Generating Test Databases for Database-Backed Applications

Database-backed applications are widely used. To effectively test these applications, one needs to design not only user inputs but also database states, which imposes unique challenges. First, valid database states have to satisfy complicated constraints determined by application semantics, and hence are difficult to synthesize. Second, the state space of a database is huge, as an application can contain tens to hundreds of tables with up to tens of fields per table. Making things worse, each test run takes time, as it involves database operations. Consequently, unhelpful tests can severely waste testing resources.

We propose DBGRILLER, a framework that supports the testing of database-backed applications by generating test database states. To effectively generate valid database states, DBGRILLER strategically injects minor mutation into existing database states and transforms part of the application-under-test into a standalone validity checker. To tackle the huge database state space and save testing time, DBGRILLER uses program analysis to identify a novel branch-projected DB view that can be used to filter out database states that are unlikely to increase the testing branch coverage. Our evaluation on 9 popular open-source database applications shows that DBGRILLER can effectively increase the testing branch coverage and expose previously unknown bugs.