CASE 01
G-Dragon concert ticket sale
Read case studyShow less
- Problem
- In October 2025, booking.vinwonders.com opened ticket sales for the G-Dragon concert, with 600,000 people trying to buy at the same time. Traffic hit in a single moment, tickets were limited and sold fast: the system had to absorb the spike, never oversell, and keep other booking services running normally.
- Approach
- Keep the 600,000 people from reaching the transaction layer directly. A virtual waiting room admitted buyers at the capacity measured by load testing, with a scenario at 1.5x the expected load. Tickets were held before payment and released automatically on expiry; hold operations were idempotent so retries could not deduct a ticket twice. Layered bot protection sat at the entrance, and read-heavy data was cached to keep it off the database.
- Trade-offs
- Reliability traded against wait time: buyers queued longer, but anyone admitted to the purchase flow could finish reliably. On scalability, stateless services could scale out, but the database could not scale in time for the spike, so it was protected by limiting inflow. Expired holds returning to stock made the remaining count fluctuate, an accepted cost for never overselling.
- My role
- Led the team behind the booking platform: capacity planning, architecture review of the purchase flow, coordinating preparation across teams, owning the release, and monitoring the system during the sale.
- Result
- Sales reached about 100x a normal day. Tickets sold out in about 30 minutes. Uptime reached 99.9% during the sale window, with 0 orders oversold.
