[ultramancool]

Well said. I recently migrated to FreeBSD after trying sytemd on Arch and seeing that Debian and Ubuntu are planning to move too.

The dead simple rc.conf file seems so much nicer than the stuff I was dealing with in the entire world of Linux-based systems, like going back to the way Arch used to be when I really liked it.

[allegory]

This. The FreeBSD rc system just works, is well documented and is small enough to understand by one person without too much effort.

[_delirium]

As an init system it works fine, but you do end up having to find or invent a bunch of additional stuff if you want similar functionality to what's driving some of the systemd use-cases. The result might still be better (I haven't done a detailed architectural comparison), but you do need something. For example one of the things I find useful about the "systemd way" of things is that it provides, finally, a story about how to apply cgroups to services in a sane way. The kernel provides the APIs, but actually using them from userspace was not fun previously, with multiple incompatible systems, largely based on tangles of shell scripts that had broken corner cases.

With FreeBSD, my impression is that manual shell scripting is still the norm. Integrating RCTL (FreeBSD's resource-limiting facility) with service management basically consists of manually writing in a bunch of imperative calls to RCTL into scripts. There's no way to configure services with limits declaratively, ensure the right thing happens when services are started/stopped, etc., precisely because there's no integration between the RCTL facility and the process-management or init facilities. Or at least I haven't found a way. The closest is that if you need such integration only for jails, you do have the option of third-party "monolithic" management systems, such as CBSD.

[allegory]

RCTL is a stateful database. It's not there yet, but the right solution for managing this declaratively as with anything else on a Unix platform is Ansible/salt/cfengine or something like that, not building those tools into a superset service that manages everything.

I will also add that managing disparate platforms is never a reality from experience. There are perhaps two core platforms at a company and they are migrated together in blocks, all together. For us, we have a couple of legacy Ubuntu machines that are being canned this month. Everything else is Windows 2012 R2 and FreeBSD 10.

The "systemd way" is to provide a monolithic abstraction over many things with a DBus API. It's the equivalent of adding WMI and a registry to a Unix platform i.e. it's against the fundamental tenets of the operating system. Having managed windows systems for years, this is really not something I want to see. Time will tell, but if I'm not right about that then I'll eat all three of my hats.

And yes I have experience with systemd as well through evaluation of RHEL7. Within two hours, I'd hit a wall with timedatectl enabling NTP on the machine. The steps to debug the mess were horrible and the issue eventually just spontaneously disappeared.

That's reminiscent of the stateful nature of windows which brings back many years of pain in the 1990's and 00's for me.

[johnny22]

don't you just have to turn off chrony to fix that?

[RexRollman]

I've been tinkering with NetBSD the last couple of weeks in a VMWare Fusion virtual machine and the RC system it uses it very nice. OpenBSD's is nice as well.

[dobbsbob]

NetBSD can run as a xen dom0 host too though normally I use Alpine Linux/OpenRC as dom0 because it's so small.

[imanaccount247]

>NetBSD can run as a xen dom0 host too

Isn't there a bunch of fine print that goes along with that?

>Alpine Linux

How? The installer doesn't even work.

[aerique]

Are there reasons why Linux couldn't just adopt it?

[vidarh]

Feel free. Most people won't, as systemd solves very real problems that people care a great deal about, whether or not you like the way it has solved them.

[bigbugbag]

15 years linux user here, systemd is pushing me hard towards leaving linux. Please tell me what very real problems people care a great dead bout systemd solved by turning log files from text to binary.

Also I care about being able to use my computer and for the first in 15 years a systemd update caused my computer to needlessly dropping into systemd emergency mode at boot and this emergency mode being broken I was effectively locked out of my computer because an optional external usb drive that was defined in fstab with no issue for a couple years now required a nofail option. Now consider that this computer is located in a remote location 1000 km away from where I live.

To me systemd already caused way more very real problems I do care a great deal about than it has solved reducing boot time by a few seconds is not something I care that much about.

[allegory]

For me Linux is pretty dead already because I can't entirely trust the direct it's going in having survived the Unix wars of the 1990s. There are so many parallels to that at the moment, it's not funny. There are large vendors pulling it in separate directions (Canonical, Redhat, Google). At the end of the day, much like back then, customers will suffer from terrible support, fragmentation and political battles.

I just want to get shit done and solve problems and anything that risks that gets outed now.

FreeBSD hits the sweet spot, probably followed by NetBSD.

[cwyers]

"There are large vendors pulling it in separate directions (Canonical, Redhat, Google)."

