There is a similar question on the site which must not be named.
My question still has a little different spin:
It seems to me that one of the biggest selling points of Nix is basically infrastructure as code. (Of course being immutable etc. is nice by itself.)
I wonder now, how big the delta is for people like me: All my desktops/servers are based on Debian stable with heavy customization, but 100% automated via Ansible. It seems to me, that a lot of the vocal Nix user (fans) switched from a pet desktop and discover IaC via Nix, and that they are in the end raving about IaC (which Nix might or might not be a good vehicle for).
When I gave Silverblue a try, I totally loved it, but then to configure it for my needs, I basically would have needed to configure the host system, some containers and overlays to replicate my Debian setup, so for me it seemed like too much effort to arrive nearly at where I started. (And of course I can use distrobox/podman and have containerized environments on Debian w/o trouble.)
Am I missing something?
Everyone here have already explained their various stances very eloquently and convincingly - so I won’t argue against that - so instead I’ll just put forth my own 2c on why I use Silverblue instead of Nix/Ansible.
The main draw for me in using Silverblue (well, uBlue to be exact) is the no-cost, cloud-based, industry-standard, CI/CD and OCI workflow. Working with these standard technologies also helps me polish up my skills for work, as we’ve started to make use of containers and gitops workflows, so the skills that I’m gaining at personal level are easily translatable for work (and vice-versa).
With Nix (the declarative way), I’d have to learn the Nix language first and maintain the non-standard Nix config files and, tbh, I don’t want to waste so much time on something that no one in the industry actually uses. Declarative Nix won’t really help me grow professionally, and whilst I agree it has some very unique advantages and use-cases, it’s completely overkill for my personal needs. In saying that, I’m happy with using Nix the imperative way though - I don’t need to learn the Nix language, and it’s great having access to a vast package repository and access my programs without having to go thru the limitations of containers.
As for Ansible, I’d have to have my own server (and pay for it, if it’s in the cloud), and spend time maintaining it too. And although we use Ansible at work as well, so the skills I gain here won’t be waste of time, it’s unfortunately too inflexible/rigid for my personal needs - my personal systems are constantly evolving, whether it is in the common packages I use, or my choice of DE (my most recent fling is with Wayfire) etc. With an Ansible workflow, I’d be constantly editing yaml files instead of actually making the change I want to see. It’s overkill for me, and a waste of time (IMO). You could argue that I’m already editing my configs on Github with uBlue, but it’s nowhere as onerous as having to write playbooks for every single thing. And as I mentioned, I like to maintain some flexibility and manual control over my personal machines and Ansible will just get in the way of that.
With the uBlue workflow, I just maintain my own fork on Github with most of my customisations, + a separate repository for specific dotfiles and scripts that I don’t want to be part of my image. Pull bot keeps my main uBlue repo in sync with upstream, and I only need to jump in if there’s some merge conflicts that cannot be resolved automatically. At the end of it all, I get a working OCI image, with a global CDN and 90 days of image archives, allowing for flexible rollback options - all of this without incurring any costs or wasting too much time on my part. Plus I can easily switch between different DEs and OCI distros, with just a simple rebase - I could go from a Steam-Deck like gaming experience (Bazzite) to a productivity-oriented workstation (Bluefin), or play around with some fancy new opinionated environments like Hyprland and River (Wayblue) - all with just a simple rebase and a reboot, without needing to learn some niche language or waste time writing config files. How cool is that?
In this comparison, the devil is in the detail.
With Ansible, you have an initial condition onto which you add additional state through automatically executed steps dictated by you until you (hopefully) arrive at a target state. This all happens through modification of one set of state; each step receives the state of the previous step, modifies it and passes the entire state onto the next step. The end result is not only dependant on your declared steps but also the initial state. A failure in any step means you’re left in an inconsistent state which is especially critical for the case of updating an existing state which is the most common thing to do to a Linux system.
In NixOS, you describe the desired target state and the NixOS modules then turn that description into compartmentalised bits of independent state. These are then cheaply and generically combined into a “bundle”; wrapping them into one big “generation” that contains your entire target state.
Your running system state is not modified at any point in this process. It is fully independent, no matter what the desired system is supposed to be. It is so independent in fact that you could do this “realisation” of the NixOS system on any other system of the same platform that has Nix installed without any information about the state of the system it’s intended to be deployed on.
This “bundle” then contains a generic script which applies the pre-generated state to your actual system in a step that is as close to atomic as possible.
A good example for this are packages in your PATH. Rather than sequentially placing the binaries into the/usr/bin/directory as a package manager would when instructed by ansible to install a set of packages, NixOS merely replaces thebinsymlink with one that points at an entirely new pre-generated directory which contains the desired packages’ binaries (well, symlinks to them for efficiency). There cannot possibly be an in-between state where only some of the binaries exist; it’s all or nothing. (This concept applies to all parts that make up a Linux system of course, not just binaries in the PATH. I just chose that as an easy to understand example.)
By this property, your root filesystem no longer contains any operating system configuration state. You could wipe it and NixOS would not care. In fact, many NixOS users do that on every boot or even use atmpfsfor/.(Immutability is a property that NixOS gains almost by accident; that’s not its primary goal.)
Nice, thank you very much for this great summary!
In my own words, you describe the difference between declarative vs. imperative configuration and the joy of atomic updates. :-)
I just want to add one point: Theoretically I totally agree, that one might have a bad state in Ansible and the updated state is spoiled, and of course configuration drift is a theoretical problem via Ansible. In practice I never run into this problems in 10+ years of using Ansible. (Of course I treat servers/desktop as cattle, so every major revision of Debian means a complete/clean new installation.)
If it works for you, that’s great.
I find it a lot more convenient to treat every change about my system as a nearly complete/clean new installation though and NixOS allows you to do so. An upgrade between major versions becomes a walk in the park due to this; it’s the M.O. The last one didn’t require any modification of the config on my NAS. I remotely rebuilt the system from my MacBook Pro with the new channel, rebooted to get the newer kernel and everything worked just like before.
An upgrade between major versions becomes a walk in the park due to this
For personal desktop use, I see nixos and guixos as spending an hour every day, so that you can save minutes every few years.
When trying out guixos, I spent like 5 days going through the docs and annoying people on irc, just to accomplish the same thing as copying a file to a non-~ directory in a normal distro. Anyways, I powered through the weeks of configuration, and got to actually doing some work in the damn thing. I started off by using the language specific PM to install the dependencies I needed for a project. Welp, guix is not FHS compliant, and can only stand the presence of isolated PMs like nix and flatpak, so that’s out of the question. Since the guix packages were too old, it was time to either:
- maintain my own guix definitions of random packages + their dependencies, and then maintain a separate source of dependencies for the project
- learn how to nix and then maintain a separate source of dependencies for the project
- start figuring out how to use guix to emulate FHS inside of a venv
So I obviously went for the third option, gave up after banging my head some more, and finally nuked the partition.
Recently I started using home-manager, and tried converting my i3 config so it’s handled by it. Halfway through, it set in just how ugly the syntax looks. I mean just look at a small part from the beginning of both files: nix formatted by nixfmt, i3 config. But ok, I might get used to it, let’s test it out: it fucking added defaults like the keybindings I’ve specifically removed from the config so that the only two i3 deals with are opening the terminal and reloading itself. Then I finally realized I’m wasting my time creating this monstrosity, when literally the only benefit is that it would allow me to skip symlinking directories after a fresh install. So now it’s only managing my packages, doing a great job freshening up stable Debian, and I can reload any config immediately after saving the file instead of going through the entire process of recreating everything home-manager handles.
I might still use it for some specific packages, like firefox (about:config and custom css), which can’t really be stowed easily.
Nix and Guix os are brilliant in specific use cases they were made for, like when you’re creating a large number of ephemeral instances for scientific computing and need extremely precise control of everything that goes into the image. Spending a week to configure something simple is worth it in that scenario because it’ll save you from much bigger headaches down the line. The issue is that they solve problems most users simply don’t face, and so the added complexity comes with practically no benefits.
I see nixos and guixos as spending an hour every day, so that you can save minutes every few years.
I see it more like spending an hour every week so that you can save hours to days of annoying and stressful time every few months.
