Ride Sharing

Designing a service like Uber/Lyft: matching, dispatch, pricing, and tracking.

Requirements

• Request → match → accept → pickup → dropoff → complete
• Real-time driver locations; surge pricing
• ETAs, cancellations, and reassignments

High-Level Design

• Matching on proximity, driver supply, and constraints
• Trip lifecycle managed by state machine; idempotent updates
• Location updates via streaming ingestion; geo-indexed store

Capacity & Sizing

• Driver location updates/sec and ingest throughput
• Trip TPS at peak, surge factors by region
• Geo index size and query QPS

Key Components

• Matching service, Dispatch queue, Driver pool
• Trip service with state machine
• Location service and geo-index (R-tree/Geohash)

Data Model

• trips (trip_id PK, rider_id, driver_id, status, origin, destination, fare, created_at)
• driver_locations (driver_id PK, lat, lng, updated_at)
• trip_events (event_id PK, trip_id, type, ts)

APIs

• Request ride: POST /api/trips with body { origin, destination }
• Driver update location: PUT /api/drivers/me/location with body { lat, lng }
• Trip status: PUT /api/trips/:id/status with body { status }

Hot Path

1. Rider request → price/ETA → dispatch offer to nearest drivers
2. Accept → start trip; update location stream → ETA updates

Matching Flow

1. Ingest rider request; compute price and ETA
2. Find nearest eligible drivers; dispatch offers with timeouts
3. Driver accepts; confirm and start trip; reassign on timeout

Scaling

• Partition geographies into cells; maintain driver pools per cell
• Geo-index for proximity queries; stream processing for locations
• Backpressure on dispatch queue with prioritization

Caching & TTL

• Short TTL cache for driver locations by cell
• Cache static map tiles and traffic segments

Failure Modes & Mitigations

• Location gaps → extrapolation and last-known fallbacks
• Driver scarcity → surge pricing and queueing
• Provider outages → degrade to simpler matching rules

Observability

• Match success rate, time-to-assign, cancel/reassign rate
• Driver online counts, location ingest lag