It's pretty clear how that's going to shake out, isn't it? Google is pretty much a non-issue here; yes, Android and ChromeOS use a Linux kernel base, but they have no impact on any mainline distros, and there's no indication Google wants them to. So it reduces down to two parties fighting for control: Canonical and Red Hat. And Red Hat is going to win. Canonical doesn't have the resources to go its own way on more than a handful of fronts (this is why when Debian switched to systemd Upstart was killed off; Canonical is far too reliant on Debian as an upstream to fight every issue), and their requirement for a CLA to accept anyone else's code means they are entirely reliant on their own coders, as nobody wants to sign Canonical's CLAs. We'll see how long they can stick it out on Mir, but they don't have the resources to fight a war with Red Hat on two fronts, so that's the only issue I expect to see them fighting over.

[vidarh]

I've not claimed that Systemd gets everything right. I've claimed it gets enough right enough that a lot of people will be entirely unwilling to give up those advantages and return to something that for many of us is now an inferior solution, just because there are things about Systemd we may not agree with.

For my part, I agree that binary logs was not necessary, though I've yet to encounter any issues with it, and journald certainly does provide a lot of functionality that makes it more pleasant to deal with logs than before. All of that could have been achieved while retaining text logs, though. But at the same time, it is still trivial to log to text files by telling journald to log to syslog if that matters to you.

Other things I do care about include getting rid of init scripts - that is a persistent source of problems. I'm inclined to believe not a single one of them are bug free, though that's probably a bit uncharitable. Unit files helps. So does cgroup containment to rid us of the abomination that is the need to rely on pid-files and hope that works reliably (it doesn't, since pretty much nobody are through enough when writing init scripts). Other things include better recoverability in cases where critical processes gets killed, and well thought out handling of early stage logging. And things like systemd-cgtop and systemd-cgls are nice.

I'm sure we'll eventually get solutions that split more of this functionality out into more cleanly separate components, and that'll be great, but until then I'm happy to stay with systemd.

As for the problems you ran into, that sucks, but any large change like this will have painful teething problems and they're not a good basis for judging whether it's a good long term solution - I've had plenty of boot failures caused by problems with init scripts as well.

Boot time is a long way down the list of benefits for me too - most of our servers have uptimes measured in years, and even my home laptop usually goes a month or two between reboots.

[jbb555]

I couldn't agree more. If it become impossible to use linux without systemd then I won't be using linux any more.

[Andys]

There's nothing stopping a Linux distro doing this.... but

It would be a step backwards: it is simpler, and does less stuff, so booting would be slower and some features are missing.

[allegory]

How often do you reboot your kit? Boot time is such a stupid metric even on laptops and stuff where you just suspend/hibernate.

My BSD systems (not front-facing and therefore on a lesser patch cycle) rarely get rebooted and neither do the processes so this is indeed moot for me.

Proof:

http://i.imgur.com/tZsM82Q.png

Yes that's a memcached uptime on a host that has had 10,185,367,932 cache hits...

[_delirium]

Suspend/hibernate on free unixes is a nightmare of incomplete support and buggy drivers, so not much of a solution. I'm don't know a single person who has a working laptop suspend/resume setup on FreeBSD (though it's theoretically possible), and it's not usually recommended to rely on it even if you could get it working. Linux has somewhat more complete support, but it's still very hit or miss, and it's common for stuff to be wonky after a resume, even when it does work.

[allegory]

I'll give you that to a degree. It does suck on FreeBSD with my X201. Nothing works but I'm being cheeky now and running it in VirtualBox on top of windows (which I need for other work).

OpenBSD however works wonderfully.

[edwintorok]

The main question is would it work? For me systemd or applications starting to support systemd only break things (something wrong with policykit/consolekit for example with sysvinit+systemd-shim) that used to work.

Also there are some peculiarities in the way the LSB init script compatibility is implemented in systemd: it tries to be 'smart' and remember their state. So you start an init script, and it failed for some reason / perhaps even exited with an error code, perhaps you are still developing that init script. Now fixing the problem and running the init script / systemctl start doesn't even try to run the script because it thinks it has run it yet. You first have to tell it to stop it (which fails), and only then you can run it again.

[rjsw]

Why would booting be slower ?

My BSD systems boot quickly enough for me.

[chousuke]

My FreeBSD system has a 30-second timeout during which the entire boot process is halted because it waits for a default route to the internet... which it won't get, because I haven't configured one.

It's pretty dumb, and not enough of a problem for me that I'd figure out how to work around it, but it's a pretty good example.

[allegory]

Just set in rc.conf:

   defaultrouter="NO"

It shouldn't hang.

[ash]

I don't think this 30-second timeout is a bug in FreeBSD or in rc. You may want your server to wait for the network to become available. Ubuntu Server has the same "waiting for network" timeout:

http://askubuntu.com/questions/63456/waiting-for-network-con...

[jamesog]

Slackware uses exactly this system.

[earlz]

The really sad thing is that Arch Linux use to pride itself on being BSD-like. It used a similar rc.conf system. Each service had exactly one init script, not the weird multiple file init script system Debian has. Before systemd, Arch was the closest you could get to BSD simplicity on Linux

[imanaccount247]

Slackware and gentoo have both been closer for a very long time.