Netflix engineers built Graph Abstraction, a high-throughput platform managing 650 TB of graph data with millisecond-level latency globally. Rather than a standalone graph database, it layers on top of existing Netflix infrastructure: a Key-Value abstraction for current state, a TimeSeries abstraction for historical data, and EVCache for distributed caching. The system supports use cases like social graphs for Netflix Gaming and service topology graphs for incident analysis. It restricts traversal depth to trade query flexibility for consistent low latency, achieving single-digit millisecond latency for single-hop traversals and under 50ms for two-hop queries at p90. Global availability is maintained through asynchronous cross-region replication with eventual consistency. A gRPC API inspired by Gremlin exposes traversal capabilities to internal services.
Sort: