Uber Backend — Production-Grade Java Microservices

Architecture

                        ┌──────────────────────────────────────────────────────┐
                        │                    API Gateway :8080                  │
                        │   JWT Auth • Rate Limiting • Circuit Breaker • CORS   │
                        └────────────────────────┬─────────────────────────────┘
                                                  │
              ┌───────────┬───────────┬───────────┼───────────┬───────────┬───────────┐
              ▼           ▼           ▼           ▼           ▼           ▼           ▼
        Driver:8081  Rider:8082  Trip:8083  Payment:8084  Notif:8085  Location:8086  ETA:8088
              │           │           │           │           │           │
              └───────────┴───────────┴───────────┴───────────┴───────────┘
                                                  │
                    ┌─────────────────────────────┼───────────────────────────┐
                    ▼                             ▼                           ▼
              Kafka Cluster                   Redis Cluster             PostgreSQL
         (12+ partitions/topic)          (GEO + Hash + ZSet)      (4 separate DBs)

Services

Service	Port	DB	Description
api-gateway	8080	—	JWT auth, rate limiting, routing, circuit breaking
driver-service	8081	PostgreSQL	Driver CRUD, status, trip acceptance
rider-service	8082	PostgreSQL	Rider CRUD, payment methods, ratings
trip-service	8083	PostgreSQL	Trip lifecycle state machine
payment-service	8084	PostgreSQL	Stripe integration, idempotency, retries
notification-service	8085	—	FCM push, Twilio SMS, WebSocket
location-service	8086	—	Redis GEO, GeoHash indexing, WebSocket streaming
matching-service	8087	—	Hungarian algo, surge pricing, driver scoring
eta-service	8088	—	Dijkstra, A*, traffic-aware ETA with cache

Key Algorithms

GeoSpatial Indexing

• Redis GEO commands (GEOADD, GEORADIUS) — O(N+log M) nearest neighbor search
• GeoHash dual-index for O(1) cell lookup + neighbor expansion
• 6-character precision (~1.2km cells), 8-character neighbor expansion

Driver Matching

• Multi-factor scoring: distance (40%) + rating (25%) + acceptance rate (15%) + trip count (10%) + wait time (10%)
• Sequential offering: top-scored driver offered first, 30s timeout, then next
• Hungarian Algorithm: batch optimal 1:1 assignment for surge periods (O(n³))
• Distributed locking: Redis SETNX prevents race conditions across instances

Surge Pricing

• Demand/supply ratio per GeoHash cell (5-char ≈ 5km cells)
• log curve: multiplier = 1 + log(ratio/threshold) × 2
• Snapped to nearest 0.1x, capped at 8.0x
• 5-min rolling window, 2-min multiplier cache

ETA

• Haversine + road factor heuristic for fast estimates
• Dijkstra for exact shortest-path on road graph
• A* for path reconstruction with geographic heuristic
• Traffic multiplier: 1.5× peak hours, 0.8× late night
• Grid-snapped cache key (0.001° ≈ 100m) with 60s TTL

Trip State Machine

REQUESTED → DRIVER_ASSIGNED → DRIVER_ARRIVING → DRIVER_ARRIVED → IN_PROGRESS → PAYMENT_PENDING → COMPLETED
     ↓              ↓                ↓                 ↓               ↓
  CANCELLED      CANCELLED       CANCELLED         CANCELLED        (never)

Kafka Topics

Topic	Partitions	Retention	Publisher	Consumers
trip.requests	12	24h	trip-service	matching-service
driver.matched	12	24h	matching-service	trip-service, notification-service
trip.status	12	24h	trip-service	notification-service, rider-service
location.updates	24	1h	driver-service	location-service
payment.events	6	7d	trip-service, payment-service	payment-service, notification-service
notifications	6	24h	any service	notification-service
driver.availability	12	1h	driver-service	matching-service
surge.pricing	6	24h	matching-service	—

Redis Key Patterns

Pattern	Type	TTL	Usage
drivers:geo	GEO	—	All active driver coordinates
driver:location:{id}	Hash	30s	lat/lon/ts/geohash per driver
geohash:drivers:{hash}	Set	—	Driver IDs per GeoHash cell
driver:active:{id}	String	4h	Lock: driver on trip
driver:metrics:{id}	Hash	24h	rating/trips/acceptance
match:pending:{tripId}	Hash	5m	In-flight match state
surge:demand:{hash}	String	5m	Ride request count
surge:supply:{hash}	String	5m	Available driver count
surge:multiplier:{hash}	String	2m	Computed surge
eta:cache:{key}	String	60s	Grid-snapped ETA result
rate:{userId}:{minute}	String	2m	Rate limit counter
payment:lock:{tripId}	String	5m	Idempotency lock
rider:payment:{id}	String	—	Default payment method

Running

# Full stack with Docker Compose
docker-compose up -d

# Individual service
cd driver-service && mvn spring-boot:run

# Kubernetes
kubectl apply -f k8s/namespace.yaml
kubectl apply -f k8s/infra.yaml
kubectl apply -f k8s/deployments.yaml

API Endpoints

Rider flow

POST /api/riders                          # Register rider
PUT  /api/riders/{id}/payment-method      # Set payment method
POST /api/trips                           # Request trip
GET  /api/eta/estimate?originLat=...      # Get fare + ETA estimate
WS   /ws/notifications                    # Subscribe to real-time updates
WS   /ws/location → /topic/trip/{id}/location  # Track driver

Driver flow

POST   /api/drivers                        # Register driver
PATCH  /api/drivers/{id}/online            # Go online
PUT    /api/drivers/{id}/location          # Push location update
POST   /api/drivers/{id}/trips/{id}/accept # Accept trip offer
POST   /api/drivers/{id}/trips/{id}/decline
PATCH  /api/trips/{id}/status             # Update trip status

Trip lifecycle

POST  /api/trips                           # Create (REQUESTED)
PATCH /api/trips/{id}/status {"status":"DRIVER_ARRIVED"}
PATCH /api/trips/{id}/status {"status":"IN_PROGRESS"}
PATCH /api/trips/{id}/status {"status":"PAYMENT_PENDING"}
PATCH /api/trips/{id}/status {"status":"COMPLETED"}
POST  /api/trips/{id}/cancel
POST  /api/trips/{id}/rate