Uniqueness Constraints for Object Stores

Object stores offer an increasingly popular choice for data management and analytics. As with every data model, managing the integrity of objects is fundamental for data quality but also important for the efficiency of update and query operations. In response to shortcomings of unique and existence constraints in object stores, we propose a new principled class of constraints that separates uniqueness from existence dimensions of data quality, and fully supports multiple labels and composite properties. We illustrate benefits of the constraints on real-world examples of property graphs where node integrity is enforced for better update and query performance. The benefits are quantified experimentally in terms of perfectly scaling the access to data through indices that result from the constraints. We establish axiomatic and algorithmic characterizations for the underlying implication problem. In addition, we fully characterize which non-redundant families of constraints attain maximum cardinality for any given finite sets of labels and properties. We exemplify further use cases of the constraints: elicitation of business rules, identification of data quality problems, and design for data quality. Finally, we propose extensions to managing the integrity of objects in object stores such as graph databases.