Online feature store

Model 02
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

Turn every goal green.

Traffic
Lookups10,000 req/s · reads onlyOrders300 req/s · writes only
Limit
Postgres answers about 7,500 reads/s
Bill
  • API server16 USD/mo
  • Postgres8 USD/mo
24 of 50 USD/mo

Goals

Not tested yet

99 of 100 requests finish faster than this.The share of requests that fail or get no answer.How old the data is when a reader gets it.What the servers and services cost each month.

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.

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

Parts you can add: Redis cache, Memory cache, Change capture. At most 6 parts. Budget: 50 USD/mo.

Tap a part to add it, or drag it onto a trace.Click a part to add it, or drag it onto a trace.