Though there are other benefits rather than just time.
When trying out guixos, I spent like 5 days going through the docs and annoying people on irc, just to accomplish the same thing as copying a file to a non-~ directory in a normal distro
In the beginning when you’re not familiar with things yet, it always takes more time to do something.
I didn’t have that particular use-case yet and I’d have to consult at least one manual to do it correctly but I’d nowadays be able to solve that particular problem with one line of relatively simple code in NixOS. (On Guix, I don’t know how I’d do it though since they don’t use systemd.)
When I started out though? No chance, it’d have taken weeks.I started off by using the language specific PM to install the dependencies I needed for a project.
Yeah, that’s like mistake #1 for non-FHS-compliant distros. ;)
- start figuring out how to use guix to emulate FHS inside of a venv
As the person who implemented a variant of this for Nix (buildFHSEnv), it’s rather straight-forward. Though if it didn’t exist, I’d rather try distrobox or other dev container thingies if there was no reasonable nix support for the thing I’m working on.
I mean just look at a small part from the beginning of both files: nix formatted by nixfmt
I mean, that’s just one particular interface. It’s actually quite flexible to do it this way though as it allows you to dumb it down if you don’t like it with a little refactor:
{ startup = let exec = command: { inherit command; }; in [ #on launch (exec "emacsclient -a '' -nc --init-directory=~/.config/emacs" // { workspace = "1"; }) (exec "nitrogen --restore") (exec "~/.screenlayout/shareni.sh") (exec "numlockx on") (exec "volumeicon") ]; }That’s the beauty of IaC (actual code, not that YAML nonsense): Software environment configuration becomes a software engineering problem and we know how to solve those.
In this specific case though, I probably wouldn’t bother with doing that stuff in Nix and would rather just keep the plain i3config text file and set the option glue to just use that file; effectively a glorified stow.
This more complex interface is only truly beneficial if there are parts of your config that vary depending on some other conditions. Some users may have the need to only run a set of commands or have certain launch options on one machine but not another. Trivial to do with
lib.optionalsand the like using this kind of interface but very hard to do if it was just a list of strings or one large string.it fucking added defaults like the keybindings I’ve specifically removed from the config so that the only two i3 deals with are opening the terminal and reloading itself
Well, then tell it to not to do that? I don’t know the module in question but any well-designed module has an option for precisely that. If it doesn’t, I’d consider that a bug. Otherwise,
lib.mkForceis usually also an option.Spending a week to configure something simple is worth it in that scenario because it’ll save you from much bigger headaches down the line. The issue is that they solve problems most users simply don’t face, and so the added complexity comes with practically no benefits.
I wholeheartedly disagree. Declarative stateless system configuration a la Nix solves a lot of issues that users face all the time.
Whether the time investment is worth it at present is debatable but there’s a clear path towards yes IMO because a project focused around proper IaC elevates operating systems onto another level because it abstracts and centralises configuration. It takes one person to figure out how to configure a certain thing in a sensible way and they can publish that work as a NixOS module for everyone’s benefit. Most of the work I put into NixOS is upstream because of this.
Right now, it’s absolutely catered towards nerds and other technologically able people like us but imagine what a further abstracted GUI could do for mere mortals.
Thanks for the detailed response.
I’d nowadays be able to solve that particular problem with one line of relatively simple code in NixOS
I can’t remember the exact details of the whole issue, but that part was for a desktop entry. If I remember correctly, in the end I had to create a system service and there were no readily available examples like for the packages. After days of researching in my spare time, I had to ask in the irc for a snippet from someone’s config.
As the person who implemented a variant of this for Nix (buildFHSEnv), it’s rather straight-forward.
Oh, it seems really cool. I’ll need to look into it.
I mean, that’s just one particular interface. It’s actually quite flexible to do it this way though as it allows you to dumb it down if you don’t like it with a little refactor:
That’s definitely an improvement, but the default config is still far better IMO.
I probably wouldn’t bother with doing that stuff in Nix and would rather just keep the plain i3config text file and set the option glue to just use that file; effectively a glorified stow.
