How Swiggy prepares for its busiest night of the year

transcript

400 people. One office. One night. 31 December. Swiggy's busiest day of the year. Here's how they get ready.

8 to 10 weeks out, planning starts. Forecast orders, hour by hour. Find the peak orders per minute. One number. Turn it into requests per minute, for every backend service. Then scale each service up for those hours.

Critical caches and databases move from one zone to several. A continuity plan: which levers to pull if something breaks. And a deploy freeze. No new code on the big day.

New Year's Eve. About 23 teams are in the office. Over 950 people in one Slack channel, called war room. Seven dashboards. 46 metrics, in near-real time. Decisions on the spot.

Around 3 pm, orders start climbing. But here's the catch. Even with all that prep, things still broke.

As orders surge, UPI payments go down. Customers can't pay. The payments team quickly sets up other ways to pay. Then the feature telling riders an order is ready stops working. 492 care staff on duty. A few thousand more, remote.

The verdict? They didn't lose a lot of orders. A known peak is a project. Not a surprise. Forecast it. Scale it. Freeze it. And plan for what breaks.

Full breakdown in the lecture.