Dial Once, Scale Infinitely: Production-Ready gRPC Without the Mess
summary
DigitalOcean describes a gRPC stack called DRPC that centralizes connection setup with a shared client-creator abstraction, uses auto-generated typed clients and interceptors, and resolves targets via a custom Endpoint Discovery Service (EDS). It runs on DOCCC with auto-mTLS, OPA-based policy checks, built-in client load balancing, and claims scale: ~4,000 services, 25,000+ instances, ~1M internal RPS, sub-10ms internal latency and 15–20 GB/s. The pattern enforces dial once, use many long‑lived connections and reduces redundant dials.
AI-written summary · about
chapters (14)
- 0:00Introduction to gRPC
- 2:07Why gRPC replaced RPC
- 8:27Protocol Buffers in gRPC
- 10:12DigitalOcean’s gRPC scale
- 11:37Hello World gRPC service
- 12:39DRPC framework features
- 13:20Interceptors in gRPC
- 14:24Client load balancing in DRPC
- 15:28Auto TLS with DOCCC
- 16:52Endpoint Discovery Service
- 17:54mTLS with DRPC
- 18:35Role-based access control in gRPC
- 19:40EDS for service discovery
- 20:23Scaling with DRPC, DOCCC, EDS
related talks
17mBuilding a gRPC Proxyless World: How Reddit Scaled Resilience...
23mA Simplified Service Mesh With gRPC
33mgRPC For Microservices: Service-mesh and Observability
14mOverview of gRPC
28mgRPC: 5 Years Later, Is It Still Worth It?
47mMonolith to Microserv. at 20M+ requests per second: Latency and Scale Challenges
31mRPC Frameworks Overview
35mgRPC Load Balancing on Kubernetes - Jan Tattermusch, Google