My New Rube Goldberg Machine

My last post was about my local COVID tracker tool. While it worked well, I found having to re-run the script every time I wanted an update annoying. Plus, I wanted to share this on Facebook, but I have non-technical friends who would not be able to run it for themselves.

So I decided to put up a hosted version, but I challenged myself. I wanted it to run entirely on someone else’s machines. And I didn’t want to pay for it.

So how to do it?

Well, the hosting is simple. I’ve been using Render for this blog, as well as my professional site and some other static sites1. And while my COVID tracker does require updated data to stay relevant, the data is just a simple JSON file and the chart is generated entirely in the browser.

So the trick was to make the data file available in a way that let me deploy it with Render every time there are updates.

Enter my Rube Goldberg machine.

The data source I’m using,, only updates their data daily, so I only need to run this once a day to stay relevant. This sounds like a job for cron. But not on my desktop machine (even though that would be way simpler).

Instead, I used GitHub Actions. It supports scheduled jobs as well as running on every push to the repo. But the trick is to then make the data available after each run. And then the trickier trick is to get that data as part of running the deploy job on Render. Oh, and every time the GitHub Action runs, I want to have the Render site deploy again.

This turned out to be not that hard.

My GitHub Actions workflow runs the script, which generates a summary.json file. Then the workflow uploads that file as a build artifact. This is all incredibly trivial, and by using caching for both my Perl prereqs and the intermediate data files, I can make it quite fast. When the cache is warm, a run takes less than a minute. When it finishes, it hits a webhook provided by Render to trigger a deploy.

Of course, GitHub has an API for artifacts like the summary.json file. So all I need to do in the Render deploy script is use the API to find the latest artifact, then download that and deploy it along with my index.html and chart.js files. With a little experimentation, I was able to create a Bash script to do exactly that. I could have written this in Perl, but the combination of curl, jq, and zcat (artifact files are always zipped) actually made this much simpler to do in Bash than Perl2. I had to use sed, which always seems weird when I know Perl, but doing this in Perl requires at least a few more characters.

The hardest part was figuring out how to securely store the Render webhook URL in GitHub and then access it in my workflow. I had to store a GitHub token in Render3 as well.

And so I present to you

Also, you might note that the chart has changed a little since last time. I made the past 7-day average line thicker and the daily numbers line thinner. The average is much more indicative of trends then the actual daily numbers, which jump around quite a bit.

  1. See my previous writeup on moving all my sites to Render. ↩︎

  2. This happens every once in a while. ↩︎

  3. I guess I didn’t have to, but the GitHub limit on unauthenticated requests is so low that I figured it was best to use authentication instead. ↩︎