In this paper, we present CONSERV - a middleware infrastructure for the development of virtual databases in pervasive computing environments. CONSERV provides an SQL front-end for posing and processing queries on information provided by ad-hoc communities of Web services hosted by peers that arbitrarily join and leave the system. The cornerstone of the proposed infrastructure is the fact that we replace the traditional treatment of databases as persistent collections of records by the assumption that a database relation is a collection of records dynamically compiled from such ad-hoc sets of peers. Each peer offers data to the relations through a workflow of Web services. Another aspect of our approach is that we confine query processing over specific sets of peers that we call communities. Communities are defined based on the current context of the peer initiating each query. Since our infrastructure departs from the traditional query processing strategies, we discuss query processing as customized in CONSERV.