How do I share autocannon results?

Run autocannon with `--json`, render the result to a small HTML table or chart, and publish it to Comma. You get a URL with the latency percentiles laid out properly, instead of an ASCII table that loses its column alignment the moment it is pasted into a chat window.

Why does the autocannon table break in Slack?

The table is aligned with spaces and drawn with box characters. Slack renders it in a proportional font unless it is inside a code fence, and even inside one it wraps on narrow windows and on mobile — so the percentile columns stop lining up with their headers.

Which number should the report lead with?

p97.5 or p99 latency, not the mean. A mean hides the tail, and the tail is what users experience as 'the app is slow'. autocannon's JSON gives you every percentile, so publish the distribution rather than one summary number.

Can I run it in CI and keep a history?

Yes. PATCH the same report id from each run. One URL per endpoint, one revision per run, and a throughput regression becomes a visible diff instead of a number nobody wrote down last time.

Can I compare two runs?

Publish both to one report as consecutive revisions. The revision list keeps the pair together, so the before-and-after does not depend on somebody finding the old terminal scrollback.

Share an autocannon report

autocannon's terminal summary is well designed for a terminal and hostile to every other destination. It is a space-aligned table with box-drawing borders, so pasting it into Slack, a pull request, or a ticket produces one of two results: a proportional-font mess, or a code fence that wraps on any window narrower than the table and puts p99 under the "Avg" header.

The numbers survive; the ability to read them does not.

Get machine-readable output

npx autocannon -c 100 -d 30 -j http://localhost:3000/api/orders > result.json

-j gives you the full result — every latency percentile, the throughput histogram, the non-2xx count — instead of the rendered table.

Render and publish

A small script turns that into something a person can read. The point is not the chart library, it is that percentiles get a proper table and the tail is visible:

npx autocannon -c 100 -d 30 -j http://localhost:3000/api/orders > result.json

node -e '
const r = require("./result.json");
const rows = Object.entries(r.latency)
  .filter(([k]) => k.startsWith("p") || k === "average")
  .map(([k, v]) => `<tr><td>${k}</td><td>${v} ms</td></tr>`).join("");
require("fs").writeFileSync("bench.html",
  `<h1>${r.url}</h1><p>${r.connections} connections · ${r.duration}s ·
   ${r.requests.average} req/s · ${r.non2xx} non-2xx</p>
   <table><tr><th>Percentile</th><th>Latency</th></tr>${rows}</table>`);
'

curl -fsS -X POST "https://commareports.com/api/v1/reports" \
  -H "Authorization: Bearer $COMMA_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d "$(jq -n --rawfile html bench.html \
        --arg t "Load — GET /api/orders" '{title: $t, html: $html}')"

Or skip the script entirely and drag the HTML into the app.

Lead with the tail

The mean is the number that gets quoted and the number that means least. A run averaging 40 ms with a p99 of 1.8 s is a service that is fast for dashboards and broken for users. Publish the distribution, and put non2xx next to it — a throughput figure achieved while returning 503s is not a throughput figure.

From CI

npx autocannon -c 100 -d 30 -j "$TARGET" > result.json
# …render bench.html as above…

curl -fsS -X PATCH "https://commareports.com/api/v1/reports/$REPORT_ID" \
  -H "Authorization: Bearer $COMMA_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d "$(jq -n --rawfile html bench.html \
        --arg t "Load — /api/orders @ $GITHUB_SHA" '{title: $t, html: $html}')"

PATCHing one report id per endpoint gives you a stable URL and a revision per run — which is the cheapest performance-regression history you can have, and the one most teams skip because the results were only ever in a job log that expired.

What review adds

Anchored threads on the row somebody disputes — see commenting on HTML.
Revisions, so "we regressed on the connection-pool change" is checkable.
Access per report — private, team, domain-gated or named reviewers. See the sharing model.

Limits

Entry HTML: 5 MB. A rendered benchmark summary is kilobytes; this is not the limit you will meet.
60 requests/minute per token.

Try it

Comma is free — unlimited reports, unlimited commenters, unlimited revision history.

Publish a benchmark →