Comment by nine_k
13 hours ago
The backlash against systemd was twofold. On one hand, when released and thrust upon distros via Gnome, it was quite rough around the edges, which caused both real problems and just understandable irritation. Fifteen years after, the kinks are ironed out, but it was sort of a long time. (Btrfs, released at about the same time, took even longer to stop being imprudent to use in production.)
On the other hand, systemd replaces Unix (sort of like Hurd, but differently). It grabs system init, logging, authentication, DNS, session management, cron, daemon monitoring, socket activation, running containers, etc. In an ideal Red Hat world, I suppose, a bare-metal box should contain a kernel, systemd, podman, IP tools, and maybe sshd and busybox. This is a very anti-Unix, mainframe-like approach, but for a big consulting firm, like Red Hat / IBM, it is very attractive.
All this lather about doing it the UNIX way, whilst neglecting to point out that the old tooling was far worse. "Do one thing well" implies it was done well to begin with.
DNS: Can you from memory recite how name lookups work on Linux? Ever had t otrack down problems with non-standard setups? `resolvectl` is not perfect, but it does let you control all of this stuff in one place, and with a nice, orderly view of what does what.
Init system: ever written the old sysV ones from scratch? Sure they're just shell script, but did you remember to make yours re-entrant? What about forking or master-slave processes? Hope you got your pid checking code just right...
Containers: Docker is quite robust nowadays but it's not like it follows the 'UNIX philosophy' either. And systemd/nspawn at least lets you do namespacing and cgroups reasonably well and in a straightforward way...
Mounts, etc.: let's not get into fstab and its idiosyncrasies.
Logging: let's hope you set up logrotate properly or you're in for a surprise.
And on it goes.
Systemd is not perfect. But what it replaces was god-awful and far worse.
> DNS: Can you from memory recite how name lookups work on Linux
Yes, I can, and I use systemd only because it's the default on debian, I have no reason to try devuan yet.
> ever written the old sysV ones from scratch?
Many, many times, and I was only an enthusiast/user, not a sysadmin.
> did you remember to make yours re-entrant?
Dealing with PID files was only mildly annoying. Init scripts were very boilerplate-y, so I wouldn't forget anything after my copy-paste-edit-delete unnecessary parts cycle. In a single afternoon, one could bash out an CLI init-script generator that uses jinja2 templates and interactively asking <10 questions about the service.
> Systemd is not perfect. But what it replaces was god-awful and far worse.
Init systems shouldn't have anything to do with managing container lifecycles beyond managing the container-runner service using the usual unix interface (signals). Call me a purist, but system services shouldn't be containerized.
An init system shouldn't be managing DNS or logging either, those should be standalone components. If they are problematic, there should be composable, domain-specific tools that solve them, instead of smooshing everything into systemd.
SystemD wasn't the only possible way to solve those logging, DNS, or security policy problems, and I'm glad other PID 1 projects that focus on being init systems are thriving.
So one could agree that something should be done, in each of these cases. But that doesn't imply that the thing to be done should have been systemd, or even systemd-shaped. But no, it has borged the lot.
No, systemd absolutely does not replace Unix.
Systemd-the-project and systemd-the-service-manager (“init”) are two different things. The former is a project with numerous components (e.g. resolved) that actually _are_ rather modular; they usually require systemd-the-service-manager, but you (or your distro) can generally pick and choose the components you want.
The service manager does indeed require some components to be gobbled up (udev comes to mind). But subsuming other subsystems shouldn’t be so anathema; the systemd people didn’t just think that “the one” thing of the Unix philosophy wasn’t being done well. Rather, the idea is that is was the wrong thing, i.e. classic Unix init was a tool operating at the wrong layer of abstraction. And in their eyes, a modern system needs a richer set of userspace primitives. So they made engineering decisions in pursuit of that goal.
Is following the "Unix way" a terminal value? I.e., is it desirable for itself, or is it just supposed to be a means to an end?
In discussions such as these, the Unix philosophy of "do one thing and do it well" is often being touted as a proxy for (and a necessary attribute of) "good design", as if all possible wisdom about the future of computing was available to the creators of UNIX in 1969.
You are repeating a bunch of "talking points" common among systemd-critics, but are not really backed up.
First of all, it wasn't "thrust upon" anyone, it was democratically selected multiple times in a ranked voting setup in case of Debian, and independently by Arch as well. It was simply because maintainers were fed up with the absolutely unmaintainable mess that predated systemd -- it seems random-ass bash scripts are not suitable for such a complex problem as booting up a system, and doing it properly is much better.
Logging sucked great time before, e.g. you didn't even get logs from before the Linux kernel is started - systemd moves it to a single place. And if you are for some reason irritated by binary logging, you can just freely pipe it to text logs.
Authentication is not done by systemd, are you thinking of pam modules? The network service is not systemd, just runs under the same group's name - KDE file browser is also different from their terminal. Also, it's not mandatory to use. Logind is not systemd itself, again. Scheduling services makes absolute sense for systemd's problem domain, so do monitoring and socket activation.
You need some kind of order to build stuff on, the Unix philosophy is more of a feel good convention than a real design guideline (and it doesn't apply in many cases)