Hi there folks, I’m still learning about Linux and have yet to dip my toes properly in any arch based distro. Have for the moment fallen in love with the immutable distros based on Universal Blue project. However I do want to learn about what arch has to offer to and plan on installing default arch when I have time. But have been wondering why I haven’t heard of any immutable distros from arch based distros yet.
So, am left wondering if there are talks within that Arch community of building immutable distros?
While writing this post I found a project called Arkane Linux, which seem to be very interesting. Does anyone have nay experience with it? Is there a specific reason why immutable wouldn’t be a good idea when based on Arch?
Project: https://arkanelinux.org/
Thank you for your reply!
I guess I’d define it as a distro where the base system is read-only and changes or updates to it are done by replacing it atomically.
Aight. I got no qualms with that definition for an immutable distro. However, small nitpick, the term “base system” can be very murky at times. And perhaps I would rephrase the part addressing changes/updates to “changes or updates to it are intended to be applied atomically”.
Btw, I think this conversation is primarily on semantics and some assumptions we’re making related to that. So, I agree with you that (strictly speaking) immutability is only part of the puzzle (perhaps I might even refer to it as an enabler) for acquiring a lot of the aforementioned benefits to the degree by which it’s attained. So, the precise implementation of immutability is at least as important.
For example, openSUSE Aeon/Kalpa, as much as I like them, have not been able to deliver most of these benefits beyond what traditional distros are capable of. Despite these distros being immutable*. However, they’ve recognized their faults and intend to move towards an image-based solution in order to improve. Similarly, Vanilla OS has recognized that their first vision of ABRoot wasn’t fit and thus overhauled it to be more in line with Fedora Atomic. We should continue to regard their initial visions as immutable distros despite ‘their failings’, but should also recognize that their failures aren’t representative of what immutable distros are or can be.
How exactly?
what can they do that you can’t on a normal distro?
Alright, let’s start:
(Note that the immutable distros will only be represented by Fedora Atomic, GuixSD and NixOS. The others are either too niche or immature)
- atomic updates; no need to explain as you’ve established this yourself.
- reproducibility; I’m not talking here simply about “reproducible builds”. Instead, I’m referring to how, on a daily basis, the system is actually built up from the ground (or at least capable of doing so without the hassle of actually committing to a complete reinstall). Some references in case you’re interested to know more about this. This is something I’ve simply never seen on a traditional distro. Though I hope you can correct me on this.
- declarative system configuration; atomicity is just a pre-cursor or enabler for this. Nonetheless, ultimately, it’s a feature found exclusively on immutable distros. In case you mention Ansible, you should be aware that Ansible (at best) offers convergent system management. The aforementioned immutables succeeds effortlessly in (the superior) congruent system management. Refer to this article in case you’re interested in the differences.
- increased security; Linux’ security standards leaves a lot to be desired. Unfortunately, whatever immutability offers in that regard, is far from enough to bridge the gap. By contrast, Qubes OS, while not being immutable (and technically not even a Linux distro), succeeds at setting an excellent security standard. But even they acknowledge the benefits that immutability provides for security (even on Qubes OS).
- built-in rollback functionality; I can’t remember how GuixSD and NixOS far surpass anything else out there in this regard. However, for Fedora Atomic’s OCI model, it’s required that some registry builds your images. Currently, these images are stored for 90 days by default. Which means that I can go back to any one of these without keeping the snapshots on my own device. While it’s technically possible to keep snapshots from the last 90 days with Snapper/Timeshift, most sane people would simply not do it as it’s overkill and/or wasteful of storage. However, Fedora Atomic allows such functionality without burdening your system. Hence, it’s superior. Though, once again, perhaps this functionality is somehow available on traditional distros. But I’m simply unaware of it*.
There’s no need to go over the “consequences” as they’re (as the name implies) consequences of what has mentioned earlier. Hence, as their causes are better than the one found on traditional distros, so are the consequences better than how they’re found on traditional distros.
Finally, minimizing bit rot, configuration drift and hidden/unknown states are direct consequences of atomicity and declarative system management. Hence, immutable distros perform better at this compared to traditional distros.
(Note that the immutable distros will only be represented by Fedora, GuixSD and NixOS. The others are either too niche or immature)
This was my issue with your original comment - I’m aware most of the work on features like these is based on immutable distros, but just being immutable doesn’t mean it will have those features.
When it comes to reproducibility and declarative system management, I think you’re right that they’re only available in immutable distros.
The security benefit of a read-only filesystem isn’t very significant IMO, and for some immutable distros, interesting parts (to attackers, like /etc for example) are mutable anyway.
And I don’t use any snapshot solution currently, but don’t most of them only store the parts that change between snapshots? According to the Arch Wiki, Snapper’s “default settings will keep 10 hourly, 10 daily, 10 monthly and 10 yearly snapshots”. This doesn’t seem like much of an advantage for immutable distros, really.