Replacing stow with home-manager has the same issues as replacing a regular distro with nixos. If I can stow all of my dotfiles, why would I use home-manager to handle them instead? In most cases it’s just going to be harder to configure anything, and you also need to rebuild your home every time you want to update a config.
This more complex interface is only truly beneficial if there are parts of your config that vary depending on some other conditions. Some users may have the need to only run a set of commands or have certain launch options on one machine but not another. Trivial to do with lib.optionals and the like using this kind of interface but very hard to do if it was just a list of strings or one large string.
What benefits does it have over just using a shell script?
I guess it’s also great for programs that aren’t following the standards like firefox.
Well, then tell it to not to do that? I don’t know the module in question but any well-designed module has an option for precisely that. If it doesn’t, I’d consider that a bug.
It’s probably a skill issue, but that ties into another problem I’ve had when messing around with home-manager: the only source of options I found was mynixos. So to configure anything I had to first guess potential keywords to search for the option I’m interested in. And that’s after learning about it from some video on youtube, because google left me high and dry.
I wholeheartedly disagree. Declarative stateless system configuration a la Nix solves a lot of issues that users face all the time.
Can you give me some examples, what issues will I face running MX + nix that I wouldn’t if I ran nixos?
As someone who works with terraform, I understand the benefits of being forced to keep a single source of truth instead of remembering to update my post-installation script and keep things synced across devices. But on the other hand this is everything I need to do to get a fresh install to where I’m currently at:
- install the single docker dependency that doesn’t work with home-manager
- clone my dotfiles and symlink them
- set up nix, home-manager, and switch
- remove the few packages I’ve replaced with nix ones
- cp the i3.desktop (home-manager and sddm were not agreeing)
- clone and install doom
It’s definitely not a lot to maintain, and the issues are either obvious and easy to solve, or just a small waste of space. For example if I forget to remove the debian version of git, it’s still going to automatically source the nix one first. Home-manager with just a list of packages makes the hardest part of that process a breeze, while still being really easy to set up.
The main problem was getting started from 0, so I’m considering writing a post about it when I get a bit more comfortable. Trying to learn nix declarative package management from the nix manual is a bad idea, and almost all of the resources are on nixos. A quickstart guide with a few commands and examples would’ve had me up and running in 10 minutes instead of days.
Whether the time investment is worth it at present is debatable but there’s a clear path towards yes IMO because a project focused around proper IaC elevates operating systems onto another level because it abstracts and centralises configuration. It takes one person to figure out how to configure a certain thing in a sensible way and they can publish that work as a NixOS module for everyone’s benefit. Most of the work I put into NixOS is upstream because of this.
Right now, it’s absolutely catered towards nerds and other technologically able people like us but imagine what a further abstracted GUI could do for mere mortals.
Oh for sure, a home-manager gui that let’s you customize every package from a single place while automatically updating your config would be a complete game changer. But I’m talking about the current state of things. In that regard, currently every linux user can enjoy simple declarative package management with stable and bleeding edge sources. Yet I never see it mentioned, while even beginner threads are being spammed with nixos recommendations. Imagine if casuals could open their software center or discover and install nix packages instead of flatpaks.
If I can stow all of my dotfiles, why would I use home-manager to handle them instead? In most cases it’s just going to be harder to configure anything, and you also need to rebuild your home every time you want to update a config.
Yes, yes indeed. That’s why my dotfiles are still in a git repo (don’t get the point of stow), not in home-manager.
If you do in fact need home-manager’s features for some of your dotfiles though, it can effectively act as a stow superset for the rest.
What benefits does it have over just using a shell script?
Declarative stateless configuration rather than imperative stateful configuration.
With a bash script, you’d have to meticulously craft together the i3config file using shell script syntax and remember to run that every time you change something. home-manager just does all of that for you with high-level data types and frameworks specifically made for that purpose.
that ties into another problem I’ve had when messing around with home-manager: the only source of options I found was mynixos. So to configure anything I had to first guess potential keywords to search for the option I’m interested in.
Yeah, it’s not great. https://search.nixos.org/options? is really useful for NixOS.
