Partial answers for unavailable data sources

Abstract

Many heterogeneous database system products and prototypes exist today; they will soon be deployed in a wide variety of environments. Most existing systems suffer from an Achilles' heel: they ungracefully fail in presence of unavailable data sources. If some data sources are unavailable when accessed, these systems either silently ignore them or generate an error. This behavior is improper in environments where there is a non-negligible probability that data sources cannot be accessed (e.g., Internet). In case some data sources cannot be accessed when processing a query, the complete answer to this query cannot be computed; some work can however be done with the data sources that are available. In this paper, we propose a novel approach where, in presence of unavailable data sources, the answer to a query is a partial answer. A partial answer is a representation of the work that has been done in case the complete answer to a query cannot be computed, and of the work that remains to be done in order to obtain this complete answer. The use of a partial answer is twofold. First, it contains an incremental query that allows to obtain the complete answer without redoing the work that has already been done. Second, the application program can extract information from a partial answer through the use of a secondary query, which we call a parachute query. In this paper, we present a framework for partial answers and we propose three algorithms for the evaluation of queries in presence of unavailable sources, the construction of incremental queries and the evaluation of parachute queries.

This work has been done in the context of Dyade, joint R&D venture between Bull and Inria.