UDA (Unified Data Architecture) at Netflix
summary
Netflix presents Unified Data Architecture (UDA), a single canonical system (called Upper) to define business concepts like Movie or Talent once and automatically project those definitions to GraphQL schemas, Avro/Flink pipelines, Iceberg tables, OpenSearch indexes, and Java APIs. The talk covers how fragmented models cause duplication and quality issues, how Netflix uses RDF-based knowledge graphs to structure shared semantics, and the “Model Once, Represent Everywhere” approach that generates GraphQL, Avro, SQL, etc., from one canonical model.
AI-written summary · about
chapters (9)
- 0:00UDA at Netflix
- 3:54Fragmented data models
- 8:00Domain models and universes
- 16:19Live domain modeling demo
- 21:29GraphQL and Avro projections
- 27:59Federation and semantic interop
- 33:13Auto-governance and formalization
- 37:12Extracting models from APIs
- 40:45Q&A
related talks
22mNetflix Data Mesh: Composable Data Processing
25mMoving Data From Graph to the Warehouse, A Configuration Driven Approach
49mData Platform Architecture, Evolution, and Philosophy
29mHigh-Quality Performant and Cost Efficient Schema-Aware Data Streams on Flink at Netflix Scale
31mFederated Data Architecture
01h 01mData-Centric in Action
37mMetadata Automation in the Data Mesh
39mData Mesh: Why, What and How