Results and outputs

Files

--out json=results.jsonl and --out csv=results.csv stream every sample in k6's JSON and CSV formats (also with --workers, whose samples all go through the manager). --summary-export summary.json writes k6's end-of-test JSON, and --no-thresholds skips threshold evaluation. k6's summaryTrendStats option (or --summary-trend-stats avg,med,p(99),count) picks the values trends show in the summary and in both JSON summaries; summaryTimeUnit (s, ms or us, --summary-time-unit) shows every time in the summary in one unit.

HTML report

--report report.html (or k9.run(..., report="report.html")) writes the run as one self-contained HTML page that works offline: the key numbers, thresholds, checks, per-second charts (VUs, requests and failures per second, response-time avg/p(95)/p(99), iterations per second) and every metric. It has light and dark themes, and it's small enough (about 30 KB) to keep as a CI artifact.

Prometheus remote write

--out prometheus-rw[=URL] (or k6's name, experimental-prometheus-rw) pushes metrics to Prometheus remote write with k6's names and units: - counters are k6_<name>_total, gauges k6_<name>, and rates k6_<name>_rate; - each trend stat gets its own series, like k6_http_req_duration_p99, with times in seconds.

That way k6's Grafana dashboards work unchanged. It takes k6's K6_PROMETHEUS_RW_* settings (SERVER_URL, PUSH_INTERVAL, TREND_STATS, USERNAME/PASSWORD, BEARER_TOKEN, HEADERS_<name>, HTTP_HEADERS, LABELS, INSECURE_SKIP_TLS_VERIFY, TREND_AS_NATIVE_HISTOGRAM, STALE_MARKERS), also spelled K9_PROMETHEUS_RW_*. With TREND_AS_NATIVE_HISTOGRAM, trends are sent as native histograms (k6_http_req_duration_seconds) in place of their stats; with STALE_MARKERS, a last push marks every series stale when the test ends. Not yet supported: client certificates and SigV4. Pushes go out on a thread of their own; a failing endpoint is reported but doesn't fail the run.

InfluxDB

--out influxdb[=URL] sends every sample to InfluxDB v1 like k6's output does: the URL names the database (http://localhost:8086/k6 by default, created first), each point is named after its metric with a value field, and the tags in TAGS_AS_FIELDS (vu:int,iter:int,url by default) become fields. That way k6's InfluxDB dashboards work unchanged. --out xk6-influxdb[=URL] does the same for InfluxDB v2 (k6's extension), with ORGANIZATION, BUCKET and TOKEN. Both take k6's K6_INFLUXDB_* settings (ADDR, DB, USERNAME/PASSWORD, PUSH_INTERVAL, PRECISION, RETENTION, CONSISTENCY, INSECURE), also spelled K9_INFLUXDB_*.

StatsD and Datadog

--out statsd[=ADDR] sends samples over UDP as StatsD lines, like k6's output: counters as counts, gauges as gauges, trends as timings in ms, and checks as check.<name>.pass/.fail counts. It takes k6's K6_STATSD_* settings (ADDR, NAMESPACE, PUSH_INTERVAL, BUFFER_SIZE, ENABLE_TAGS, TAG_BLOCKLIST). --out datadog[=ADDR] is the same with DogStatsD tags on, for the Datadog agent (K6_DATADOG_*).

OpenTelemetry

--out opentelemetry[=URL] (or k6's experimental-opentelemetry) exports OTLP metrics over HTTP (protobuf) to a collector, the way k6's OpenTelemetry output does: - counters become monotonic sums, and gauges stay gauges; - rates become one counter with a condition attribute (zero or nonzero); - trends become histograms in ms or By.

Values are cumulative, and tags become attributes. It takes k6's K6_OTEL_* settings (HTTP_EXPORTER_ENDPOINT, HTTP_EXPORTER_URL_PATH, HTTP_EXPORTER_INSECURE, EXPORT_INTERVAL, METRIC_PREFIX, SERVICE_NAME, SERVICE_VERSION, HEADERS, HTTP_EXPORTER_USERNAME/PASSWORD, TLS_INSECURE_SKIP_VERIFY), also spelled K9_OTEL_*. Unlike k6, it only speaks http/protobuf (no gRPC yet), and service.name defaults to k9.