Netflix Real-Time Distributed Graph: Query Engine Design
Understand the key design decisions behind Netflix's graph query serving layer — BFS traversal, async thread pools, volatility-aware caching, configurable filtering, and adaptive concurrency — and know when to apply these patterns in your own systems.
4 sections · 8 lessons
Course outline
Graph Traversal at Scale
- Why BFS beats DFS for distributed graphs
- From sequential calls to parallel batches
Async-First Concurrency
- Why thread-per-request doesn't scale
- 16-24 threads serving thousands of concurrent queries
Volatility-Aware Caching
- The problem with uniform TTLs
- Mapping TTL to data change rate
Configurable Filtering and Adaptive Concurrency
- Filtering hierarchy: global, per-hop, per-edge-type
- Asymmetric concurrency limiting