What a Festival On-Sale Teaches You About Traffic Spikes

A festival on-sale is the purest load test in existence. The time is published weeks ahead. The audience is motivated, because supply is genuinely limited. Everyone arrives in the same few seconds, and they all want to perform the same expensive, transactional, non-cacheable operation at once.

Ordinary web traffic does not behave like this. Normal load arrives as a curve that a sensible autoscaling policy can follow comfortably. An on-sale arrives as a step function, going from near-idle to peak faster than most scaling systems can add capacity, because provisioning a new instance takes longer than the spike takes to arrive.

This is why serious ticketing platforms use queue systems rather than trying to absorb the spike directly. A queue converts an unmanageable instantaneous load into a controlled stream, admitting people at a rate the transactional systems can genuinely sustain. It is a worse experience than instant access and a far better one than a site that falls over, because the failure mode is waiting rather than losing your basket at the payment step.

The design principle underneath is worth carrying into far more ordinary projects. When demand exceeds capacity, you get to choose how the system degrades. You can choose deliberately, by queueing, shedding non-essential load, serving cached content and protecting the transactional path. Or you can decline to choose, in which case the system will pick for you, and it will pick the worst option available.

Most businesses will never run an on-sale. Plenty will run a product launch, a campaign that lands better than expected, a piece of coverage that arrives without warning, or a seasonal peak that is entirely predictable and still catches them out. The question is not whether you can serve peak load. It is whether you have decided, in advance, what happens when you cannot.