Grafana k6 Documentation

What it is

Grafana k6 is an open source performance testing tool maintained by Grafana Labs. You write tests as JavaScript files using the k6 API; k6 runs them with many virtual users and reports latency percentiles, throughput, error rates and custom metrics. The core tool is licensed under AGPL-3.0 and runs locally or in CI. Grafana Cloud k6 is a separate hosted option for larger, distributed runs.

What the documentation covers

• Installing k6 and writing your first test.
• Options for virtual users, duration, stages and scenarios.
• Checks, which assert on responses without stopping the test, and thresholds, which turn a run into a pass or fail, such as keeping p95 latency under 500 ms.
• Test types: smoke, load, stress, spike, soak and breakpoint testing, with guidance on when to use each.
• Metrics, results output and integrations, including CI.
• Browser-level performance testing with the k6 browser module.

Is it worth your time?

For a tester who needs performance testing at awareness level, yes. The scripting is plain JavaScript, a first meaningful test is only a few lines, and thresholds produce the exit code a CI pipeline needs. The guides on test types are also a clear introduction to performance testing concepts in general, whichever tool you end up using.

Prerequisites

Basic JavaScript and HTTP. An application you own or have permission to load test, such as a local copy of your practice app. Never load test sites or APIs you do not own.

Free or paid?

k6 itself is free and open source. Grafana Cloud k6 has free and paid tiers; check the official pricing page. You do not need the cloud product to learn.

What you can do after

Write and run a load test with sensible virtual users and duration, read the summary, set thresholds for latency and errors, and run a small performance check in CI.

Common questions

k6 or JMeter? JMeter is older, Java-based and GUI-driven, and many companies still use it. k6 is script-first and fits code review and CI well. Learn one properly and recognise the other.

Can I trust numbers from my laptop? Only roughly. Local machines and home connections limit how much load you can generate, so treat early results as learning, not a capacity report.