Preserving an ancient but reliable website via Docker and AI

2026-09-01 [Petri]

After creating this website around Christmas in 2017, I have been spending both my paid and free time on learning a multitude of new things like cloud management, production-grade 3D-printing, hardware design and subcontracting, drone battery technologies and charging systems, and lately also cloud and local AI.

So no wonder that my “hands-on” experience of both Django and Bootstrap have totally faded in the background.

There’s only so much you can keep in the active memory these days, and the tool chains are so complicated that you simply can’t remain fully fluent on something you are not actively using.

What I did back in 2017 was just an exercise to learn the (then) latest web trends, and to get a web site that I knew and owned fully, instead of ending up as just another customer of some content creation setup that would have its own quirks and bugs that would be out of my control.

And throughout all these years, the site has just chugged along amazingly reliably. All the praise I wrote in my very first blog entry about Django and Bootstrap still stands.

I have had constant heart beat monitoring of the web server health, and I recall only one crash being reported. The server had been up and running for almost two years non-stop at that point, so most likely the failure had nothing to do with my Django setup at all.

And as usual, the classic IT mantra for transient problem solving of “have you tried turning it off and on again” worked.

The nice side effect of such a ”DIY” approach is the fact that the multitude of hacking exploits based on more widely used web site management systems are totally non-issues:

For the hackers, a system hand-written in Django is basically a black box, and the management interface is accessible through my personally managed VPN environment only. So there is no indication that any of the logged tens of thousands of hacking attempts against the web server ever got in over the almost nine years of being online.

Seriously, the amount of obscure daily (failing) web pings that point at flaws in WordPress and other common tools is quite staggering.

Because my setup worked so well, I had also dutifully followed the second IT mantra of “if it works, don’t fix it”.

I did not see any pressing reason to update the tool set to the latest versions, and thus I eventually ended up in a situation where the support for various components had either ended or was ending soon.

At this point, the biggest backdoor for possible intrusion was the outdated, no longer security-maintained Linux server, and I really had to start seriously considering on a system-wide upgrade.

After charting out my alternatives, the easiest solution became obvious:

As my setup is still in a perfectly working condition, why not put it all in a Docker container that matched the currently active components of Django, Bootstrap and Python? This container can then be run as a process on the newest Linux with the spiffiest OS-level hacker protections.

So, in short: the site itself would be running on a dedicated, non-root user ID inside a Docker instance, frozen in terms of functionality, which still is totally sufficient for my needs.

The worst that could happen is someone trashing the contents of the Django media database, which has been copied daily to my backups since day one, so restoring data would be an easy task.

I did have Docker experience by running my mail server setup, but doing it all from scratch against Docker container template configuration that matches my ancient toolkit set was still something that was not a standard "Docker cook book" event.

So I piggybacked the help from ChatGPT and Claude for the transition, and after spoon-feeding the AI tools one step at a time during what turned out to be a three-day session, I had a working and tested Dockerized carbon copy of my trusty website.

Therefore, I give my sincere thanks to OpenAI and Anthropic. Although I have to admit that I am personally quite certain that the ongoing AI boom will crash under its own weight due to the fact that you can already get your own, "good enough" local AI server setup for just a few thousand euros or dollars.

And the elephant in the room for these companies is that we all know how the price of computing power has only one direction. Down.

Maybe they all will find a way out from the multi-billion dollar circular investment hole that they, in my humble opinion, have dug themselves in, but they may just as well have all fallen into the first-mover trap in this.

With an enormous debt burden.

One interesting detail here is that whenever I ran out of the generous free tokens, I started a couple of intermediary upgrade steps locally via Unsloth Studio on an AI model running on a separate, old laptop in my home network.

And although that was purely CPU-bound and thus very slow, I usually got 1-3 additional steps done locally until I was allowed to burn more free tokens.

Yes, it was about small, auxiliary stuff, but still free, local AI.

In a way, the free tier setups of these companies have the traditional "first shot is free" lure, but increasingly the next step is not about the customer getting hooked, but starting to run her "own stash" instead.

I assume that I could have done all this without ever running out of tokens and within a reasonable (5-10x slower) time frame on a local AI node costing maybe €5000 today.

This, in a corporate context, would be peanuts to add to the cost of otherwise proficient coder, and a small price to pay for an environment that certainly won’t leak anything to anywhere outside of the company.

Add to this the newest wave or running mega-models containing hundreds of billions of parameters in limited memory via predictive expert caching/loading algorithms, like FreeToken, and the depth of what can be done locally may suddenly expand, putting the hundreds of billions spent on running today's models in data centers under the threat of becoming obsolete.

So, if you just read this story, you were accessing the new and improved carbon copy of my 2017 work, with just two small changes against the original: new font family (so I could distinguish between old and new sites during testing) and better handling of text and image sections on small screens (see the attached image).

To summarize: based on my experience, using Docker to “crystallize” an old, fully working setup, together with all standard container precautions to stop any escalation after a potential breach due to the underlying ancient tool kits, certainly appeared to be a low-effort, manageable solution.

Let’s see if that takes me through the next nine years.

Permalink: https://bhoew.com/blog/en/170

Earlier entries