: Making excess too easy

As I was chatting with a former colleague the other day, I realized that I have been doing some pretty diverse work as part of my home lab. A quick scan of my posts in this category reveal a host of topics ranging from home automation to Python monitoring to Kubernetes administration. One of my colleague’s questions was something to the effect of “How do you have time to do all of this?”

As I thought about it for a minute, I realized that all of my Kubernetes research would not have been possible if I had not first taken the opportunity to automate the process of provisioning Hyper-V VMs. In my Kubernetes experimentation, I have easily provisioned 35-40 Ubuntu VMs, and then promptly broken two-thirds of them through experimentation. Thinking about taking the time to install Ubuntu and provision it before I can start work, well, that would have been a non-starter.

It started with a build…

In my corporate career, we have been moving more towards Azure DevOps and away from TeamCity. To date, I am impressed with Azure DevOps. Pipelines-as-code appeals to my inner geek, and not having to maintain a server and build agents has its perks. I had visions of migrating from TeamCity to Azure DevOps, hoping I could take advantage of Microsoft’s generosity with small teams. Alas, Azure DevOps is free for small teams ONLY if you self host your build agents, which meant a small amount of machine maintenance.. I wanted to be able to self-host agents with the same software that Microsoft uses for their Github Actions/Azure DevOps agents. After reading through the Github Virtual Environments repository, I determined it was time to learn Packer.

The build agents for Github/Azure Devops are provisioned using Packer. My initial hope was that I would just be able to clone that repository, run packer, and viola! It’s never that easy. The Packer projects in that repository are designed to provision VM images that run in Azure, not on Hyper-V. Provisioning Hyper-V machines is possible through Packer, but requires different template files and some tweaking of the provisioning scripts. Without getting too much into the weeds, Packer uses different builders for Azure and Hyper-V. So I had to grab all the provisioning scripts I wanted from the template files in the Virtual Environments repository, but configure a builder for Hyper-V. Thankfully, Nick Charlton provided a great starting point for automating Ubuntu 20.04 installs with Packer. From there, I was off to the races.

Enabling my excess

Through probably 40 hours of trial and error, I got to the point where I was building my own build agents and hooking them up to my Azure DevOps account. It should be noted that fully provisioning a build agent takes six to eight hours, so most of that 40 hours was “fire and forget.” With that success, I started to think: “Could I provision simpler Ubuntu servers and use those to experiment with Kubernetes?”

The answer, in short, is “Of course!” I went about creating some Powershell scripts and Packer templates so that I could provision various levels of Ubuntu servers. I have shared those scripts, along with my build agent provisioning scripts, in my provisioning-projects repository on Github. With those scripts, I was off to the races, provisioning new machines at will. It is remarkable the risks you will take in a lab environment, knowing that you are only 20-30 minutes away from a clean machine should you mess something up.

A note on IP management

If you dig into the repository above, you may notice some scripts and code around provisioning a MAC address from a “Unifi IP Manager.” I created a small API wrapper that utilizes the Unifi Controller APIs to create clients with fixed IP addresses. The API generates a random, but valid, MAC Address for Hyper-V, then uses the Unifi Controller API to assign a fixed IP.

That project isn’t quite ready for public consumption, but if you are interested, drop a comment on this post.