Design_uber
Chapter 48: Designing Uber
Section titled “Chapter 48: Designing Uber”Real-Time Ride Matching at Global Scale
Section titled “Real-Time Ride Matching at Global Scale”48.1 Uber Overview
Section titled “48.1 Uber Overview”Uber connects riders with drivers through a real-time marketplace, processing millions of rides daily across 10,000+ cities.
Uber by the Numbers =================
┌─────────────────────────────────────────────────────────────┐ │ 130M+ monthly active users │ │ 6M+ driver partners │ │ 10,000+ cities globally │ │ 2B+ rides completed │ │ 10M+ ride requests per day at peak │ └─────────────────────────────────────────────────────────────┘Requirements Analysis
Section titled “Requirements Analysis”| Requirement | Challenge | Solution |
|---|---|---|
| Real-time matching | < 10 seconds | Geospatial indexing |
| High availability | 99.99% uptime | Multi-region, microservices |
| Location updates | 100K+ per second | WebSocket, Redis |
| Surge pricing | Complex calculations | Real-time analytics |
| Payment | Global compliance | Stripe integration |
48.2 High-Level Architecture
Section titled “48.2 High-Level Architecture” Uber Architecture ===============
┌─────────────────────────────────────────────────────────────┐ │ Mobile Apps │ │ (Rider App / Driver App) │ └────────────────────────────┬────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────┐ │ API Gateway │ │ (Kong / Nginx) │ └────────────────────────────┬────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────┐ │ Core Services │ │ │ │ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐ │ │ │ Dispatch │ │ Map │ │ Trip │ │ User │ │ │ │ Service │ │ Service │ │ Service │ │ Service│ │ │ └─────────┘ └─────────┘ └─────────┘ └─────────┘ │ │ │ │ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐ │ │ │ Payment │ │Pricing │ │Supply │ │Rating │ │ │ │ Service │ │ Service │ │Service │ │Service │ │ │ └─────────┘ └─────────┘ └─────────┘ └─────────┘ │ └────────────────────────────┬────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────┐ │ Data Layer │ │ │ │ ┌──────────┐ ┌──────────┐ ┌──────────┐ │ │ │ MySQL │ │Cassandra │ │ Redis │ │ │ │(Trip data)│ │(Locations)│ │(Cache) │ │ │ └──────────┘ └──────────┘ └──────────┘ │ │ │ │ ┌──────────┐ ┌──────────┐ ┌──────────┐ │ │ │ Kafka │ │PostGIS │ │ Blob │ │ │ │(Events) │ │(Geo) │ │ (Images) │ │ │ └──────────┘ └──────────┘ └──────────┘ │ └─────────────────────────────────────────────────────────────┘48.3 The Dispatch System
Section titled “48.3 The Dispatch System”The heart of Uber - matching riders with drivers in real-time.
Dispatch System Flow ===================
┌─────────────────────────────────────────────────────────────┐ │ Step 1: Rider Requests Ride │ │ ─────────────────────────────────────────────────────────│ │ │ │ ┌──────────┐ │ │ │ Rider │──────▶ Request: Pickup Location, │ │ │ App │ Destination, Ride Type │ │ └──────────┘ │ │ │ │ │ │ WebSocket │ │ ▼ │ │ ┌──────────────────────────────────────────────────────┐ │ │ │ Dispatch Service Receives Request │ │ │ └──────────────────────────────────────────────────────┘ │ └─────────────────────────────────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────┐ │ Step 2: Find Nearby Drivers │ │ ─────────────────────────────────────────────────────────│ │ │ │ Query: Drivers within 5km of pickup │ │ │ │ ┌──────────────────────────────────────────────────────┐ │ │ │ Geospatial Index │ │ │ │ │ │ │ │ (lat, lng) │ │ │ │ │ │ │ │ ● ● ● ←── Drivers │ │ │ │ ● ● ● │ │ │ │ ● ● ● │ │ │ │ 📍 ←── Pickup location │ │ │ │ │ │ │ │ Using: H3 (Hexagonal hierarchical indexing) │ │ │ └──────────────────────────────────────────────────────┘ │ └─────────────────────────────────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────┐ │ Step 3: Calculate ETA & Price │ │ ─────────────────────────────────────────────────────────│ │ │ │ • Distance to pickup (route API) │ │ • Estimated trip time │ │ • Base fare + time + distance │ │ • Surge multiplier (if applicable) │ │ │ │ Return to rider: "5 min away, $15-20" │ └─────────────────────────────────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────┐ │ Step 4: Send to Drivers │ │ ─────────────────────────────────────────────────────────│ │ │ │ ┌──────────────────────────────────────────────────────┐ │ │ │ Send push notification to top 5 drivers │ │ │ │ │ │ │ │ Driver 1: "Ride request! 5 min away, $18" │ │ │ │ Driver 2: "Ride request! 5 min away, $18" │ │ │ │ ... │ │ │ └──────────────────────────────────────────────────────┘ │ │ │ │ │ │ Drivers have ~15 seconds to accept │ │ ▼ │ │ ┌──────────────────────────────────────────────────────┐ │ │ │ First to accept wins! │ │ │ └──────────────────────────────────────────────────────┘ │ └─────────────────────────────────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────┐ │ Step 5: Match & Start Trip │ │ ─────────────────────────────────────────────────────────│ │ │ │ • Driver accepts → Create Trip record │ │ • Notify rider: "Driver is on the way" │ │ • Track driver location in real-time │ │ • Begin trip when driver arrives │ │ │ └─────────────────────────────────────────────────────────────┘48.4 Geospatial Data Handling
Section titled “48.4 Geospatial Data Handling”H3: Uber’s Geospatial Index
Section titled “H3: Uber’s Geospatial Index”Uber developed H3 for efficient geospatial queries.
