Netflix built a real-time service topology map called Service Topology to help engineers understand microservice dependencies at scale. The system combines three complementary data sources: eBPF network flows for comprehensive coverage, IPC metrics for application-level endpoint details, and distributed tracing for actual request paths. Each source creates a separate graph stored in Netflix's custom graph database, which can be queried independently or merged into a unified view with sub-second response times. The architecture uses Kafka ingestion, Apache Pekko Streams for distributed processing, and a three-stage aggregation pipeline to resolve network intermediaries and reconstruct true app-to-app connections. Engineers can visualize dependencies, assess blast radius, overlay health status, query programmatically via gRPC, and even time-travel through historical topology states. Future plans include change event overlays and automated root cause analysis powered by the topology graph.
Table of contents
The Puzzle with a Thousand PiecesThe Three Questions Every Engineer AsksWhy This Matters More Than EverBuilding on What We LearnedWhat We Needed: A Living MapOur Approach: Three Sources of TruthGet Netflix Technology Blog ’s stories in your inboxFrom Flows to Graph: How We Built ItWhat Engineers Can Do NowThe Living Map: Always CurrentThe Journey ContinuesWhy This Matters for Our MembersWhat’s Next in This SeriesAcknowledgements29.4K Impressions2 Comments