How Netflix Leverages gRPC for Resilience at Scale
summary
Netflix runs gRPC for over a thousand clients and services handling tens of millions of gRPC requests per second and builds resilience as middleware: retries are implemented with a retry interceptor wrapped by a dynamic client interceptor so changing max retries (e.g., from 0 to 1) instantly converts a 50% flaky service into far more successes, hedging is 95th-percentile hedging so only ~5% trigger extra requests, fallbacks use proto options or dynamic logic, deadlines propagate via a request‑expiry header across REST/GraphQL, and production validation uses failure injection testing (FIT) and CRR squeeze tests with cached redirect channels.
AI-written summary · about
chapters (12)
- 0:00Introduction to resilience
- 1:28gRPC resiliency features
- 2:11Retries and hedging
- 4:58Dynamic retry configuration
- 6:26Hedging implementation
- 8:11Fallbacks for degraded UX
- 10:16Deadline propagation
- 11:42Validating production resilience
- 12:04Failure injection testing
- 14:51Custom request routing
- 17:21Summary and future work
- 18:12Q&A session
related talks
28mHow Netflix Makes gRPC Easy to Serve, Consume, and Operate - Benjamin Fedorka, Netflix
23mReducing gRPC Call Volume Through Caching and Batching
07mKeynote: State of gRPC
53mHow Netflix Uses Java - 2026 Edition
44mArchitectural Patterns of Resilient Distributed Systems
35mEvolving Critical Systems With gRPC: Bidirectional at Scale — What Works, What Hurts, What’s Next?
45mgRPC vs REST: let the battle begin!
55mEmbracing gRPC in .NET