$ k6 featuresNAME STATUS DESCRIPTIONnative-histograms experimental Use native histograms for trend metrics
$ k6 run --features native-histograms script.jsk6 v2.1.0, released June 30,
adds a feature-flag system for experimental behavior in the core engine. You opt
in with --features, the K6_FEATURES environment variable, or a features key
in config.json, and you list what exists with k6 features. The release ships
one flag, native-histograms, which switches trend metrics to native Prometheus
histograms.
One switch, and a command that lists them
An experimental feature used to reach you as a specific K6_ environment variable
named in release notes, with no command to tell you what was available or how
finished each one was. 2.1 gives every experimental feature the same switch and a
single place to read them.
k6 features prints the catalog: each flag’s name, its lifecycle status, and a
one-line description of what it does. k6 features --json gives you the same list
in a shape a script can read, so a CI job can assert which flags a pipeline runs
with. To turn one on, pass --features native-histograms to k6 run, set
K6_FEATURES=native-histograms in the environment, or add the flag to the
features array in config.json. The three routes cover the CLI, the container,
and the checked-in config without a custom build.
The lifecycle status is the useful part. A flag marked experimental warns you
the behavior can change before it becomes the default, so you read it as a preview
and not a stable API. When the k6 team promotes a feature, the status moves and the
flag stays where you already know to look.
native-histograms, the first flag
k6 trend metrics track a latency distribution: the p95 and p99 you gate a load test on. Export those to Prometheus and the bucket layout decides how accurate the tail reads. A classic Prometheus histogram needs its buckets declared up front, so a p99 that lands past your highest bucket comes back as an estimate against the nearest boundary you happened to pick.
Native histograms drop the guess. Prometheus stores high-resolution buckets whose boundaries grow with the observed values, so a request that runs slower than anything you planned for still lands in a real bucket. Turn on the flag and k6 exports trend metrics as native histograms through the remote-write output, which feeds a native-histogram dashboard that reads accurate percentiles straight from the stored distribution. For a team that gates a release on p99, the flag removes a source of quiet error in the number the gate checks.
Also in 2.1
The browser module gains a context-level proxy option, so you route a context’s
traffic through a proxy from the test instead of building a custom binary:
const context = await browser.newContext({ proxy: { server: 'http://proxy.test:8080', bypass: 'localhost,127.0.0.1', },});locator.isInViewport() lands alongside it, with an optional ratio between 0 and
1 to assert how much of an element sits on screen. k6 cloud test list prints the
tests in a Grafana Cloud k6 project and takes --project-id and --json. The
OpenTelemetry HTTP exporter accepts basic auth through
K6_OTEL_HTTP_EXPORTER_USERNAME and K6_OTEL_HTTP_EXPORTER_PASSWORD, and
Selection.single(selector) gives a faster path for a single-element HTML lookup.
The bug fixes cover the rough edges. --vus works on its own as an execution
shortcut, an invalid threshold percentile fails with a message that names the
problem instead of a silent parse, and k6 writes the config file with owner-only
0o600 permissions so a shared runner does not leave credentials readable.
The part that outlives the release
Every experimental feature k6 ships next gets one switch and one line in
k6 features.
native-histograms is the first flag, and the flag system is the change that keeps
paying out. The next preview k6 wants to try arrives as a named entry with a
lifecycle status you can query, not a new environment variable buried in a
changelog. Run k6 features on your current binary and you see what your team can
turn on today.
