System Architecture & Scaling

High-Level Architecture Flowchart

---

100x Scale-Up & The First 3 Bottlenecks

When traffic scales from 1,000 RPS to 100,000 RPS:

1. Bottleneck 1: Database Row Contention on Hot SKU: 100,000 requests attempting to update the same single database row will hit row-level write lock queues.
   • Fix: Offload inventory counters to Redis Lua, maintaining in-memory stock counters and flushing to PostgreSQL in asynchronous batched commits.
2. Bottleneck 2: Virtual Waiting Room Memory Footprint: Storing 1,000,000 queued user tickets in FastAPI process memory causes high RAM usage and prevents horizontal scaling.
   • Fix: Move queue state to Redis Sorted Sets (ZADD, ZRANGEBYSCORE) with user timestamps as scores.
3. Bottleneck 3: Third-Party Payment Gateway Rate Limits: Payment gateways (e.g. Stripe) cap API throughput at 100-500 requests per second.
   • Fix: Decouple payment submission via an asynchronous message queue (Apache Kafka or AWS SQS) and process payments via consumer worker pools.