Inside Swiggy's New Year's Eve war room.

For Swiggy, the food and grocery delivery app, 31 December is the busiest day of the year. On New Year's Eve 2022, more than 400 employees gathered at its Bengaluru headquarters, backed by more than 3 lakh (300,000) delivery partners, for customers in 540 cities.

The night started weeks earlier. Engineering readiness planning began 8–10 weeks ahead. The team forecast the day's orders hour by hour to estimate the peak orders per minute, and used that number to work out how many requests per minute each backend service had to handle, then scaled the services up for those hours. They moved critical caches and databases from a single availability zone (one isolated data center location) to several. They wrote a business continuity plan: which levers to pull if something broke before or after an order was placed. And they froze deployments, so nobody shipped new code on the big day.

On the evening itself, about 23 teams from engineering, analytics, data science and operations were in the office. Orders started climbing around 3 pm. More than 950 employees coordinated in a single Slack channel called #warroom. The analytics team had built seven dashboards with 46 metrics in near-real time so leaders could make decisions on the spot.

Things still broke. As orders surged, UPI payments went down and customers couldn't pay, so the checkout and payments team quickly set up other ways to pay. A feature that lets restaurants tell riders an order is ready stopped working. At another Swiggy office, 492 customer-care staff were on duty, with a few thousand more supporting remotely. The analytics lead said they "didn't lose a lot of orders".

Why it matters: A known peak is a project, not a surprise: forecast it, turn the forecast into per-service capacity, freeze changes, and plan for what will break, because something always does.