Performance testing best practices with k6

1. Preparing #

A test without an objective proves nothing.

Before running #

• Define the objective (SLO, target throughput, peak date) and the type of test.
• Model real traffic: journey mix, think times, varied data.
• Test a representative (size, data, configuration) and isolated environment.
• Warn the teams and providers (CDN, third-party APIs, WAF).

2. Running and observing #

Measuring on both sides.

During the test #

• Start with a smoke test, then ramp up gradually.
• Monitor the test machine (CPU, network) and dropped_iterations.
• Export k6 metrics with a testid and display them next to those of the system under test.
• Note events (scaling, restarts) during the test.

3. Automating #

Performance as a test like any other.

Continuously #

• Thresholds aligned with SLOs, blocking in CI.
• Smoke on every pull request, load every night.
• Scripts versioned and reviewed with the code.
• Result history to detect slow regressions.

Common mistake: Never compare two tests run on different environments: environment variation often exceeds the effect of the change being tested.

Reference diagram and quick check #

Checklist.

Checklist #

1. Quantified objective and test type chosen
2. Realistic scenarios and varied data
3. Thresholds aligned with SLOs
4. k6 and system-under-test metrics on the same dashboard
5. Test automated in CI
6. Report with conclusions and actions