Skip to content
Journal

Engineering

Getting Sites Ready for a World Cup of Traffic

Headshot of Grace Nakamura
Grace Nakamura
June 10, 2026 · 3 min read

Tomorrow afternoon, the World Cup kicks off in Mexico City, and for the next six weeks, tens of millions of people across this continent will reach for their phones in the same synchronized moments: at full time, after every goal, at halftime of the matches everyone watches at work. If your brand has anything planned this summer, promotions, drops, reactive campaigns, or just a normal e-commerce business that happens to sell things people want in summer, your infrastructure is about to experience the spikiest traffic pattern of its life.

We have spent the spring running readiness work for clients, including a substantial program with Northwind's retail platform, and the checklist has settled into a shape worth sharing. None of it is heroic. All of it is the difference between a summer of opportunity and an incident channel full of screenshots.

Respect the synchronized thundering herd

Tournament traffic is not big so much as it is instantaneous. A match ends, a push notification lands, a promo goes live, and you receive an hour of normal traffic in ninety seconds. Systems sized for averages die at the p99.9, and autoscaling that takes three minutes to react is a postmortem with extra steps.

The architectural answer is boring and proven: put everything you possibly can on the edge, and make origin traffic an exception rather than the rule.

  • Static-first pages for anything campaign-related. A reactive campaign page should be prebuilt and cached globally, with dynamic elements isolated into small islands that degrade gracefully.
  • Cache with intent. We audit every major route and assign an explicit caching policy. The most common finding is pages that are 99 percent cacheable being served dynamically because of one personalized widget nobody would miss.
  • Queue the truly dynamic. Checkout and inventory need an origin. Protect them with a queue that admits honestly rather than failing mysteriously. Users forgive a styled thirty-second wait. They do not forgive a spinner that lies.
# fail soft: serve stale while origin struggles
proxy_cache_use_stale error timeout updating http_500 http_502 http_503;
proxy_cache_background_update on;

Those two lines, or their equivalent in your CDN's dialect, have saved more summer campaigns than any capacity plan.

Degrade by design, not by accident

Every system has a behavior under overload. The only question is whether you chose it. We build clients an explicit degradation ladder: which features switch off, in what order, triggered by what signals, owned by which human. Recommendations off first, then search sorting simplifications, then personalization, with core browsing and checkout defended to the last core. Written down, rehearsed, and wired to feature flags that marketing is not allowed to touch.

The rehearsal matters more than the document. We ran load drills with each client in May, generating match-end spike shapes against staging, and every single drill found something: a connection pool sized in 2024, an image resizer without a cache in front of it, a third-party tag that blocked rendering while its vendor melted. Third-party scripts deserve special paranoia this summer. Your uptime is currently the minimum of every vendor in your tag manager, and their infrastructure is about to have a World Cup too.

The performance dividend

Here is the part I enjoy. Almost everything on this checklist is just aggressive web performance discipline with a deadline attached. Edge-served pages, minimal blocking JavaScript, and honest caching do not only survive spikes. They make the site faster every ordinary day, cheaper to run, and more legible to the crawlers and agents that increasingly mediate discovery. Two clients came for resilience and left with their best Core Web Vitals in company history.

Kickoff is tomorrow. If you are prepared, the next six weeks are a free lesson in what your infrastructure can really do. If you are not, the good news is that the tournament runs until mid-July, and the checklist above is executable in a week of focused work. The whistle does not care either way.

Building something this could apply to?

We take on a small number of flagship projects each quarter.

Start a project