Diffy - Predicting and Preempting Regressions in Services - Puneet Khanduri & Prashant Khanduri
summary
Diffy replays sampled production traffic and multicast requests to three instances—candidate, primary, secondary—then compares candidate vs primary and subtracts primary vs secondary differences to do noise cancellation. It supports Thrift, HTTP and Kafka replays, reduced the combinatorial testing problem from trillions to actionable diffs, and enables automatic promotion in CI/CD when no regressions appear. Open source at open-diffy/diffy, used at Twitter to catch a signup bug pre-deploy, and plans include PII scrubbing for production deployments.
AI-written summary · about
chapters (7)
- 0:00Start
- 3:20The Tipping Point *Does my new microservice behave the same as my monolith?
- 4:14Unit Tests
- 5:13Component Tests
- 5:45Integration Tests A request path going through
- 9:05Diffy Topology
- 15:12Automation
related talks
38mDIFFY Catching Bugs without Writing Tests
22mDiffy: Data-Driven Bug Finding for Configurations
26mServer-Side Diff: Future-Proofing ArgoCD's Comparison Engine
28mWhat the Diff?
38mProgressive Delivery: Patterns & Benefits of Decoupling Deploy from Release
52mPerformance & Stability Anti patterns
30mIt's Down! Simulating Incidents in Production
20mNetwork Monitor: A Tale of ACKnowledging an Observability Gap