Uber Architecture

​

Overview

Uber has revolutionized transportation by connecting riders with drivers through a real-time marketplace. Operating in 10,000+ cities across 70+ countries, Uber’s platform handles millions of trips daily while maintaining sub-second response times. This case study examines Uber’s technical architecture, from frontend to backend, and how it has evolved to support massive scale and reliability.

​

Overall Architecture

​

Frontend Technologies

​

Web Frontend

Fusion.js Framework

Uber built Fusion.js as a modern React-based framework for creating robust web applications.Key Features:

• Plugin-based architecture
• Universal rendering (SSR + CSR)
• Built-in performance optimization
• Type-safe with TypeScript support

Why Custom Framework:

• Standardization across teams
• Optimized for Uber’s specific needs
• Better developer experience

Visualization.js

Custom geospatial visualization library for:

• Real-time driver locations
• Route visualization
• Heatmaps for demand
• ETA calculations

Built on top of deck.gl and WebGL for high-performance rendering.

​

Mobile Architecture

​

RIBs Architecture Pattern

Uber developed RIBs (Router, Interactor, Builder) as a cross-platform alternative to MVC: Components:

• Router: Manages navigation between screens
• Interactor: Contains business logic
• Builder: Handles dependency injection
• Presenter: Translates business logic to view updates
• View: Displays UI and captures user input

Benefits:

• Cross-platform consistency (iOS and Android)
• Better testability than MVC/MVVM
• Clear separation of concerns
• Scalable for large teams

​

Service Mesh and Gateway

​

Uber Gateway

Uber built a custom gateway on top of NGINX with dynamic configuration capabilities: Features:

• Dynamic routing rules without restarts
• Rate limiting and throttling
• Authentication and authorization
• Request/response transformation
• Service discovery integration

​

Communication Protocols

gRPC

Used for internal service-to-service communication:

• HTTP/2 based
• Binary protocol (Protocol Buffers)
• Bi-directional streaming
• Automatic code generation

Use Cases:

• Real-time location updates
• Driver dispatch
• High-frequency microservice calls

QUIC Protocol

Used for client-server communication:

• Built on UDP instead of TCP
• Reduced latency for mobile networks
• Better handling of network switches
• 0-RTT connection establishment

Why Important:

• Mobile users frequently switch networks
• Lower latency for location updates
• Better user experience in poor network conditions

Apache Thrift

Used for API definition and RPC:

• Cross-language support
• Efficient binary serialization
• Code generation for multiple languages
• Used before gRPC adoption

​

Backend Services

​

Configuration Management

Flipr (later UCDP - Unified Configuration Data Platform):

• Centralized configuration store
• Real-time configuration updates
• No deployment needed for config changes
• Versioning and rollback capabilities
• A/B testing support

​

Location Services

H3 - Hexagonal Hierarchical Geospatial Index

Uber’s geospatial indexing system (open-sourced):How It Works:

• Divides Earth into hexagonal cells
• Hierarchical levels (0-15)
• Each level provides different resolution
• Efficient proximity queries

Use Cases:

• Finding nearby drivers
• Geofencing
• Surge pricing zones
• Market analytics

Advantages:

• Better than grid-based systems
• Hexagons have uniform neighbor distances
• Efficient storage and queries
• Open-source for community use

​

Service Framework

• Spring Boot: Primary framework for Java services
• uAct: Event-driven architecture framework
  • Pub/sub messaging
  • Event sourcing support
  • Saga pattern implementation

​

Workflow Orchestration

Cadence - Distributed workflow engine:

• Async workflow orchestration
• Long-running business processes
• Fault-tolerant execution
• Workflow versioning

Use Cases:

• Trip lifecycle management
• Payment processing
• Driver onboarding
• Refund workflows

​

Database Architecture

​

OLTP Databases

DocStore

Uber’s strongly-consistent document store:Built on:

• MySQL for transactional data
• PostgreSQL for specific use cases
• Consistent across regions

Features:

• ACID transactions
• Multi-document transactions
• JSON document storage
• SQL-like query capabilities

RocksDB

Embedded key-value store:

• High-performance storage engine
• Used within other databases
• LSM tree based
• Optimized for SSDs

​

Big Data Platform

Uber’s data platform is built on the Hadoop ecosystem: Storage:

• Apache Hudi: Incremental data lake storage
• Apache Parquet: Columnar file format
• Alluxio: Distributed cache layer

Time-Series Databases:

• Pinot: Real-time analytics (OLAP)
• AresDB: GPU-powered time-series database
  • Real-time analytics on ride data
  • Sub-second query latency
  • GPU acceleration for aggregations

