200,000 tickets in an hour. 400 agents.

transcript

Friday, 8:10 pm. A food app's payments break. In one hour, 200,000 support tickets. 400 agents online. The candidate: let me do the division before I design.

56 tickets a second from one tenant. The whole platform's normal peak is 42. 130,000 customers over 400 agents. About 325 each. And a one-hour target. About 130,000 SLA breaches, each an escalation.

WhatsApp allows 80 messages a second. 100,000 acks take 21 minutes. And every real agent reply waits behind them. Keep 20 a second for replies. Acks now take 28 minutes.

First, per-tenant quotas. Her backlog grows. Nobody else's does. Merge: one open ticket per customer. 200,000 become about 130,000. Incident mode: children follow the parent's SLA clock. No 130,000 escalations.

The customer writes: money debited, order not placed. They get one honest message. We know, and refunds are automatic. When it's fixed, one resolution message. Through the same paced lane.

The interviewer asks: does the system declare the incident? It suggests. The lead confirms. Pausing 130,000 SLA clocks needs a name.

Every fix came from doing the division first. Do the division before you design anything.

Full breakdown in the lecture.