There’s no need to go over the “consequences” as they’re (as the name implies) consequences of what has mentioned earlier. Hence, as their causes are better than the one found on traditional distros, so are the consequences better than how they’re found on traditional distros.
I disagree with this though. “Better” is very subjective - I for one consider being able to have an up to date system that can have parts of it updated without rebooting to be much nicer than using something like rpm-ostree, even if it is safer to use in theory (I can’t remember the last time I had an issue when installing a package; rebooting to apply an install atomically will likely make no difference to me other than wasting my time). I know I can use containers to get around this, but once again, this just adds to the hassle.
Thanks for the quick reply!
This was my issue with your original comment - I’m aware most of the work on features like these is based on immutable distros, but just being immutable doesn’t mean it will have those features.
As alluded by the following in my previous comment:
Btw, I think this conversation is primarily on semantics and some assumptions we’re making related to that. So, I agree with you that (strictly speaking) immutability is only part of the puzzle (perhaps I might even refer to it as an enabler) for acquiring a lot of the aforementioned benefits to the degree by which it’s attained. So, the precise implementation of immutability is at least as important.
So, to conclude this point; yes, an immutable distro is not required to come with all those features by strict virtue of its immutability.
Arguably, our talk might have resolved a lot earlier if in your original comment;
That’s because most of these benefits are not a result of a distro being immutable.
, you had replaced “immutable” by “atomic”. To be clear, the “immutable” in “immutable distro” is not the correct adjective if we want to be descriptive. That’s probably why you chose to give the (current) definition of “immutable distro” rather than “being immutable” when prompted. Hence, the name “immutable distro” is continuously being redefined and rehashed based on the distros that are represented by them. The popular definition for “immutable distro” right after SteamOS 3.0 was released, was very different from the definition you gave it earlier. Which was again very different when we had only NixOS and Guix System as our points of references. Just like how I mentioned to not have any qualms with your earlier definition, I likely wouldn’t have any qualms with earlier shifts of the definition. Therefore, I’d argue, the notion of "immutable distro" is perhaps best defined by the distros that it represents. And currently, within the discourse, Fedora Atomic is its flag bearer. Hence, why a lot of other comments found under this post make assumptions based on that as their point of reference. But, I see Fedora Atomic merely as an iteration of NixOS but image-based (Colin Walters has even reported to be inspired by NixOS). And, the other (notable) immutable distros are heading that way. (And some, like blendOS, might already have come very close to that vision already.)
The security benefit of a read-only filesystem isn’t very significant IMO, and for some immutable distros, interesting parts (to attackers, like /etc for example) are mutable anyway.
It may not be very significant, but it is significant enough that even Qubes OS (with their excellent model) aspires to it. Btw, I never implied or said that security became perfect (quite the opposite actually) just by virtue of becoming immutable. Instead, I only said it improved*. Finally, I suppose it’s worth mentioning that e.g. Fedora Atomic does track the changes to /etc, keeps a pristine copy of /etc and allows you to flush /etc.
And I don’t use any snapshot solution currently, but don’t most of them only store the parts that change between snapshots? According to the Arch Wiki, Snapper’s “default settings will keep 10 hourly, 10 daily, 10 monthly and 10 yearly snapshots”. This doesn’t seem like much of an advantage for immutable distros, really.
No space occupied on your machine is better than some space occupied on your machine. I only said it’s better, its significance is definitely up to debate though.
There’s no need to go over the “consequences” as they’re (as the name implies) consequences of what has mentioned earlier. Hence, as their causes are better than the one found on traditional distros, so are the consequences better than how they’re found on traditional distros.
I disagree with this though. “Better” is very subjective
To be clear, I didn’t intend to imply that literally all consequences are better. With “consequences”, I actually implied the points that were mentioned in the comment you first replied to; rock solid system even with relatively up to date packages, possibility to enable automatic updates in background without fearing breakage, (quasi) factory reset feature, setting up a new system in just a fraction of the time required otherwise.
I for one consider being able to have an up to date system that can have parts of it updated without rebooting to be much nicer than using something like rpm-ostree, even if it is safer to use in theory
Let’s not disregard NixOS and Guix System 😅. Furthermore, I understand the frustration. Thankfully, even in Fedora Atomic, there’s a plethora of alternative package managers you can use to suit your needs; AppImage, Flatpak, Guix, Homebrew, Nix etc. Besides, I don’t think you install new software every single day. FWIW, systemd does offer the soft-reboot functionality; though, the biggest problem for me personally is restarting all the programs that were open. So yeah… Though, this might be an issue of the past with the upcoming systemd-sysext.
(I can’t remember the last time I had an issue when installing a package; rebooting to apply an install atomically will likely make no difference to me other than wasting my time). I know I can use containers to get around this, but once again, this just adds to the hassle.
This criticism is absolutely fair. I know you felt compelled to said this only due to a misinterpretation of what I meant with “consequences”. Nevertheless, I am totally with you that the ‘Fedora Atomic’-model is not perfect. And, perhaps, never will be. For all we know, it will coexist with traditional distributions perpetually.
