RETURN TO SYSTEM DESIGN VAULT (55)BLUEPRINT #16 // Very Hard
#16Very HardLocation, Geospatial & Ride-Hailing

Design Yelp or Nearby Friends

ASKED IN INTERVIEWS AT:AmazonMetaGoogleUber

EXECUTIVE SUMMARY & PROBLEM STATEMENT

Design a geospatial search engine capable of locating nearby places or friends within a radius using spatial indexing algorithms (QuadTrees, Geohash, Google S2).

SCALE ESTIMATION & TRAFFIC ENVELOPE

DAILY ACTIVE USERS / TPS100 Million DAU, 50,000 spatial queries/sec
READ / WRITE TRAFFIC RATIOYelp is 99% Read-heavy; Nearby Friends is 50/50 write/read
5-YEAR STORAGE CAPACITY50 Million places with coordinates ~ 5 GB spatial index in memory

FUNCTIONAL REQUIREMENTS

  • [1]Search places/friends within R kilometers radius
  • [2]Filter places by rating, category, and open status
  • [3]Update user live location (for Nearby Friends)

NON-FUNCTIONAL REQUIREMENTS

  • •Query response latency < 50ms
  • •High throughput spatial lookups
  • •Smooth handling of high-density urban areas

CORE ARCHITECTURAL COMPONENTS

Location Update Gateway
Geospatial Index (Redis Geo / PostGIS / QuadTree Node Cache)
Aggregation Filter Worker
CDN Map Tiles

DATA SCHEMA & PERSISTENCE

Places: place_id, name, lat, lng, geohash (char(8)), rating

DISTRIBUTED TRADE-OFFS & FAILURE MODES

  • ⟡Geohash (base32 string prefixes, easy B-Tree indexing) vs QuadTree (hierarchical 2D grid, dynamically splits dense cities).