Maple Stage asks for a launch design for a fictional 8,000-seat venue and 60,000 checkout attempts in one minute. Deliver a waiting-room board, rate estimate, seat model and checked recommendation. Prove that concurrent buyers can't share a seat and a successful payment after hold expiry can't take the reservation held by a later buyer.
Maple Stage sells tickets for a fictional 8,000-seat venue. The launch brings 60,000 checkout attempts in one minute, and many buyers choose the same front row. Adding random shards can distribute request traffic, but it can't let two shards independently sell one seat. Keep seat ownership serialized.
Use a waiting room or admission limit to bound the work reaching inventory. A hold needs an expiry checked by the authority, including when a delayed payment result arrives after expiry. Payment and seat ownership are separate state machines. You'll size admitted traffic, model a conditional hold and choose how a late payment is reconciled. The final recommendation must protect inventory even when that means a buyer waits or receives a refund.