Summary
This blog post, part of a series on Netflix's Real-Time Distributed Graph (RDG), discusses the design and implementation of a query serving layer using gRPC to provide efficient, low-latency access to a constantly evolving graph. It details the challenges faced in querying a billion-edge graph while maintaining sub-100ms response times for diverse workloads, and how architectural decisions, including a breadth-first traversal approach and smart caching strategies, were implemented to optimize performance. The authors share technical decisions, challenges encountered, and principles for designing I/O-heavy distributed systems.