System Design

Ride Matching

Design the ride-hailing dispatch and matching system for a service like Uber or Lyft.

Functional requirements

• Riders request a ride from a pickup location and receive a match with a nearby available driver, along with an ETA and a price quote.
• Drivers continuously report their location and availability; the system tracks them in real time.
• The system matches a ride request to the best available driver (proximity/ETA-based) and handles the driver's accept/decline within a timeout, falling back to the next candidate.
• Dynamic (surge) pricing adjusts the quoted price per geographic zone based on real-time supply and demand, and the quoted price is honored for the duration of the trip.
• Riders and drivers see each other's live location from match acceptance through pickup.

Non-functional requirements

• Matching latency: a rider should receive a driver match (or a 'no drivers available' response) within ~10 seconds end-to-end; nearby-driver lookup itself should complete in <200ms.
• Consistency for matching: a driver must never be double-assigned to two concurrent ride requests, even if that costs some additional latency on the assignment step.
• Availability bias elsewhere: location ingestion and ETA/price display can tolerate slightly stale data (seconds); prefer availability for these read paths.
• Location update ingestion must handle sustained high write throughput without backpressure onto driver apps; losing an occasional individual ping is acceptable, losing a driver's presence entirely is not.
• Trip and payment-relevant records (accepted quote, match, trip state transitions) must be durable and auditable.

Scaling & constraints

• 10 million riders DAU and 1 million drivers DAU globally, concentrated in ~500 cities with heavy skew toward the top 20 metros.
• Peak concurrent online drivers: ~300,000, each sending a location update every 4 seconds while online.
• ~15 million ride requests per day, with rush-hour peaks at roughly 5x the daily average rate in a given city.
• Each location update is ~100 bytes; raw location history must be retained for 30 days (safety/audit), trip records retained indefinitely.
• Roughly 20 rider-side reads (nearby-driver map views, ETA checks, price quotes) for every ride actually requested.
• Assume 20% year-over-year growth in both riders and drivers.

Out of scope

• Payment processing, fraud detection, and payout to drivers.
• Routing/navigation engine internals (assume an ETA service exists that returns travel time between two points).
• Ratings, reviews, and driver onboarding/background checks.
• Pool/shared-ride matching (single rider to single driver only).