No systemd discussion?

[Anthony de Boer]

Peter wrote:
> Let's be honest (warning: I am into VERY small systems which probably have
> no room for systemd - but see below):

I've got an early-revision Beaglebone Black with Angstrom Linux running
systemd, and it runs quite nicely in that constrained environment.

(Granted, the later-rev with Debian and the additional-package options
behind that comes a lot closer to being actually _useful_.)

> The scripts are not the problem. The problem is the daemons and other things
> they start/stop, which can be programmed in ways which make them
> nonresponsive for a variety of reasons. init(1) has a sane way to cope with
> runaway processes, by making them sleep for a bit if they go insane. Zombies
> should be dealt with by process group mechanisms which exist, but are not
> used by anyone. It is really easy to write a bit of code which registers
> "children to be killed in case of process group leader demise" perhaps
> directly into /var/run/thedaemon.pid files, to be used by a slightly modded
> init(1) when a process dies or runs away.

I haven't seen very many distros using inittab to respawn anything
beyond the traditional getty lines.

Most boxes these days use SysV init scripts, which start a daemon once
and leave you without if it dies.  And the finagling they do to try and
find the process they started if you want to stop or restart it does NOT
always work.  I've seen an attempt at a clean shutdown wedged because of
a stop script sucking mud.  And other daemons have had strange issues
because of crap inherited from the shell of the sysadmin who restarted
them.

Daemontools did a good job, starting daemons in a clean environment and
restarting if needed, but it never got enough traction mostly because
of people having issues with the author.

I'm not quite out of my third decade as a Unix admin, and was really
hoping we'd be past a lot of those very traditional old problems by now.
The status quo is NOT perfect, various bits and pieces of solutions have
been tried, and at least conceptually systemd pulls all that together
and looks a lot like a 21st century init system should look like.
However, there's a lot of inertia in the status quo, a *LOT* more systems
around than when we moved from the old single rc script to SysV scripts,
and it's a major flag day to move to anything new.

> The way it is now, I see systemd as an unneeded complication which will
> break many many things before starting to work "right" for most people. And
> by most people I explicitly exclude "ready made" distribution users a la
> ubuntu etc., who are end users, and, who, by their own (!) definition,
> should do nothing more than push buttons and be rewarded with actions,
> reagrdless how those actions are achieved.

The biggest problem I see with systemd is all the bad blood that's sprung
up.  Lots of people have gotten crossed and are fighting it.  People
aren't going to upgrade to the next rev of their distro that uses it,
daemon authors are going to resist integrating compatibility, nobody is
going to publically admit to using it, and people will forget that there
ever was an emacs-vs-vi difference of opinion in the nuclear heat of this
new flamewar.

> Let's say I will be interested in systemd on small systems *after* openwrt
> and other embedded distributions adopt it *and* the inevitable anguished
> help cries on relevant forums die down a bit. That could take a year or two
> after they start using it, judging by how things went in the past. I so wish
> I am wrong about the time-frame.

I tried an in-place upgrade on an existing system, just to start getting
some systemd experience, and you do NOT want to deal with a system with
half-baked not-ready-yet daemon scripts for systemd.  That box got
repaved.

> Until then, I see systemd based kernels as a fork... harsh, but a serious
> problem for people who need to tinker under the hood frequently, as I have to.

And the other problem is that the hooks go all *over* your system
without any sensible justification, so it's too easy to conclude that
they're trying to be the new Borg and force-assimilate you.  See also
the Rob Landley posting Len already mentioned.

You catch more flies with honey than salt.

(Years ago I got a Linux box to behave by showing it a FreeBSD install
diskette and having a word or two.  Linux should not make me do that
again.)

-- 
Anthony de Boer
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists