This interactive demo requires JavaScript. Please enable it to run the demo in your browser.

Applications & services · Go

loadgun

HTTP load testing — configure RPS, duration and concurrency; the simulator reports nearest-rank percentiles on a live histogram.

GitHub

Configure the run

URL

Limit mode

Requests (-n)

Concurrency (-c)

Rate cap RPS (--rate, 0 = none)

Base latency ms

Server error %

Simulates a target: latency ≈ base + queueing under concurrency + noise. Percentiles use the same nearest-rank rule as the Go code — a p95 is a latency that actually happened. Requests that error before a response (timeouts) are left out of the latency distribution; a 5xx still has a latency and stays in, matching internal/report.

Results

Requests

—

Succeeded

—

Failed

—

Throughput

—

Received

—

min

—

mean

—

p50

—

p90

—

p95

—

p99

—

max

—

Latency distribution

bucket count p95 marker 10 equal-width buckets, min→max (same as Histogram())

$ loadgun -n 500 -c 25 http://localhost:8080/health

Four decisions that make the numbers meaningful

Percentiles are nearest-rank, not interpolated. p95 is a latency a request actually experienced.
A request that errors has no latency. Timeouts stay out of the distribution; a 503 is still measured.
Duration stops new requests; it does not cut in-flight ones.
The body is drained before the clock stops. Keep-alive measures the server, not the handshake.

Status codes

Code	Count
Run a test

Exit code is 1 if anything failed — use it as a one-line CI gate.