Uber Mobility Platform

Appearance

System Architecture & Scaling to 100x ​

End-to-End Request & Event Lifecycle ​

mermaid
sequenceDiagram
    participant Rider
    participant Gateway as Express / Socket.io
    participant Maps as Google Maps Service
    participant Captain

    Note over Rider,Captain: Phase 1: Fare Quote & Booking
    Rider->>Gateway: GET /maps/get-distance-time
    Gateway->>Maps: Query Matrix API
    Maps-->>Gateway: Matrix Response
    Gateway-->>Rider: Fare Breakdown (Car, Moto, Auto)
    Rider->>Gateway: POST /rides/create
    Gateway->>Gateway: Generate 4-digit OTP
    Gateway-->>Captain: WS emit('new-ride')

    Note over Rider,Captain: Phase 2: Acceptance & OTP Handshake
    Captain->>Gateway: POST /rides/confirm
    Gateway-->>Rider: WS emit('ride-confirmed', { captain, otp })
    Captain->>Gateway: GET /rides/start-ride?rideId=...&otp=1234
    Gateway->>Gateway: Verify OTP Match
    Gateway-->>Rider: WS emit('ride-started')
    Gateway-->>Captain: 200 OK (Ride Ongoing)

    Note over Rider,Captain: Phase 3: Transit & Completion
    Captain->>Gateway: POST /rides/end-ride
    Gateway-->>Rider: WS emit('ride-ended')
    Gateway-->>Captain: 200 OK (Ride Completed)

Scaling to 100x Load: The First Three Bottlenecks ​

  1. WebSocket Connection Saturation on Single Node: A single Node.js instance cannot hold 100,000 concurrent driver WebSocket connections. Mitigation: Introduce Redis Pub/Sub Adapter (@socket.io/redis-adapter) to distribute connections across an auto-scaling cluster of Node instances.
  2. Google Maps API Rate Limits and Cost Explosion: Direct external calls for every drag-and-drop autocomplete or distance recalculation will incur astronomical API bills. Mitigation: Cache geocoding results and common route distances in Redis with a 24-hour TTL; introduce client-side debouncing on autocomplete inputs.
  3. Database Write Pressure on Driver GPS Updates: Storing raw GPS coordinate updates to MongoDB on every ping will saturate disk IOPS. Mitigation: Store transient driver coordinates in Redis Geospatial indexes (GEOADD, GEORADIUS) and persist only the final trip polyline to MongoDB upon ride completion.