A load test without a traffic model is a lottery. We build scenarios from analytics: login, checkout, webhooks, night reports. Then a longer run, because a connection leak shows after hours, not after two minutes.
Scenario and observation
We bind the result to what you see in production: API response time, database locks, CPU saturation, queues. A node-failure test is a separate session. We do not mix it with the first performance measurement.
A harness before the campaign
For software houses we give a repeatable set that can be run on every client before Black Friday or a fintech campaign. Scope: load testing with k6 and JMeter.
In short: you test the cart and payment on numbers from analytics before the campaign starts. A homepage under load says nothing about checkout.