gemini://perso.pw/blog/articles/nixos-workflow.gmi

After successfully switching my small computer fleet to NixOS, I'd like to share about the journey.

That sums up to 6 computers running NixOS, half of them is running the development version, and the other is running the latest release.

All the computers above used to run OpenBSD, let me explain why I migrated. It was a very complicated choice for me, because I still like OpenBSD despite I uninstalled it.

NixOS offers more software choice than OpenBSD, this is especially true for recent software, and porting them to OpenBSD is getting difficult over time.

After spending too much time with OpenBSD, I wanted to explore a whole new world, NixOS being super different, it was a good opportunity. As a professional IT worker, it's important for me to stay up to date, Linux ecosystem evolved a lot over that past ten years. What's funny is OpenBSD and NixOS share similar issues such as not being able to use binaries found on the Internet (but for various reasons)

NixOS helps me to squeeze more from my hardware (speed, storage capacity, reliability)

systemd: I bet this one will be controversial, but since I learned how to use it, I really like it (and NixOS make it even greater for writing units)

Security is hard to measure, but it's the main argument in favor of OpenBSD, however it is possible to enable mitigations on Linux as well such as hardened memory allocator or a hardened Kernel. OpenBSD isn't practical to separate services from running all in the same system, while on Linux you can easily sandbox services. In the end, the security mechanisms are different, but I feel the result is pretty similar for my threat model of protecting against script kiddies.

I give a bonus point for Linux for its ability to account CPU/Memory/Swap/Disk/network per user, group and process. This allows spotting unusual activity. Security is about protection, but also about being aware of intrusion, OpenBSD isn't very good at it at the moment.

One issue I had migrating my mail server and the router was to find what changes were made in /etc. I was able to figure which services were enabled, but not really all the steps done a few years ago to configure them. I had to scrape all the configuration file to see if it looked like verbatim default configuration or something I changed manually.

This is where NixOS shines for maintenance and configuration, everything is declarative, so you never touch anything in /etc. At anytime, even in a few years, I'll be able to exactly tell what I need for each service, without having to dig up /etc/ and compare with default files. This is a saner approach, and also ease migration toward another system (OpenBSD? ;) ) because I'd just have to apply these changes to configuration files.

Working with NixOS can be disappointing. Most of the system is read-only, you need to learn a new language (Nix) to configure services, you have to "rebuild" your system to make a change as simple as adding an entry in /etc/hosts, not very "Unix-like".

Your biggest friend is the man page configuration.nix which contains all the possible configurations settings available in NixOS, from Kernel choice and grub parameters, to Docker containers started at boot or your desktop environment.

The workflow is pretty easy, take your configuration.nix file, apply changes to it, and run "nixos-rebuild test" (or switch if you prefer) to try the changes. Then, you may want something more elaborated like tracking your changes in a git or darcs repository, and start sharing pieces of configuration between machines.

But in the end, you just declare some configuration. I prefer to keep my configurations very easy to read, I still don't have any modules or much variable, the common pieces are just .nix files imported for the systems needing it. It's super easy to track and debug.

After a while, I found it very tedious to have to run nixos-rebuild on each machine to keep them up to date, so I started using the autoUpgrade module which basically do it for you in a scheduled task.

But then, I needed to centralize each configuration file somewhere, and have fun with ssh keys because I don't like publishing my configuration files publicly. Which isn't optimal either as if you make a change locally, you need to push the changes and connect to the remote host to pull the changes and rebuild immediately instead of waiting for the auto upgrade process.

So, I wrote bento, which allows me to manage all the configuration files in a single place, but better than that, I can build the configuration locally to ensure they will work once shipped. I quickly added a way to track the status of each remote system to be sure they picked up and applied the changes (every 10 minutes). Later, I improved the network efficiency by central management computer as a local binary cache, so other systems are now downloading packages from it locally, instead of downloading them again from the Internet.

The coolest thing ever is that I can manage offline systems such as my work laptop, I can update its configuration file in the weekend for an update or to improve the environment (it mostly shares the same configuration as my main laptop), and it will automatically pick it up when I boot it.

Moving to NixOS was a very good and pleasant experience, but I had some knowledge about it before starting. It might be confusing a lot of people, and you certainly need to get into NixOS mindset to appreciate it.

My NixOS workflow after migrating from OpenBSD

Introduction

Migration

From OpenBSD to NixOS

NixOS modules

Workflow

Bento

Conclusion