In data-integration systems, information sources often have various constraints such as “all houses stored at a source have a unique address.” These constraints are very useful to compute answers to queries. In this paper we study how to describe these constraints, so that they can be utilized in query processing and optimization. We consider the local-as-view approach to data integration (under the open-world assumption), in which source contents and user queries are formulated on predefined global predicates. In this approach, source contents and constraints often exist before the global predicates are designed. We discuss two different levels of describing constraints: local constraints are defined sources, while global constraints are on global predicates. We formally define two types of global constraints, namely general global constraints and source-derived global constraints. We present the advantages of having these constraints, and discuss open problems that need more ...