Adding Privacy Friendly tracking to my Blog
10 Dec 2024I’ve recently caved into the idea of adding some tracking onto my blog, but rather than cave completely to the likes of Google Analytics, I’ve opted to investigate some minimal, privacy friendly tracking solutions.
In the past I’ve been happy blogging to the void and not knowing how popular my posts are or where they’re being linked from, but recently, with my Advent of Code posts, I’ve been interested in how many people if any have actually seen them. Additionally, the basic analytics you get from platforms like X and LinkedIn aren’t always sufficient.
So my first port of call was to figure out what was out there. Recently, we’ve been evaluating ChatGPT Enterprise at my work, so I thought I’d challenge it to find me a privacy respecting analytics solution. At first it mentioned Plausible and Fathom as options, but when I drilled it for open-source, self hosted options it listed several more:
Of the second batch, GoAccess and Umami looked like the simplest to get started with, along with Plausible from the first two suggestions, that gave me 3 options to investigate.
ChatGPT did provide some limited information on each option, but I still don’t fully trust its results and in general wanted to do further research on each option. So I fired up the web browser and opened up each options home page, as well as did a quick search for each.
Resource Usage
One of the reasons I had been reluctant to install analytics, other than apathy, was that any service would need to consume some resources, and the 3 options are no different. Each needs some amount of memory and CPU time, but they also need somewhere to store their data. Umami and Plausible both need some kind of SQL database, whereas GoAccess stores its data in memory and persists locally to disk.
Recommended resources seemed to be around 2GB of memory for Plausible CE. Umami didn’t list exact system requirements, but it does need both a NodeJS server and MySQL/PostgreSQL server running. GoAccess seemed to be the most lightweight in terms of resources, according to it’s documentation its memory footprint is dependent on the log size it parses.
I use DigitalOcean for my hosting. Specifically I use their App Platform for my blog, and one of their droplets for hosting some subdomains. This complicates things. My deployment is done automatically upon pushing to my Blog repository which is a very nice feature, the SSL is all automated, and the cost is non-existent from the blog side of things (the App is free in DigitalOcean).
With relative ease, I can deploy whatever solution I come up with to a subdomain on a droplet, in fact I already do similar for some other projects. This will inevitably incur some kind resource cost but I’d prefer to keep this to a minimum.
GoAccess
So, of the 3 I highlighted, GoAccess has the least amount of dependencies and resource usage.
In my exploration of articles related to GoAccess, I found this blog post by Ben Hoyt from 2019, about replacing his Google Analytics with GoAccess. I also found this thread from Hacker News linking to a blog post by Deni Bačić.
Ben Hoyt’s post was quite extensive at explaining the process he took. He also had a similar situation with how his blog was hosted compared to mine, except he uses GitHub Pages, something I migrated away from. He opted to host a special tracking pixel on Amazon S3 via CloudFront. This would produce logs he’d process with GoAccess.
Hoyt’s approach had the benefit that some analytics would still be collected if you block JavaScript, as the tracking pixel (sometimes called a web beacon) will still be loaded, barring any blocking software.
Hoyt also mentioned at the top of his article that he later replaced GoAccess with something called GoatCounter.
If I wanted to apply a similar process to Ben Hoyt, I could host a tracking pixel at a subdomain and process the logs it produces. Specifically, I could do this in a lightweight way with an existing DigitalOcean droplet and Nginx instance. This would be a little simpler than Hoyt’s approach as I would not need to convert logs and could run GoAccess on the same location.
GoatCounter
Before I jumped down the rabbit hole of implementing a GoAccess solution, I paused to look into GoatCounter. There was obviously a reason why Ben Hoyt opted to migrate to GoatCounter within 2 years of using GoAccess. Hoyt actually wrote another article for LWN.net about GoatCounter and Plausible. In it he highlights the benefits of these two open-source analytics solutions, one such benefit of GoatCounter is that it is free for personal use. In theory this means I would not need to host anything at all
Of course, one issue using any third-party analytics, even non-invasive ones,
is that they are easily blocked by ad-blockers and hosts lists.
GoatCounter is no
different. However, as that GitHub issue highlights, you can host the
GoatCounter count.js script elsewhere, and proxy your requests through your
own domain to GoatCounter if desired, or self host it entirely.
After looking into GoatCounter, it seems even more lightweight than GoAccess. This means that self-hosting it won’t break the bank or take up inordinate resources.
Conclusion
With all that investigation done, I opted to use GoatCounter. For now I am not self-hosting it, instead I will be using the simple support provided while I evaluate it. This does mean it can be blocked by ad-blocking scripts, so I may get only limited analytical information from it, but it can still be a useful tool.
I will likely revisit the idea of analytics in the new year, perhaps I will migrate to GoAccess, self-host GoatCounter, or even implement something incredibly basic myself, based on a simple tracking pixel and/or custom script. Until then I’d love to know what privacy conscious solutions others have used for tracking their personal blogs.
Comments