Feedback-guided Adaptive Testing of Distributed Systems Designs
summary
Fest uses feedback-guided adaptive schedule generation for formal distributed-system models, implemented in the P framework and evaluated on 94 models. It replays a schedule to a chosen prefix via deterministic replay, performs suffix resampling (indirect mutation) to produce valid alternative schedules, and scores runs with combined diversity feedback (novel happens-before patterns) and compliance feedback (match to developer-specified scenarios). Results: up to 41× behavioral coverage, 278× scenario coverage, and 33% more bugs found versus prior sampling methods.
AI-written summary · about
chapters (7)
- 0:00Distributed system testing intro
- 2:27Traditional verification limits
- 4:53Schedule exploration challenges
- 7:21Feedback-driven fuzzing for distributed systems
- 9:32Schedule interestingness semantics
- 12:46Evaluation results and impact
- 13:28Conclusion and future work
related talks
14mModel-guided Fuzzing of Distributed Systems
18mEfficient Exposure of Partial Failure Bugs in Distributed Systems with Inferred...
23mRandomized Testing of Distributed Systems with Probabilistic Guarantees
19mCAFault: Enhance Fault Injection Technique in Practical Distributed Systems via...
27mEfficient and Correct Test Scheduling for Ensembles of Network Policies
08mUnlucky Simulation
26mDifferential Testing with Foundry
40mTesting Distributed Systems w/ Deterministic Simulation