You have to either use your browser’s dumb search on https://nix-community.github.io/home-manager/options.xhtml or your pager’s dumb search in
man home-configuration.nix.Can you give me some examples, what issues will I face running MX + nix that I wouldn’t if I ran nixos?
All the issues which declarative immutable stateless system configuration solves such as atomic updates, configuration rollback in case you messed something up and trivial recovery. I’m sure I’m forgetting some since I’m so used to having them.
The main problem was getting started from 0, so I’m considering writing a post about it when I get a bit more comfortable. Trying to learn nix declarative package management from the nix manual is a bad idea, and almost all of the resources are on nixos. A quickstart guide with a few commands and examples would’ve had me up and running in 10 minutes instead of days.
Yeah, docs are a pain point. If you think that section is bad (I think so too), everyone will thank you for rewriting it. Feel free to shoot a PR to Nixpkgs and ping a few people from the docs team if you’re motivated.
Yet I never see it mentioned, while even beginner threads are being spammed with nixos recommendations.
I don’t get it either. NixOS is the best thing since sliced bread for a certain kind of person (experienced hacker who has felt the pain points which NixOS relieves) but I’d never recommend it to an inexperienced user in its current state.
I wonder now, how big the delta is for people like me: All my desktops/servers are based on Debian stable with heavy customization, but 100% automated via Ansible.
Close to none. Immutable solve the same problem that was solved years ago with Ansible and BTRFS/ZFS snapshots, there’s an important long-term difference however…
Immutable distros are all about making thing that were easy into complex, “locked down”, “inflexible”, bullshit to justify jobs and payed tech stacks and a soon to be released property solution. We had Ansible, containers, ZFS and BTRFS that provided all the required immutability needed already but someone decided that is is time to transform proven development techniques in the hopes of eventually selling some orchestration and/or other proprietary repository / platform like Docker / Kubernetes does. Docker isn’t totally proprietary and there’s Podman but consider the following: It doesn’t really matter if there are truly open-source and open ecosystems of containerization technologies. In the end people/companies will pick the proprietary / closed option just because “it’s easier to use” or some other specific thing that will be good on the short term and very bad on the long term.
“Oh but there are truly open-source immutable distros” … true, but again this hype is much like Docker and it will invariably and inevitably lead people down a path that will then require some proprietary solution or dependency somewhere (DockerHub) that is only required because the “new” technology itself alone doesn’t deliver as others did in the past. Those people now popularizing immutable distributions clearly haven’t had any experience with it before the current hype. Let me tell you something, immutable systems aren’t a new thing we already had it with MIPS devices (mostly routers and IOTs) and people have been moving to ARM and mutable solutions because it’s better, easier and more reliable.
The RedHat/CentOS fiasco was another great example of this ecosystems and once again all those people who got burned instead of moving to a true open-source distribution like Debian decided to pick Ubuntu - it’s just a matter of time until Canonical decides to do some move.
Nowadays, without Internet and the ecosystems people can’t even do shit anymore. Have a look at the current state of things when it comes to embedded development, in the past people were able to program AVR / PIC / Arduino boards offline and today everyone moved to ESP devices and depends on the PlatformIO + VSCode ecosystem to code and deploy to the devices. Speaking about VSCode it is also open-source until you realize that 1) the language plugins that you require can only compiled and run in official builds of VSCode and 2) Microsoft took over a lot of the popular 3rd party language plugins, repackage them with a different license… making it so if you try to create a fork of VSCode you can’t have any support for any programming language because it won’t be an official VSCode build. MS be like :).
All those things that make development very easy and lowered the bar for newcomers have the dark side of being designed to reconfigure and envelope the way development gets done so someone can profit from it. That is sad and above all set dangerous precedents and creates generations of engineers and developers that don’t have truly open tools like we did.
This is all about commoditizing development - it’s a negative feedback loop that never ends. Yes I say commoditizing development because if you look at it those techs only make it easier for the entry level developer and companies instead of hiring developers for their knowledge and ability to develop they’re just hiring “cheap monkeys” that are able to configure those technologies and cloud platforms to deliver something. At the end of the they the business of those cloud companies is transforming developer knowledge into products/services that companies can buy with a click.
I would separate NixOS from other immutable distros. NixOS is really about giving you blank slate and letting you fully configure it.
