How would you design Amazon's category sales ranking?

Question

Accepted Answer

## Step 1: Use Cases, Constraints, Assumptions, Back-of-Envelope

### Use cases
We'll scope to handle only the core flows that expose the bottlenecks; everything else is deferred.

- **Service** ingests orders — order events (item, category, qty) into durable log (Kinesis/Kafka).
- **Service** aggregates rank — sliding window (hourly/daily) count per category, sorted set, rank = position in sorted order.
- **User** reads rank — GET `/category/{id}/rank?item_id=...`, <50ms p95, stale <1m acceptable.
- **Service** backfills — reprocess historic log for corrections, idempotent upserts.

### Out of scope
- Pricing, inventory, recommendations, fraud detection (separate consumers of order stream).

### State assumptions
- 1B items, 10k categories, 50M orders/day, avg 2 items/order → 100M item-sales/day.
- Ingest QPS: 100M/86k≈1,160 events/s avg, 10k peak (Prime Day 5x).
- Rank update QPS: per-category sort on write is heavy → windowed aggregation 1/s per category.
- Storage: events 100M*200B

How would you design Amazon's category sales ranking?

The problem

Try it yourself

Solution

Step 1: Use Cases, Constraints, Assumptions, Back-of-Envelope

Use cases

The solution is hidden so you can try first

Discussion

Keep your momentum going.