Case study

DropZone

Ticket sales that survive the rush

0 double-sold seats2,000 users for 1,500 seats, in tests

Live demo

No public demo yet.

What it is and what it solves

Ticket sales built for the moment thousands of people click at once: a waiting room, 5-minute seat holds, one seat per buyer guaranteed by the database, and card payments with manual capture and 3D Secure.

Business version: Reservas →

Architecture

DropZone: architectureholdconfirmBuyersbrowserWaiting roomanti-botCheckout APIRedis + Lua5-min holds (TTL)PostgreSQLunique seat indexStripemanual capture · 3DSKafkaoutbox
  • Buyers → Waiting room
  • Waiting room → Checkout API
  • Checkout API → Redis + Lua (hold)
  • Checkout API → PostgreSQL (confirm)
  • Checkout API → Stripe
  • PostgreSQL → Kafka

Simplified from the project description; the repository is private.

Three technical decisions

  1. Seat holds in Redis + Lua

    A Lua script checks and reserves a seat atomically with a 5-minute TTL, so abandoned carts free seats on their own.

  2. A unique index as the last line of defence

    Even if every other layer failed, PostgreSQL refuses a second sale of the same seat.

  3. Manual capture with 3D Secure

    The card is authorised first and charged only once the seat is confirmed, so nobody pays for a seat they did not get (verified in tests).

Tests, with numbers

  • 500 simultaneous purchases with 0 duplicates.
  • 2,000 users against 1,500 seats with 0% errors.

Verified

  • Anti-bot measures and Kafka outbox in place.

Not claimed

  • Figures come from test runs, not from production traffic.

Stack

  • NestJS
  • Next.js
  • Redis + Lua
  • PostgreSQL
  • Stripe
  • Kafka

Private repository · code walkthrough on request