How to Size Load Generators in CI

If the load generator is CPU-bound, your latency numbers measure the runner, not the service, so size the generator to have headroom.

Watch the generator's CPU during a run. A standard runner handles a few thousand virtual users for HTTP; beyond that, move to a larger runner or split the load across pods with the k6 Operator.

Steps

• Run a baseline and watch generator CPU (k6 prints dropped iterations when saturated).
• If CPU pegs at 100 percent, the results are unreliable.
• Scale up the runner or scale out across pods.

Choosing a scale strategy

Gotchas

• k6 warns "Insufficient VUs" or drops iterations when the generator cannot keep up.
• Network egress limits on the runner can cap throughput before CPU does.

Related guides