DDLumos: Understanding and Detecting Atomic DDL Bugs in DBMSs
summary
Analyzes 207 real-world Atomic DDL bugs and builds DDLUMOS, a tester that uses metadata conflict-guided DDL synthesis and graph-based consistency analysis to generate concurrent DDL sequences and verify schema state. Finds 73 previously unknown bugs across six DBMSs (PostgreSQL, MySQL, MariaDB, etc.), with 20 availability, 31 incorrect-result, and 22 post-recovery inconsistency bugs; vendors fix 14. Attributes root causes to incomplete rollbacks (32%), bad signalization (22%), concurrency errors (18%) and error-handling faults (22%).
AI-written summary · about
chapters (15)
- 0:00DDLumos introduction
- 0:40Atomic DDL overview
- 1:23Non-atomic DDL risks
- 2:05Atomic DDL implementation
- 3:10Atomic DDL bugs impact
- 4:35Testing tools limitations
- 5:18Real-world atomic DDL bugs
- 6:46Bug triggering conditions
- 7:31Test oracles for atomic DDL
- 8:18DDLumos design
- 10:28Metadata conflict-guided synthesis
- 11:54Graph-based consistency analysis
- 12:58Evaluation results
- 13:42Tool comparison
- 14:48Summary
related talks
36mSQLancer: Automatically Finding Bugs in Database Management Systems
19mAtomic DDL in MariaDB
11mDetecting Transactional Bugs in Database Engines via Graph-Based Oracle Construction
52mData Corruption Bugs: Diagnosis and Lessons
10mHow to test software without writing tests
18mValScope: Value-Semantics-Aware Metamorphic Testing for Detecting Logical Bugs in DBMSs
06mSQLancer: Finding Logic Bugs in Database Management Systems
19mPinolo: Detecting Logical Bugs in Database Management Systems with Approximate...