Fable Books survived a 40x traffic spike on launch day

An independent publisher with a modest, steady audience had one title picked up by a national broadcaster. They had four hours of warning.

The problem

Their usual traffic was around 2,000 visits a day on a small shared plan. The broadcast segment put 80,000 people on the site in ninety minutes, most of them landing on the same product page.

What we did

The site was already on Canvas, so the fix was a vertical resize and a cache rule rather than a migration. We scaled the instance from two cores to eight, forced full-page caching on the product page for anonymous visitors, and put a load balancer in front with a second node.

• Peak concurrent visitors: 6,400
• Requests served: 1.2 million in two hours
• Errors returned: none
• Time from call to scaled: 11 minutes

Afterwards

They scaled back down to a CX21 two days later and were billed for the hours at the larger size, which came to under nine dollars.

In their words

"We spent four hours panicking and eleven minutes fixing it. The bill for the biggest day in our history was less than lunch."