Recently I setup in my home lab an Ubuntu server with Docker and Portainer. Works great. Thank you and a shout out to Network Chuck with these YouTube videos that inspired me to get engaged with Docker finally. (Learning Docker is Hard!! (this makes it easy) , 18 Weird and Wonderful ways I use Docker).
For a while I have seen some challenges and some potential solutions that would fit our family needs, and would get me back to hands-on work building a family website and web applications. Yay! I have Docker now, so, why not use that to host. Brilliant.
I need a development box. I have my daily driver - Raspberry Pi4 with Ubuntu desktop installed, and VSCode. Great. I can use that for developing in Python, possibly using Django, leveraging some recent learning that I've been through. Build things and help cement new knowledge in place is a win-win. But I've never built code to deploy to Docker before.
So, what tools do I need and how do I bring my development and deployment chain together to make this happen? Diving down the rabbit hole of research I quickly find the application Docker Desktop, a developer tool specifically for building and deploying to Docker. But, there is no installation package for Docker Desktop to install on an ARM based system like the Raspberry Pi. I was keen to get the source code for Docker Desktop and build my own installation package for the Raspberry Pi. From my research, this would not work because ARM chips cannot sufficiently support the virtualisation that Docker Desktop requires. My search for alternative solutions crossed the paths of about 20 different websites and blogs, none of which offered a complete and coherent answer to my challenge.
Finding a New Path
I found myself going in circles trying to cobble together a solution, getting absolutely no traction. Every time I started to go back to look at the research results, I felt like my problem set was getting more confused not less. After a coffee (No coffee, no workee!), I decided to try a new approach that I have not used before. AI.
I have since 2023 used ChatGPT once or twice just to see what the hype was with some simple fluff-piece prompts that were of no real value. But I could see here a circumstance in which I might use it to go do the research for me, and to provided some clarity that I was not finding for myself.
Note, I was not going to trust the machine learning of ChatGPT or any other large-language model (LLM) straight out of the box. The machine learning tools are made by humans, and they are feed and trained on information created by humans. Potentially, that doubles their points of failure and therefore doubles the chance of getting an incorrect or insufficient answer, from my point of view.
I also did not wish to give too many details regards my home lab to ChatGPT. I am trying to improve my OpSec. So crafting my prompts thoughtfully and carefully was necessary. This was the first prompt that I gave to ChatGPT:
Assume:
- you are an experienced web developer,
- and I am a junior web developer.
- we have developer workstations that are ARM64 architecture with Ubuntu 24.04 for the operating system and Visual Studio Code as our IDE
- Our docker engine server already exists and does not need to be setup
- Our job is to develop websites for deployment to docker containers.
Noting these constraints, I want you, the senior developer, to write a list of instructions for me, the junior developer, describing step-by-step how to setup my workstation and build my first website into a docker file to be compiled into a docker image and deployed onto the docker engine server.
A simple, neat and succinct prompt, that tells ChatGPT the basics of what I want, and does so within a meaningful context that informs ChatGPT of how I want it to approach the task, and what output I expect.
From the first response, ChatGPT was on the money with how it laid out the response, neat step-by-step instructions that were easy to read and follow, and included the rationalisation for each step. Great!
However, I had told ChatGPT, that my development workstation was ARM64 architecture. The response it came back with did not recognise that ARM chips would not support the required virtualisation of hosting Docker containers from the developer workstation.
So, now ChatGPT and I needed to reverse roles, where I was taking it's work and providing constructive and corrective critique - somewhat like providing guidance and mentorship to a junior team member. I did this by feeding a new prompt in the same workspace.
Please revise your entire instructions noting that our ARM64 workstations do not support virtualisation and therefore will not support docker CLI installation. Please republish the entire list of instructions.
In the new response, ChatGPT acknowledged the fault and gave an improved solution.
