RETURN TO SYSTEM DESIGN VAULT (55)BLUEPRINT #25 // Very Hard
#25Very HardLocation, Geospatial & Ride-Hailing

Surge Pricing System: Uber - Stream Processing, etc.

ASKED IN INTERVIEWS AT:UberLyft

EXECUTIVE SUMMARY & PROBLEM STATEMENT

Design Uber's dynamic surge pricing engine analyzing real-time supply (available drivers) and demand (app opens & ride requests) in geospatial hexagon cells (H3).

SCALE ESTIMATION & TRAFFIC ENVELOPE

DAILY ACTIVE USERS / TPSMillions of concurrent app pings and driver telemetry pings
READ / WRITE TRAFFIC RATIOContinuous stream processing
5-YEAR STORAGE CAPACITYFast spatial memory grid in Redis Cluster

FUNCTIONAL REQUIREMENTS

  • [1]Calculate surge multiplier per H3 hexagon cell every 15-30 seconds
  • [2]Lock surge multiplier for user for 2-5 minutes once quote is requested

NON-FUNCTIONAL REQUIREMENTS

  • •Calculate city-wide surge in < 15 seconds
  • •Zero sudden cliff drops across adjacent hexagons (spatial smoothing)
  • •High resilience

CORE ARCHITECTURAL COMPONENTS

Driver Location Stream (Kafka)
Rider Request Stream
Flink Complex Event Processor
Surge Multiplier Cache (Redis)
Quote Pinning Service

DATA SCHEMA & PERSISTENCE

SurgeCell: h3_index, supply_count, demand_count, multiplier, updated_at

DISTRIBUTED TRADE-OFFS & FAILURE MODES

  • ⟡Spatial smoothing algorithms prevent drivers from gaming the system by waiting on border lines between adjacent hexagons.