Breaking Valkey on Purpose: Chaos Fuzzing a High-Performance Key-Value Store With Agentic AI
summary
Demonstrates a custom Chaos Fuzzer that spins up Valkey clusters (example run: seven shards, five replicas per shard = 42 nodes), injects SIGKILL/SIGTERM and validates seven checks including 16,384 slot coverage and data replication, finds real bugs (Valgrind report) and quorum-loss failures, and closes the loop with Agentic AI (Amazon Bedrock + Kiro agent) to parse logs, classify root causes, suggest new fuzz scenarios and auto-generate inputs; code and daily runs live on GitHub.
AI-written summary · about
chapters (10)
- 0:00Introduction to chaos testing
- 1:51Understanding Valkey
- 5:50Valkey architecture and design
- 8:12Failure detection flow
- 10:56Identifying system bugs
- 14:10Live fuzzer demonstration
- 22:05Reproducing failure scenarios
- 28:49Leveraging AI for testing
- 32:10Best practices and conclusions
- 35:54Questions and answers
related talks
19mFuzzing in the GenAI Era
50mWhat the Fuzz
53mFuzz Testing for Fun and Profit
01h 00mClusterFuzz: Fuzzing at Google Scale
23mStop Performance Regressions Before Release Valkey’s CI Benchmarking
19mLearnings from Chaos experiments
19mShaking Up Quantum Simulators with Fuzzing and Rigour
35mChaos Engineering WG Deep Dive