Hard · 13 steps · ~85 min

Design a delivery dispatch system

Pick the right rider for every order, as riders and orders move.

Customers place orders, thousands of riders move around a city, and the system must pair each order with a good free rider within seconds, never give one order to two riders, and keep working at the dinner rush.

Same problem as SwiggyDoorDashUber Eats

How it works

• Riders stream their location, the system keeps an index of where everyone is.
• A new order triggers matching: find nearby free riders, offer the order, and record whoever accepts.

What you'll build

1. The API and the data behind it
2. A system for 400 requests a second, then a cheap design for 5,000 location pings
3. A durable queue and workers that absorb the dinner rush
4. A realtime tier, replicas and a limiter for 40,000+
5. A design that survives losing a gateway, a cold cache and the primary database
6. The matching decisions: batching, atomic claims and geo-indexing

Steps

1. 1Design the API
2. 2Design the data model
3. 3Handle 400 requests/sec
4. 4Rider pings at 5,000 requests/sec for $800 a month
5. 5Absorb the dinner rush: 12,000 requests/sec
6. 6Handle 40,000 requests/sec
7. 7Survive a retry storm on the same budget
8. 8Lose a realtime gateway
9. 9Survive a cold position cache
10. 10Lose the primary database
11. 11Match the nearest rider, or batch?
12. 12Two riders got the same order
13. 13Deep dive: finding riders near a point