[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: I'm not a huge fan of systemd

On Sat, Jul 12, 2014 at 11:26 AM, Miles Fidelman
<mfidelman@meetinghouse.net> wrote:
> Tom H wrote:
>> On Fri, Jul 11, 2014 at 1:21 PM, Brian <ad44@cityscape.co.uk> wrote:
>>> On Fri 11 Jul 2014 at 12:21:57 -0400, Steve Litt wrote:

>>>> A bizarre thought just popped into my head, in the form of a little
>>>> voice. The little voice told me that if they guys who controlled the
>>>> decision to go to systemd had been the decision makers in 1990, Linux
>>>> would have a microkernel today.
>>> This thread has been quite interesting. In it a number concerns about
>>> system have been comprehensively answered.
>>> * Double-forking is not a problem.
>>> * Logs are in text format.
>>> * libsystemd-daemon0 doesn't affect users of other init systems.
>>> All useful stuff and (given the lack of substantial counter-argument)
>>> cause for regarding these questions as settled.
>> The s/n ratio would be higher in threads such as these if people read
>> the documentation...
> Now that is a rather gratuitous and disingenuous statement.

It's neither gratuitous nor disingenious. There's a pattern of someone
finding that something doesn't work the way that someone expects and
rather than going through the documentation or google first, there's a
post here with a more or less gleeful "I've found something broken in

There's a lot of systemd criticism in any systemd thread. I understand
that people are expressing their frustration with this change but it's
nothing but noise. I work in finance and the traders have a saying
"you trade the market that you have and not the market that you wish
you had."

We now have systemd as the jessie init default and I wouldn't be
surprised if it weren't required for all jessie+1 DEs given how things
are evolving. Two anti-systemd GR proposals were launched. I'm not
sure what happened with the first but the second was withdrawn because
it didn't couldn't the five required co-sponsors. So even the DDs have
accepted systemd, perhaps/often grudgingly, and for us users to be
complaining in every systemd thread about systemd is illogical. As
Debian users we've implicitely accepted Debian's governance rules.
There were competing proposals, there was a vote, and there was an
opportunity to override that vote; and, in the end, systemd was
Debian's choice. There are things that I dislike about systemd, its
Debian maintainers, and its upstream maintainers but to paraphrase
what I said above about markets, these are the people whom we have.

> If one is installing a new program, the first time, reading the
> documentation is highly recommended.  But this is a different situation
> entirely:  Systemd is part of the o/s, and one that has a number of
> potential impacts - both intended and unintended -  on all kinds of things,
> in non-obvious ways.

It's because it's a major building block of the OS that reading the
documentation ought to be required of anyone who wants to administer a
system (imposed upon oneself, not the distro!).

> First off, one doesn't expect to have to pour through documentation every
> time one does an o/s upgrade.  Just the release notes and the "before you
> upgrade" part of the installation instructions. Right now, the release notes
> for Wheezy simply say:
> -------
> Debian 7.0 introduces preliminary support for systemd, an init system with
> advanced monitoring, logging and service management capabilities.
> While it is designed as a drop-in |sysvinit| replacement and as such makes
> use of existing SysV init scripts, the |systemd| package can be installed
> safely alongside |sysvinit| and started via the |init=/bin/systemd| kernel
> option. To utilize the features provided by systemd, about 50 packages
> already provide native support, among them core packages like udev, dbus and
> rsyslog.
> systemd is shipped as a technology preview in Debian 7.0. For more
> information on this topic, see the Debian wiki
> <http://wiki.debian.org/systemd>.
> -------
> and the "preparing for update" section says nothing about systemd impacts to
> watch out for.
> Of course the release notes and installation instructions for Jessie have
> yet to be written - it is, after all, "testing."
> The documentation on the Debian wiki is likewise kind of sparse when it
> comes to "known issues and workarounds."

It's all in a beta state including the notes and wiki. The whole point
of a distro is that it lessens the installation and upgrade burdens on
its users. Whether they'll be as easy and uneventful as this
documentation promises will be proven when jessie is released. You can
package a leaf application without much integration but an init
system's going to necessitate a lot of work and a lot of testing.

RHEL 7 has been released with systemd as its init so releasing an
enterprise, stable distro with systemd is possible.

> Documentation at the upstream site is a bit more comprehensive, if not well
> organized.  You have to dig a bit to find things like:
> --- "If your distribution removes SysV init scripts in favor of systemd unit
> files typing "/etc/init.d/foobar start"
> --- as a server admin, that's going to impact me, but
> --- but.. there's nothing in the Debian documentation that tells me if my
> "distribution removes SysV init files" (Wheezy, obviously not; but what
> about Jessie?)

"/etc/init.d/foobar start" has been deprecated since "service ..." was
introduced to Debian. As has been covered before here, using "service
..." cleans up the environment in which the daemon's launched. You can
use "systemctl ..." with sysvinit scripts and systemd units but I
can't remember whether you can use "service ..." with systemd units.
AFAIR the plan's to have that working in jessie.

If a daemon has both a systemd unit and sysvinit script and you're
running systemd as pid 1, it's the systemd unit that'll launch that

> Now, I would hope that by the time Jessie gets released as stable, bugs and
> impacts have been minimized, and that the release notes and installation
> instructions are comprehensive when it comes to systemd impacts - but we're
> from there.
> The discussions on this list are precisely what will lead to such
> documentation by:
> a. alert many of us to issues and impacts before we migrate (particularly
> impacts on things that won't be caught by package maintainers)
> b. provide input to package maintainers, and maybe upstream developers, on
> things that might need some work as a result of systemd
> c. provide raw material for the release notes and installation instructions
> for Jessie

This list doesn't have "discussions." It has rants interspersed with
some technical points. Good luck to whoever (if there is such a
person) who'll rummage through debian-user@ threads for release notes
and documentation material.

> In other words, a pox on your "read the documentation" snipe.

You selfishly expect others to read the documentation for you...

Reply to: