Evolving Critical Systems With gRPC: Bidirectional at Scale — What Works, What Hurts, What’s Next?
summary
Mastercard runs gRPC bidirectional streaming across a global backbone, handling 159.4 billion transactions in 2024 and hundreds of millions per day, using Protobuf schemas in a central repo and tens of internal APIs. Engineering builds telemetry at TCP, HTTP/2, stream and application layers, runs WireShark + a TCP proxy to mutate TLS records and confirm fatal alerts force reauthentication, and shifts to client-side load balancing with per-message routing, message IDs, redundancy connections, XDS for discovery, and experiments toward HTTP/3/QUIC.
AI-written summary · about
chapters (9)
- 0:00Evolving critical systems with gRPC
- 2:08Why Mastercard chose gRPC streaming
- 7:31Developer productivity with gRPC
- 11:44gRPC streaming security
- 13:07gRPC bidirectional streaming pain points
- 18:44Load balancing persistent gRPC
- 22:20Customer view on gRPC streaming
- 23:47gRPC and HTTP/3 future plans
- 25:35Q&A
related talks
29mThe Story of Why We Migrate to gRPC and How We Go About It
35mThe Story of why we migrate to gRPC and how we go about it
16mKeynote: Overview of gRPC
33m"RPC frameworks overview"
20mKeynote: Overview of gRPC
22mA Case Study of Enhancing gRPC Micro-Services for Supporting High Throughput, Low Latency Trading
24mgRPC Communication Patterns – A Deep Dive
36mAdopting gRPC at Spotify