materialized and virtual approaches. In Jagadish and Mumick JM96]. 1 number of occurrences in V 2. Thus the size of the materialization of V with counts is bounded by the size of V 1. The staggered algorithm can then be extended in a straightforward fashion. A detailed description of this is beyond the scope of this paper. The SQL UNION operation can be a bag or a set union (by specifying ALL in the rst case). In the case of a view deened using UNION ALL, the change set for the view is simply a bag union of the change sets of the operand expressions. Therefore, the propagation can be done without referring to the deriving tables. In the case of the set union, we treat the view as a UNION ALL view for the purpose of view maintenance. In other words, we maintain multiplicities in the view and perform change propagation as in the UNION ALL case, but ignore the multiplicities for all querying to the view. Alternatively, the view can be maintained without counts using an approach similar to the one for EXCEPT ALL views described above. Other operations that return sets, such as project-distinct can be handled by always maintaining counts in the view and by carefully distinguishing the cases when multiplicities must be retained and when they should be ignored. An algorithmsimilar to the staggered algorithm was developed independently by Hull and Zhou HZ96]. They developed the algorithm for maintaining views in a warehousing environment, where all views are maintained immediately upon receiving the updates. They considered fully materialized, partially materialized and virtual views and did not consider deferred and snapshot views. Unlike in our model, all the updates to all the tables in the warehouse were under the control of the maintenance process. The views in their system are derived from set-based relational operations as opposed to bag-based operations like Except. They did not consider the problem of self-joins. In this paper, we presented a staggered view maintenance algorithm that interleaves refresh and propagate operations. Each invocation of the algorithm causes all (non-deferred) views in a view-group to be refreshed. The interleaving guarantees that a table is in the correct state when a propagate equation referencing the table is evaluated provided that the referenced table is not a base table or a table outside the viewgroup that …