Checking Transaction Isolation Violations with Graph Queries

Abstract.

Distributed databases provide different transaction isolation levels for higher performance and fault tolerance. However, implementing isolation models is challenging, and database systems can produce executions that violate their isolation guarantees. In this work, we propose GRAIL, a new approach that uses graph databases and queries to detect isolation violations expressed as anti-patterns in transactional dependency graphs. We implement the approach on top of the popular ArangoDB and Neo4j graph databases and show its efficiency through an experimental analysis of real executions of ArangoDB as a system under test.