🚕

System Design of Uber

Jun 22, 2024

System Design of Uber

Introduction

  • Presenter: Anshul Sadariya, Software Engineer 3 at Google
  • Previous system design videos: Twitter, Tinder
  • Focus on understanding large-scale products
  • Video topic: System design for Uber (applicable to similar services like Ola, Lyft, Kareem)
  • Plug: Scalar's free master classes by industry experts

Customer User Journeys (CUJs)

  • User Types: Normal user (rider) and driver
  • Driver CUJs:
    • Create profile
    • Show availability
    • Accept or reject rides
    • Navigate to pickup location
    • Start the ride
    • Navigate to drop location
    • End the ride
    • Accept payment and optionally rate the rider
  • Rider CUJs:
    • Create profile
    • See nearby cabs
    • Add start/end location
    • Show ETA and approximate price
    • Book cab
    • Make payment and optionally rate the driver
    • See past trips and payment history
    • Check current trip status

Functional Requirements

  • Each CUJ translates into a functional requirement of the app
  • Mapping CUJs to functional requirements, e.g., showing ETA or booking a cab

Non-Functional Requirements

  • Minimal latency: Fast response times
  • High availability: Service should rarely, if ever, go down
  • Durability: Data should persist and not be lost
  • Scalability: Able to handle peak times and expansion to new locations
  • Consistency: Strongly consistent where needed (e.g., active rides), eventually consistent where acceptable (e.g., past payments)

Operational Scale

  • Metrics:
    • 100 million MAU (monthly active users)
    • 10 million daily rides
    • 1 million daily active drivers
    • 100 QPS (queries per second)
  • Considerations:
    • Handling high QPS for ride requests
    • Efficiently finding and pinging available drivers
  • Optimization Techniques:
    • Geohashing, Quadtree, and Google S2 Library for managing lat/long data
    • Distributed Hash Ring for managing server responsibilities

Major Technical Concepts

Communication

  • HTTP Requests: Synchronous
  • WebSocket Servers: Asynchronous, peer-to-peer connection

Mapping and Location

  • Map Service:
    • Converts lat/long to chunk ID
    • Computes ETA
    • Determines pricing
  • Location Service:
    • Stores latest lat/long data
    • Historical location data for analytics and machine learning
    • Uses Redis for fast data retrieval
    • Cassandra cluster for historical data

Finding Nearby Cabs

  • Cab Finder Service (or Dispatch Service): Matches riders with drivers based on proximity
  • Consistent Hashing and Gossip Protocol: For managing server responsibilities dynamically

Data Management and Analysis

  • Trip Service:
    • Tracks trip states (accepted, started, ended)
    • Stores active and inactive trip data
    • Uses SQL database for active trips and Cassandra for historical trips
  • Payment Service:
    • Calculates fare based on trip data
    • Integrates with payment gateways (Stripe, Razorpay)
  • Profile Service: Manages CRUD operations for user and driver profiles
  • Additional Analytics:
    • Preferred access points based on aggregate location data
    • Machine learning for fraud detection and driver profiling
    • Use of Kafka queues for real-time data processing

Additional Concepts

  • ELK Stack: For log analysis and visualization

Conclusion

  • Covered major technical concepts and services required to build an Uber-like application
  • Emphasized understanding of functional and non-functional requirements
  • Discussed real-time data processing and machine learning integrations
  • Encouragement to check out scalable and insightful content

Full transcript