You do that configuration using a static config language that is able to be far more idempotent than Andible. It’s also able to define packages that are well contained and don’t require dynamic linking setup by manually installing other packages.
Immutable distros, on the other hand, really have no advantage to your setup and will probably feel more restrictive. The main use I see for them is for someone new or lazy that wants to get a working system up and running quickly.
My favorite example of how idempotent NixOS is has to do with the DE. If you’ve ever looked at switching from gnome to KDE, or the other way around, most distros suggest to just re-install because each DE leaves so much cruft around and it’s so hard to remove everything in a safe manner.
With NixOS, you just change one line in your config, and the DE is cleanly swapped.
Or you can add specialisations, which to be fair might require a reboot (system accounts might change during specialisations switch which will confuse the script trying to reload services for the now non-existent user) but it is how I have multiple DEs installed without their applications flooding the other ones, each with their own login manager (SDDM for plasma, gdm for gnome, greetd for sway).
Well maybe you youself are too new to recognize some of the appeals ;)
One large advantage with silverblue is, that the whole composition of the OS does not take place on the target machine. That means that all the issues that could arise will not take place on the target machine, and can be dealt with beforehand. In the simple case this could mean just enjoying vanilla silverblue without having to think about possibly borking the machine. In an advanced usecase this could mean for example building the os images in a GitLab CI/CD pipeline (with well working tooling that exists already for docker etc), then having automatic tests in the pipeline ensure that everything important works as expected. And only if the tests pass, the image will be added to the repositorie’s image registry, where the target machines will fetch it from automatically and rebase to it.
No, I fully understand it. But if you build the whole system where every package is isolated, none of the packages interfere with each other, and every package is tested across a wide array of architectures, you can just as safely put together your ideal OS setup and don’t have to deal with being locked into very simple and bare system.
The right place for immutable OSes is if you’re using it as a server for container workloads, where you will never customize the base system. Or if you never want to customize your system. Yes, you can customize the system image, but it breaks all the guarantees that the images gives you because the packages themselves are not isolated and by bumping a wrong dependency for a custom packages you can still break the whole system.
Partly yes, but just installing a package without running into conflicts does not yet guarantee a working system. You have to cater for the right configurations too, for example when you think about a corporate setting with all kinds of networking whoes (like shares, vpns and such). I think you could get this to work with Nix somehow, but you want to test these things beforehand, and if you do so using images then you have the thing to ship to machines in your hands already, there’s no need to compose the OS and configurations over and over again for every machine.
Another aspect with non-atomic OS composition on the target is that you have to deal with the transient phase from one state to the next. In this phase all kinds of things could happen, for example an update of nvidia drivers would render cuda disfunctional until the next reboot, as the userspace and kernelspace parts do not fit together anymore. With something like any of the fedora atomic variants, transient phases with basically undefined behaviour do not exist, and the time the system is not guaranteed to be in working order gets reduced to just the reboot.
Nix is cool and definetely better than any traditional package manager. But it is not an ultimate solution, to be honest so far it seems to me like it is living in a nieche of enthusiasts that are smart enough to put up with its unique declaration language. And below that niche you have ordinary linux users that may just be happy with silverblue without any modifications, and above that niche you have corporate doing their own images in CI/CD, CoreOS and all that jazz.
Your summary of the language is spot on. I still hope that more distros take inspiration from the declarative config and try to move in the direction, or nix supports a better language in the future. I think that ultimately that’s what the average linux user would want. The ability to still customize in a safe manner. Silverblue, and others, are and will remain a great option for the new or indifferent user.
On your point about the transient phase, nix actually does that by default already. It installs everything at a separate path and then flips over in one go. You can even pick the mode, either try to do a live switch as you describe, or on boot. I don’t know if I see many benefits to images there.
I am at a second place now that uses NixOS in a corporate setting, and it is much easier than maintaining the CoreOS images, or similar. I’ve had some many broken builds of CoreOS images because something goes wrong between the custom packages and the base CoreOS images, I would rather just run an Ansible script at this point. Also, you end up using the exact same test suite for NixOS images as for your other images, so the same guarantees end up being met.
