Abstract

Data-processing systems increasingly face data that is eclectic—it spans many heterogeneous schemas and its schemas evolve over time. Unfortunately, existing approaches for processing and querying data are not ideal for eclectic data since they impose a tradeoff between efficient querying and simplicity. We argue that this limitation stems from the very foundations of data processing: data models and their corresponding query languages. No existing approach—whether relational, document, or hybrid—is designed to enable ingesting, querying, and reasoning about heterogeneous types of data. In this paper we propose Zed, a new approach to data processing that centers around data types. Zed elevates data types to be first-class members of both the data model and query language, and by doing so offers a promising path towards easing the processing of eclectic data.