On Bridging Relational and Document-Centric Data Stores

Big Data scenarios often involve massive collections of nested data objects, typically referred to as “documents.” The challenges of document management at web scale have stimulated a recent trend towards the development of document-centric “NoSQL” data stores. Many query tasks naturally involve reasoning over data residing across NoSQL and relational “SQL” databases. Having data divided over separate stores currently implies labor-intensive manual work for data consumers. In this paper, we propose a general framework to seamlessly bridge the gap between SQL and NoSQL. In our framework, documents are logically incorporated in the relational store, and querying is performed via a novel NoSQL query pattern extension to the SQL language. These patterns allow the user to describe conditions on the document-centric data, while the rest of the SQL query refers to the corresponding NoSQL data via variable bindings. We give an effective solution for translating the user query to an equivalent pure SQL query, and present optimization strategies for query processing. We have implemented a prototype of our framework using PostgreSQL and MongoDB and have performed an extensive empirical analysis. Our study shows the practical feasibility of our framework, proving the possibility of seamless coordinated query processing over relational and document-centric data stores.