Publication
Boosting DBMS Test Coverage via LLM-Driven SQL Generation
Eslam Abdelkarim; Carsten Binnig; Anupam Sanghi
In: Proceedings of the 2026 11th International Workshop on Testing Database Systems, DBTest 2026, Bengaluru, India, 31 May 2026 - 5 June 2026. International Workshop on Testing Database Systems (DBTest), Pages 49-53, ACM, 2026.
Abstract
Database management systems require comprehensive testing, but
achieving high code coverage in complex DBMS implementations
remains challenging. Manual test writing is time-consuming, while
automated query generation tools struggle with schema flexibility
and coverage guidance. This paper presents Quover, an automated
approach for improving DBMS test coverage through large lan-
guage model (LLM)-generated SQL queries. Specifically, Quover
implements an iterative coverage-guided methodology that iden-
tifies uncovered functions in the target DBMS source code and
generates targeted SQL queries. In each iteration, an LLM call is
made with a rich context, including function descriptions and code
snippets, and their effectiveness is validated. Our experiments show
that over 57% coverage can be achieved within a few hours–already
representing a 14% improvement over state-of-the-art automated
systems. Furthermore, we evaluate Quover across multiple data-
base schemas and DBMSs, demonstrating its effectiveness as an
automated approach alongside handcrafted test suites.