​

Data Processing

​

API Gateway Evolution

​

First Generation: Organic Evolution (2014)

Uber started with two key services: Dispatch Service:

• Connects riders with drivers
• Real-time matching algorithm
• Location-based routing

API Service:

• Stores user data
• Trip history
• Long-term data persistence

Limitations:

• Monolithic components
• Difficult to scale independently
• Limited flexibility

​

Second Generation: All-Encompassing Gateway (2015-2018)

Uber adopted microservices early, reaching 2,200+ microservices by 2019. Gateway Responsibilities:

• Routing to appropriate services
• Authentication and authorization
• Rate limiting
• Request/response transformation
• Monitoring and logging

Challenges:

• Gateway became a bottleneck
• Difficult to maintain
• Tight coupling between teams
• Deployment coordination required

​

Third Generation: Self-Service, Decentralized, Layered (2018-Present)

As Uber expanded to new business lines (Freight, ATG, Elevate, Groceries), a new approach was needed: Key Features:

• Self-Service: Teams can define their own API contracts
• Decentralized: Multiple gateways for different domains
• Layered: Different layers for different concerns
  • Edge layer: TLS termination, DDoS protection
  • Gateway layer: Routing, authentication
  • Service layer: Business logic

Benefits:

• Teams move faster independently
• Better isolation and fault tolerance
• Easier to maintain and scale
• Domain-specific optimizations

​

DevOps and CI/CD

​

Development Workflow

Monorepo Strategy

Uber uses a monorepo for code organization:Build System:

• Bazel: Large-scale build system
• Incremental builds
• Cached artifacts
• Deterministic builds

Benefits:

• Atomic commits across services
• Easier refactoring
• Shared libraries
• Consistent tooling

Challenges:

• Large repository size
• Build complexity
• Requires investment in tooling

Development Environment

Devpod:

• Simplified local development
• Cloud-based development environments
• Consistent setup across team
• Reduces onboarding time

​

Code Quality

Custom Tooling:

• NullAway: Eliminates NullPointer exceptions in Java
• NEAL: Code linting and style enforcement
• Piranha: Cleans outdated feature flags automatically

​

Testing

SLATE:

• Manages short-lived testing environments
• Spins up isolated test clusters
• Automatic cleanup

Ballast:

• Ensures smooth user experience
• Load shedding during overload
• Graceful degradation

​

Experimentation Platform

Uber’s experimentation platform enables data-driven decisions: Features:

• A/B testing framework
• Deep learning-based analysis
• Real-time metrics
• Statistical significance calculations

Open Source:

• Pyro: Probabilistic programming in Python
• Used for Bayesian inference in experiments

​

Build and Deploy

Build - uBuild

Uber’s containerization platform:

• Built on Buildkite
• Packages services into containers
• Optimized build caching
• Multi-architecture support

Deploy - Spinnaker

Uses Netflix Spinnaker for deployments:

• Multi-cloud support
• Canary deployments
• Automated rollbacks
• Integration with monitoring

​

Monitoring and Observability

uMetric Platform:

• Built on Cassandra
• Consistent metrics across services
• Real-time dashboards
• Alerting integration

Monitoring Stack:

• Metrics: uMetric
• Logs: ELK Stack
• Traces: Distributed tracing
• Alerts: uMonitor

OnCall Dashboard:

• On-call rotation management
• Incident tracking
• Automated escalation
• Post-mortem workflows

​

Special Tooling

​

Key Architecture Decisions

​

1. Microservices at Scale

• Early adoption of microservices (2,200+ services)
• Service per business capability
• Clear ownership model
• Independent deployment

​

2. Real-Time Processing

• Location updates every few seconds
• Sub-second matching algorithms
• Real-time pricing (surge)
• Event-driven architecture

​

3. Geospatial Optimization

• Custom H3 indexing system
• Efficient proximity searches
• Geofencing for regions
• Market-specific configurations

​

4. Observability First

• Custom monitoring tools (uMetric, uMonitor)
• Distributed tracing
• Real-time alerting
• Data-driven decisions

​

5. Mobile Optimization

• Custom RIBs architecture
• QUIC protocol for better mobile performance
• Offline capabilities
• Battery and data usage optimization

​

Lessons Learned

​

Scale Statistics

• 10,000+ cities worldwide
• 70+ countries
• Millions of trips daily
• 2,200+ microservices
• Sub-second response times for most operations
• Thousands of engineers

​

References

• Uber Engineering Blog
• H3 Geospatial Index (Open Source)
• RIBs Architecture (Open Source)
• Uber Tech Stack Overview