πŸ’¬
βœ‰οΈ
πŸ“Έ
▢️

Loading…

Design Uber: location, matching, and trip service

UberGuided Mode

Step 4 β€” Geo Index (Redis Geo)

Request Flow

1

GEOADD on ping

Location Service calls GEOADD fleet:NYC <lng> <lat> driver_id. Redis converts coordinates to a geohash integer score and stores driver_id in a sorted set. Cost: ~1 Β΅s per write. Capacity: 200K writes/second per instance.

2

GEORADIUS query

Matching Service will call GEORADIUS fleet:NYC <lng> <lat> 3 km ASC COUNT 20 on the Cache. Redis checks the center geohash cell + 8 surrounding cells β€” returns 20 nearest drivers in <2ms, even with 1M drivers in memory.

3

Why not PostGIS?

PostGIS with a GIST index handles ~10-20K location updates/second. Redis Geo handles 200K+/second. For radius queries, PostGIS takes ~10ms; Redis takes <2ms. 100x improvement in both write throughput and query latency.

Why We Need This

Geohash converts 2D space into a 1D sorted key. Nearby points share geohash prefixes. A radius search becomes: find all points in the 9 geohash cells surrounding the query point β€” O(9 sorted-set range scans) instead of O(n) distance computations.

Key Insight

Geohash turns nearest-neighbour search from O(n) to O(1). A 3 km radius at precision-6 covers 9 cells of ~1.2 kmΒ² each β€” checking ~20-50 drivers instead of 100,000.

Step 4 of 8