// 01 — The problem
What's wrong with Ticketmaster
Ticketmaster is the world's leading ticketing company. Yet for fans, every big on-sale feels like an ordeal. Here's what I hold against their system.
Sites that give up
Blank pages, errors and slowdowns at the exact moment everyone logs in.
An opaque queue
You wait without knowing whether you stand a real chance, only to find out there's nothing left.
The expiring cart
The countdown hits zero during bank verification, and your seats are gone.
"Sold out" way too late
You learn there are no seats left after a long wait, sometimes at checkout.
Bots and scalpers
Seats grabbed by automated scripts, then resold for far more.
A lottery in disguise
"First click wins" comes down to milliseconds: your connection decides, not you.
Badly split allocations
Sold out at one reseller while another still has seats for the same show.
No competition
The same group promotes the tours and sells the tickets. The US Department of Justice filed an antitrust lawsuit in 2024.
What's broken under the hood
I took apart a checkout page running on Ticketmaster France's infrastructure, in September 2026. Here's what the architecture gets wrong.
- A blank page first. The server sends an almost empty shell and the whole page is built in your browser. Until the JavaScript finishes, you stare at white.
- API calls in a chain. Each request waits for the previous one before starting. The server is fast, but prices still take seconds to appear.
- Every visitor hits the inventory. Each page load fires dozens of API calls, many of which can't be cached. Multiply that by a million fans and the shared ticketing system chokes.
- The waiting room shows up late. The virtual queue only loads after the rest of the page. It's supposed to protect the system, not arrive after the crowd.
- Things nobody asked for. Menus in every language, lists of cities and music genres, cross-selling suggestions: all loaded before or alongside the prices.
- Heavy for no reason. Hundreds of requests, JavaScript split into far too many files, oversized images and a CAPTCHA loaded twice.
// 02 — It's been done
Other industries already handle worse
Selling 200,000 seats in a few minutes isn't science fiction. Every year, online retail, payments and streaming absorb much bigger rushes, and they hold.
Selling out 200,000 seats means about 1,000 reservations per second: less than 0.2% of Alibaba's peak. The real issue isn't server power. It's how the sale is designed:
- A randomized queue. Everyone arrives before the opening, the order is drawn at random, and each person sees their real position.
- Admission matched to inventory. Only as many people get in as the remaining seats can serve.
- Seats locked before payment. Once your card is out, your seats are yours. No "sold out" at checkout.
- Shared inventory, no quotas. An open registry lets several resellers sell the same seats without ever selling one twice.
- An instant "sold out". The moment everything is gone, the whole queue knows within seconds.
// 03 — The proof
The live stress test
I'm designing a ticketing platform built for these moments, and I'll put it to the test in public, with a real simulated sale. No real tickets, no real payments.
| Test conditions | Value |
|---|---|
| Tour on sale (fictional) | Ticketmaster's Retirement Plan Tour |
| People online at the same time | 200,000 |
| Seats on sale (fake, 4 categories) | 20,000 |
| Simulated resellers sharing the same inventory, no quotas | 3 |
| Share of traffic from simulated bots | 20% |
| Declined payments (bank test mode) | 5% |
| Buyers who give up halfway | 10% |
| Participants | Sign-ups + virtual users |
| Success criteria, published in advance | Target |
|---|---|
| Seats sold twice | 0 |
| Server errors seen by participants | 0 |
| Real queue position displayed | < 2 s |
| Reservation time (99% of requests) | < 200 ms |
| "No seats left" announced at checkout | Never |
| Time to sell every seat | < 10 min |
| "Sold out" shown to the whole queue | < 5 s |
Raw results, load-testing scripts and the registry code will be published as open source, whether the test succeeds or not.