Staying online through the Dashain shopping rush

Annapurna runs 14 shops and an online store. During Dashain, traffic rises about twelve-fold over two weeks. The old platform had crashed at the peak two years running, each time on the most profitable days of the year.

The cause was not the number of visitors. Every page view checked live stock levels in a single database table, while all 14 shops were constantly updating the same table, and the two kept blocking each other.

We separated reading stock levels from updating them. Shoppers now see stock levels from a fast cache that is updated continuously, and stock is reserved, through a queue, only when someone reaches the checkout.

• A catalogue optimised for browsing, with stock levels for each shop held in Redis.
• A reservation queue that holds items for 15 minutes during checkout.
• Load tests that replayed the previous year’s real traffic, rather than an artificial pattern.
• A supplier portal where vendors update their own stock and prices.

The new platform handled the next Dashain with a peak of 14,200 simultaneous visitors and no downtime. The share of shoppers completing checkout rose from 61% to 78%, mostly because the delay caused by stock checks on the basket page had gone.