[SahAssar]

> The general perception is that no one wants it (systemd)

I disagree. A few loud voices strongly do not want it, but a few others like it, and the vast majority seems pretty content using it, not caring as long as it works well enough.

[alex3305]

I'm only using systemd as a user, but for me the overall experience is rather painless after some trial and error. Except the learning curve for more advanced usages is steep and documentation seems way too complete and complicated. But I tend to experience that more with (larger) open source projects anyway, so maybe that's on me.

[m-p-3]

Same here. I can't say I'm comfortable with it, but for the amount of services I manually need to create (that aren't pre-made by package maintainers) I normally find enough info online to figure it out by myself.

[hdjjhhvvhga]

"No one wants it" is obviously a hyperbole. "It caused significant problems," "several design decisions were unfortunate," "a significant percentage of sysadmins were upset by it," and so on - all these are true statements, whereas "no one wants it" is not.

[User23]

Popular with people who don’t care is pretty faint praise. But it’s probably enough to run a business so there it is.

[SahAssar]

It's also popular with quite a few people who actually really like it and use it extensively, like me. We just tend not to be as vocal about it, probably because we don't need to be since that side is the default in all mainstream distros.

My main complaint about systemd is that most people don't use it enough. the units for sockets, timers, paths can be really powerful, and the sandboxing capabilities are great. They are just underused.

For basic system components like an init, network manager, etc. I'd also say that "Popular with people who don’t care" is great praise. It means that the people who don't want to care can choose to just not. Being great with people who do care is of course great too (and being popular with both is the best), but don't discount the value of a utility just being a utility.

[User23]

It wasn’t meant as a negative. As you clearly imply, people have built a trillion dollar company on it. You don’t have to care about the implementation details is basically Apple’s value proposition. E.g. I just want my airpods to work with all my stuff is a great user story.

And that’s not meant as a positive.

[cassianoleal]

> the sandboxing capabilities are great

I find the sandboxing to be very subpar. It works for the most basic use-cases but for anything with slightly more complicated requirements you're better off scripting your way through cgroups and namespaces.

[SahAssar]

Out of interest, got examples of those use-cases where it does not work?

[cassianoleal]

I like systemd's service units and its subsystem, and related units like sockets. I dislike pretty much everything else about it.

[SahAssar]

What "else" are you referring to? networkd? resolved? journald? Something else?

[cassianoleal]

All of those, yes.

Networkd is a pain to configure and operate, and interferes with the lower level subsystems. Resolved is a mess that only causes headaches. It might be ok-ish on desktops but honestly I never had an issue with regular /etc/resolv.conf on any of my laptops and desktops. Journald is a (terrible) solution looking for a problem and makes everything more opaque and harder to troubleshoot.

Containerisation and namespacing is a big mess and lacking on basic features. It also doesn't play nice with the core tooling (ip netns, etc).

I like the service units but honestly we could all have done without it. There were already other perfectly usable solutions in that space. I used runit for a while before nearly every distro out there caved to systemd.

[SahAssar]

Alright, gotcha. They have all worked well for me, but I understand that they might not for you. The good thing then is that all of those are replaceable with their traditional counterparts, so you can pick and choose while still keeping the core systemd that you seem to find acceptable.