Design a Taxi Hailing App (Uber)

Core Challenges

• Location updates — drivers send GPS every 4 seconds; 1M drivers = 250K location updates/sec
• Geospatial queries — find all drivers within 5km of a rider in < 100ms
• Matching — optimally match rider to nearest available driver
• Real-time ETA — live traffic-aware routing; update ETA as driver moves

Location Tracking Architecture

// Driver app sends location every 4 seconds
app.post('/driver/location', async (req, res) => {
  const { driverId, lat, lng } = req.body;

  // Redis GEO commands provide geospatial indexing
  // Uses a sorted set with geohash as score
  await redis.geoadd('drivers:active', lng, lat, driverId);

  // Expire driver from index if no update in 30 seconds
  await redis.expire(`driver:heartbeat:${driverId}`, 30);

  res.sendStatus(200);
});

// Rider requests a ride — find nearby drivers
async function findNearbyDrivers(riderLat: number, riderLng: number) {
  // GEORADIUS: find all drivers within 5km
  return redis.georadius(
    'drivers:active', riderLng, riderLat,
    5, 'km', 'WITHCOORD', 'WITHDIST', 'ASC', 'COUNT', 10
  );
}

Matching Algorithm

Uber uses a matching engine that considers: proximity, driver rating, ETA to rider, and driver heading (a driver already moving towards you is preferred). The matching service batches requests every 500ms to find globally optimal assignments.

Surge Pricing

H3 hexagonal grid divides cities into cells. When the ratio of rider requests to available drivers in a cell exceeds a threshold, surge multiplier is applied. Surge data is precomputed every 30 seconds and cached in Redis — it's read-heavy and eventual consistency is acceptable.