Skip to main content
netflixdata-gatewayreferencearchitecture

Netflix Data Gateway Reference

This section contains research and learnings from Netflix's Data Gateway architecture, which serves as inspiration for the Prism data access layer.

What is Netflix's Data Gateway?​

Netflix's Data Gateway is a battle-tested platform that provides data abstraction layers to simplify and scale data access across thousands of microservices. It decouples application logic from database implementations, enabling:

  • Unified APIs for diverse data stores (Cassandra, EVCache, DynamoDB, etc.)
  • Operational resilience through circuit breaking, load shedding, and failover
  • Zero-downtime migrations via shadow traffic and dual-write patterns
  • Massive scale: 8M+ QPS, 3,500+ use cases, petabytes of data

Why This Matters for Prism​

Prism adopts many of Netflix's proven patterns while improving on performance and operational simplicity:

AspectNetflix ApproachPrism Enhancement
Proxy LayerJVM-based gatewayRust-based for 10-100x performance
ConfigurationRuntime deployment configsClient-originated requirements (apps declare needs)
TestingProduction-validatedLocal-first with sqlite, local kafka
DeploymentKubernetes-nativeFlexible: bare metal, VMs, or containers

Core Concepts​

Data Abstractions​

Netflix built multiple abstraction layers on their Data Gateway platform:

  • Key-Value: Primary abstraction for 400+ use cases
  • TimeSeries: Handles 10M writes/sec for event tracking
  • Distributed Counter: Scalable counting with tunable accuracy
  • Write-Ahead Log (WAL): Durable, ordered mutation delivery

Scale & Performance​

Netflix's Data Gateway operates at massive scale:

  • 8 million queries per second (key-value abstraction)
  • 10 million writes per second (time-series data)
  • 3,500+ use cases across the organization
  • Petabyte-scale storage with low-latency retrieval

Read more about scale metrics →

Migration Patterns​

Netflix's approach to zero-downtime migrations:

  • Dual-Write Pattern: Write to old and new datastores simultaneously
  • Shadow Traffic: Validate new systems with production load
  • Phased Cutover: Gradual migration with rollback capability
  • Schema Evolution: Automated compatibility checking

Key Learnings​

1. Abstraction Simplifies Scale​

Managing database API complexity becomes unmanageable as services scale. A robust data abstraction layer:

  • Shields applications from database breaking changes
  • Provides user-friendly gRPC/HTTP APIs tailored to access patterns
  • Enables backend changes without application code changes

2. Prioritize Reliability​

Building for redundancy and resilience:

  • Circuit breaking and back-pressure prevent cascading failures
  • Automated load shedding protects backends during traffic spikes
  • Rigorous capacity planning prevents resource exhaustion

3. Data Management is Critical​

Proactive data lifecycle management:

  • TTL and cleanup should be designed in from day one
  • Cost monitoring: Every byte has a cost
  • Tiering strategies: Move cold data to cost-effective storage

4. Sharding for Isolation​

Product/feature sharding prevents noisy neighbor problems:

  • Dedicated proxy instances per product or SLA tier
  • Independent scaling and capacity planning
  • Clear ownership and blast radius containment

Read full lessons learned →

Reference Materials​

PDF References​

Original blog posts and articles are archived in the references/ directory:

  • Data Gateway platform overview
  • Key-Value abstraction deep dive
  • Time-series architecture
  • Real-time data processing

How Prism Uses These Learnings​

Prism incorporates Netflix's battle-tested patterns:

  1. Data Abstractions (ADR-026): KeyValue, TimeSeries, Graph, Entity patterns
  2. Client Configuration (ADR-001): Apps declare requirements, Prism provisions
  3. Backend Plugins (RFC-008): Clean abstraction for adding new backends
  4. Shadow Traffic (ADR-031): Zero-downtime migrations like Netflix
  5. Sharding Strategy (ADR-034): Product/feature isolation

See our ADRs and RFCs for implementation details.

Note: This documentation is derived from Netflix's public blog posts, conference talks, and open-source contributions. All credit goes to Netflix for pioneering these patterns at scale.