Second solution was workable. I would still test it before giving it a passing grade just to be sure. What I didn't like was that ChatGPT was guiding me toward using Java node.js. In part my fault as my prompts had not specified that I want to use Python. I will give ChatGPT bonus marks though, it identified that a developer working through the instructions it had given, would need to go through a series of repetitive steps every time that they wished to deploy code to the Docker server to test, so in the last step, it included the following instructions:
I felt that this inclusion was helpful, but I still needed to teach ChatGPT more about what I wanted and expected from it, so that it could do a better job of teaching me the solution I wanted. Again, using the same workspace, I pushed ChatGPT to do better.
Taking the solution last provided, I would like you to modify and republish in full based upon the following feedback. Assume that our application security team has recently identified that node.js has too many known and exploitable vulnerabilities and is therefore too much of a security risk. Our corporate policy allows the use of Python3. Please consider what changes are required across all of the steps to deliver the same solution using Python, not Java/JavaScript on the server side.
The solution it came back with removed the use of node.js and was re-crafted to work with Python. It was an excellent result. I'm not going to reproduce it here, but it did give me the following steps to work through:
- Step 1: Update Your System
- Step 2: Install Required Development Tools
- Step 3: Set Up Visual Studio Code
- Step 4: Create Your First Python Website
- Step 5: Create a Requirements File
- Step 6: Create a Dockerfile
- Step 7: Transfer Files to the Remote Docker Engine Server
- Step 8: Build and Run Docker on the Remote Server
- Step 9: Automating Deployment (Optional)
The layout of the steps was nice. It allowed me to quickly cast an eye over it and get a real feel for whether or not the solution was worthy of going to the next stage of testing it out. It was. There was an inclusion of an IP address of '0.0.0.0' in the app.py file in Step 4 that I wondered if it was a placeholder, given that I had not told ChatGPT what my home lab IP ranges were. I queried ChatGPT and it told me that it was the correct literal value, and not a placeholder.
For full disclosure, when I typed that value in, despite clarifying it with ChatGPT, I made the error of using the IP address of my Docker server. Whilst my web site files were archived and deployed to the server, then built into a Docker image and instantiated in a container, the website did not work. Looking through Portainer, I could see that the container was failing to start.
Normally, I would have walked away, taken time to calm down, slowly thought it through, returned, researched, conducted troubleshooting, and eventually some days down the track, maybe I would have found my error. This time however, what I could do, was go back to ChatGPT, and tell it what error messages I was getting. What amazed me next was the way that in each iteration of troubleshooting, ChatGPT came back with a number of steps to work through to resolve the problem starting with what it felt to be the most likely cause.
Having this almost immediate (less than 5 minutes think time on each interaction) feedback and assistance, was valuable to my headspace, to getting me to think the problem through more fully, and to helping me to get back into the grove of troubleshooting my own mistakes. It kept me motivated and focused. In the end, after a number of iterations kept circling around the same potential causes, it dawned upon me that I had ignored the clarification that I had asked ChatGPT for. It had given me a highly workable solution, and a simple PEBCAK (Problem Exists Between Chair And Keyboard) issue had got in the way.
In summary:
Is AI smarter than humans? No, and it is not 'Artifical Intelligence' as such. It is Machine Learning built and trained upon LLMs. It is capable of consuming and processing information to produce outputs faster than a human. But I cannot declare that it is faster at thinking/processing than a human brain. Essentially the 'AI' can do the research and reading faster, and it can 'type' or produce the answers in a highly readable and neatly formatted form faster than a human can. Whilst the IO is faster, I'm not convinced the processing is.
Do I feel that 'AI' will replace IT workers? No, but it can augment and help them. I believe that my experience above gives evidence of how this may occur.
I feel that it in certain circumstances it is of great benefit. For example I can see benefits for team members who are working solo on a task, who need timely answers to queries and feedback, or who might be working remotely with less opportunity for immediate feedback from colleagues and mentors. Having worked with many people who are 'on the spectrum' with varying degrees of ADHD, autism, or other mentally divergent capacities, I have experienced that some have a lot of questions for their mentors, and some have difficulty expressing questions in a way that a mentor can quickly assist with. Whilst I would not advocate replacing a human mentor, I feel there may be cases where using a human mentor and 'AI' mentoring side-by-side would have benefits for both the mentee and the mentor.
