Status checks, not payments, took UPI down

transcript

Half of India's UPI payments failed. For nearly two hours. And the payments weren't the problem at all.

On a normal day, about 99 percent of UPI payments succeed. The twelfth of April, 2025. About 50 percent. Nationwide. A big bank crashing? A rush of payments? Neither.

The culprit was a status check. A cheap, read-only call. Some PSP banks called it excessively. Without waiting for answers. Some re-checked transactions from previous days. Over and over. Here's the catch. The switch wasn't enforcing usage limits on that API. Every extra call got through. Real payments failed with them.

Look where it hit. Questions and payments, through the same switch. Cheap per call isn't cheap in a flood. The checks swamped the capacity real payments needed.

Afterwards, NPCI added rate limiters. The lessons go further. One: rate-limit every API, at the switch. Don't trust callers. Two: keep status checks off the critical payment path. Three: breakers per API, not just per bank.

Here's the irony. Status checks exist to settle stuck payments. The tool that resolves doubt must never become the flood. Status checks, not payments, took UPI down.

Full breakdown in the lecture.