Tracking ISP performance with Playwright and Prometheus

We’ve all been there: the internet feels slow, but when you call the ISP, they say “everything looks fine on our end.”

I wanted to monitor my connection speed consistently over time, not just when I felt a lag. My ISP, Unifique, provides a speed test page, but I wanted those numbers in Prometheus.

The Tool

Since the speed test is a web application, I couldn’t just curl an endpoint. I needed something that could interact with the page, click “Start”, and read the results.

Enter Playwright.

Playwright is fantastic for browser automation. I built a simple Node.js application that:

1. Launches a headless browser.
2. Navigates to the speed test page.
3. Clicks the start button.
4. Waits for the test to complete.
5. Scrapes the Download, Upload, and Ping results.

Architecture

The unifique-speedtest-exporter exposes a metrics endpoint. When scraped, it triggers the Playwright flow.

Note: Because speed tests take time (30s+), you’ll need to increase the scrape timeout in your Prometheus configuration.

Implementation Details

Here is a snippet of the core logic using Playwright:

await page.goto('https://speed.unifique.com.br');
await page.click('#start-button');
await page.waitForSelector('.result-finished');

const download = await page.$eval('.download-speed', el => el.innerText);
const upload = await page.$eval('.upload-speed', el => el.innerText);

await page.goto('https://speed.unifique.com.br');
await page.click('#start-button');
await page.waitForSelector('.result-finished');

const download = await page.$eval('.download-speed', el => el.innerText);
const upload = await page.$eval('.upload-speed', el => el.innerText);

This isn’t the exact code, but it gives you an idea of how simple the interaction is.

Running it

I packaged this as a Docker container to make it easy to deploy.

docker run -p 3000:3000 ghcr.io/mateuxlucax/unifique-speedtest-exporter:latest

docker run -p 3000:3000 ghcr.io/mateuxlucax/unifique-speedtest-exporter:latest

It exposes metrics on port 3000.

Prometheus Configuration

This is the tricky part. Since the speed test takes a while to complete, you must increase the scrape_timeout.

scrape_configs:
  - job_name: 'unifique_speedtest'
    scrape_interval: 1h # Don't run this too often!
    scrape_timeout: 2m  # Give it plenty of time to finish
    static_configs:
      - targets: ['192.168.1.100:3000']

scrape_configs:
  - job_name: 'unifique_speedtest'
    scrape_interval: 1h # Don't run this too often!
    scrape_timeout: 2m  # Give it plenty of time to finish
    static_configs:
      - targets: ['192.168.1.100:3000']

Future Plans

Currently, this is a Proof of Concept (PoC) written in Node.js. It works, but running a full headless browser is resource-intensive.

In the future, I plan to:

1. Reverse engineer the WebSocket/API calls the speed test page uses.
2. Rewrite the exporter in Go or Rust to remove the browser dependency.
3. Add more robust error handling for network flakes.

Conclusion

Now I have a graph that shows my internet speed every hour. If I ever need to call support, I have the data to back up my claims!

Check out the repository: MateuxLucax/unifique-speedtest-exporter