Online feature store

  • Build
  • Model 02

Difficulty 2 of 5

A fraud model reads 10,000 user features/s. Today it reads them from the offline order database, which keeps history for training and takes 300 new orders/s. Serve lookups with p99 under 10 ms. A value may lag new orders by at most 60 s. Budget 50 USD/mo.

Play level

SQLHTTP/2SQLUsers300 req/sPartner apps10,000 req/sAPI server4 vCPU · 4 GBPostgres0.5 vCPU · 2 GB
SQLHTTP/2SQLUsers300 req/sPartner ap…10,000 req/sAPI server4 vCPU · 4 GBPostgres
The board at the start of the level
Each worker adds one network cardDrag or use arrow keys to change the workers

2 workers · 1 TB moves in 1.1 h · 1,000 Mbps per card

Goals and limits

p99 latency (Lookups)
under 10 ms
Error rate (Lookups)
under 0.1%
Data age
under 60 s
Monthly cost
under 50 USD/mo
Parts you can add
Redis cache, Memory cache, Change capture
Part limit
6
Traffic
10,000 req/s

Concepts

How players did

Player results load from the server.

All levels