H3: Hexagonal Hierarchical Indexing =================================
Why H3? ───────
Traditional: Geohash ─────────────────── • Square cells • Edge distortion • Variable coverage
H3: Hexagonal Grid ───────────────── • Hexagonal cells (better proximity) • Consistent shape • Hierarchical (levels of detail)
─────────────────────────────────────────
H3 Resolution Levels ────────────────────
Level 0: ~1M km² (entire countries) Level 1: ~150k km² Level 2: ~20k km² Level 3: ~2.5k km² Level 4: ~400 km² ← City Level 5: ~50 km² ← Neighborhood Level 6: ~8 km² ← Block Level 7: ~1 km²
─────────────────────────────────────────
Driver Location Storage ──────────────────────
In Redis: ───────── Key: "drivers:index:5:123:456" Value: Set of driver IDs within H3 cell
Query: ───────── 1. Calculate H3 for pickup location 2. Query Redis for nearby cell 3. Expand to neighboring cells 4. Return nearby driversReal-Time Location Updates
Section titled “Real-Time Location Updates” Driver Location Pipeline =====================
┌─────────────────────────────────────────────────────────────┐ │ Driver App │ │ (Every 3 seconds) │ │ │ │ │ │ Location update (GPS) │ │ ▼ │ │ ┌──────────────────────────────────────────────────────┐ │ │ │ WebSocket Connection │ │ │ │ (Persistent, bidirectional) │ │ │ └──────────────────────────────────────────────────────┘ │ │ │ │ │ ▼ │ │ ┌──────────────────────────────────────────────────────┐ │ │ │ Location Ingestion Service │ │ │ │ │ │ │ │ • Validate location │ │ │ │ • Update Redis (driver position) │ │ │ │ • Publish to Kafka │ │ │ └──────────────────────────────────────────────────────┘ │ │ │ │ │ ▼ │ │ ┌──────────────────────────────────────────────────────┐ │ │ │ Kafka Topic: location-updates │ │ │ │ │ │ │ │ Consumers: │ │ │ │ • Dispatch (matching) │ │ │ │ • ETA calculation │ │ │ │ • Analytics │ │ │ └──────────────────────────────────────────────────────┘ │ └─────────────────────────────────────────────────────────────┘48.5 Real-Time Infrastructure
Section titled “48.5 Real-Time Infrastructure”WebSocket Implementation
Section titled “WebSocket Implementation”// Uber's Real-time Communication// Using WebSocket for bidirectional communication
// Client: Driver Appconst ws = new WebSocket('wss://api.uber.com/ws');
ws.onopen = () => { // Register driver ws.send(JSON.stringify({ type: 'driver_register', driver_id: 'driver-123', location: { lat: 37.7749, lng: -122.4194 } }));};
// Receive ride requestws.onmessage = (event) => { const message = JSON.parse(event.data);
if (message.type === 'ride_request') { // Show ride request UI showRideRequest(message.data);
// Auto-expire after 15 seconds setTimeout(() => { if (!accepted) { ws.send({ type: 'expire' }); } }, 15000); }};
// Send location updatessetInterval(() => { ws.send(JSON.stringify({ type: 'location_update', driver_id: 'driver-123', location: getCurrentLocation(), timestamp: Date.now() }));}, 3000);Handling High Concurrency
Section titled “Handling High Concurrency” Scaling Dispatch System ====================
┌─────────────────────────────────────────────────────────────┐ │ Per-City Dispatch │ │ ─────────────────────────────────────────────────────────│ │ │ │ Each city runs independent dispatch cluster │ │ • NYC dispatch → NYC rides only │ │ • SF dispatch → SF rides only │ │ │ │ Benefits: │ │ • Isolation (failure in one city doesn't affect others) │ │ • Lower latency (regional) │ │ • Independent scaling │ └─────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────┐ │ Splitting Large Cities │ │ ─────────────────────────────────────────────────────────│ │ │ │ NYC (high demand) split into multiple dispatch regions: │ │ • Manhattan │ │ • Brooklyn │ │ • Queens │ │ • Bronx │ │ • Staten Island │ └─────────────────────────────────────────────────────────────┘48.6 Key Learnings from Uber
Section titled “48.6 Key Learnings from Uber” Uber Engineering Principles ==========================
1. REAL-TIME FIRST ─────────────── • WebSocket for bi-directional communication • Location tracking every 3 seconds • Sub-second matching requirements
2. GEO-SPATIAL NATIVE ────────────────── • Custom indexing (H3) • Redis for real-time queries • PostGIS for spatial analytics
3. PER-REGION ISOLATION ───────────────────── • Independent dispatch per city/region • Failure isolation • Independent scaling
4. MARKETPLACE DYNAMICS ──────────────────── • Supply/demand balancing • Surge pricing algorithm • Driver incentives
5. TRUST & SAFETY ─────────────── • Real-time ID verification • Trip tracking • Emergency featuresSummary
Section titled “Summary”- Dispatch system - Real-time rider-driver matching
- Geospatial indexing - H3 for efficient location queries
- WebSocket - Real-time bidirectional communication
- Per-region architecture - City-level isolation
- Redis - Fast location lookups
- Kafka - Event-driven location updates