On Fri, Jan 16, 2015 at 1:39 PM, Lonnie Olson <[email protected]> wrote:

> Before you call me ignorant, BS'er, or a hater, I am not a systemd
> lover or hater.  Lennart has great technical chops, and has
> implemented many great ideas into systemd.  It is way better than the
> old sysvinit systems.  Though I have concerns with what it is becoming
> over time.

I'm not sure why so many people feel compelled to voice so loudly
their concerns about systemd, especially when they involve a lot of
guessing about the future.  Mountains are being constructed from...
well, maybe not molehills, but certainly lesser mounds.  I appreciate
that you're not name-calling or personally slagging anyone here, but I
don't think you're articulating your concerns very well.

> The very first question asked about the biggest misconceptions.
> Lennart answered with two of them, monolithic and not like UNIX.  He
> completely ignored talking about the first one, probably the most
> important one.  Then he, seemingly willfully, misunderstood the not
> UNIX argument, and went off on a tangent.  What is generally meant by
> this argument is that systemd goes against the general UNIX philosophy
> of small things that do one thing well working together.
> (http://en.wikipedia.org/wiki/Unix_philosophy).   The interviewer did
> not call him on it, nor did he ask about the complaint of being
> monolithic.

It seems to me that people making the "UNIX philosophy" argument are
extremely inconsistent in applying that particular aspect of
criticism. Quite a lot of Linux is made of big monolithic things. And
most of the big things that are taking on systemd as a dependency
(i.e. Gnome stuff) are certainly not very adherent to the "UNIX
philosophy" either. If you want a system that's more philosophically
pure, Plan 9 is right over there.

I think Pottering is right to dismiss this argument, and the
interviewer is right to not press him on it. The purity of philosophy
ship sailed a long time ago, and it's not coming back. You can still
build a Linux system mostly out of pieces that adhere to the UNIX
philosophy (aside from the kernel itself, which is itself a big
monolithic thing with all sorts of feature creep going on) but it's
going to look and feel pretty different than what most users of Linux
want it to look and feel like.  Once you start adding in Gnome stuff,
the weight of arguments about Unix philosophy diminishes rapidly.

[... stuff I have no comment on elided...]

> Finally in the sixth question he was asked about his feature creep
> problem.  Which was ignored for most of the answer.  It wasn't until
> toward the end when he just said those are just optional, which is
> mostly true, but still somewhat troubling for me.

Why are you so troubled by this? Systemd has taken a lot of
tangentially-related projects under its wing, but a lot of them are
"associated systems" and not part of the core PID 1 daemon. There's a
huge demand today for a lot of complex services and programs to run in
a Linux system, from small embedded single-purpose systems to
smartphones to desktops and headless servers.  There's a huge
variation in *which* systems run on any particular install, and how
the various pieces rely on one another. It's an incredibly difficult
problem to solve well across a wide range of configurations, and
solving it well depends on a lot of pieces of core system
configuration functionality coordinating their work. Flat text files
and shell scripts just don't cut it for a lot of these systems.  I
work with, or have worked with, Linux systems pretty much throughout
this entire range; I am personally really excited about the parts of
the Linux system that systemd cleans up for most of the
non-"traditional server" range of Linux systems.

Naturally, a program designed to be at the nexus of coordination and
configuration is going to have to either re-implement a bunch of
existing things or else influence the existing maintainers of those
things to ensure the coordination works properly in some cases. As far
as I can tell, systemd goes out of its way to provide the ability to
continue to use the existing, less-integrated programs and methods for
any areas that a system integrator values over whatever cases the
systemd maintainers needed a more integrated solution for.  But when
closer coordination is needed, the complaints of random people on the
internet that systemd is suffering from "feature creep" take a distant
second seat to real-world issues regarding configuration and system
startup race conditions.

The fact that groups like Gnome are switching to the systemd services
by default over the previous semi-solutions has less to do with the
monolithic nature and scope creep of systemd and more to do with the
need Gnome has for tightly-integrated management of system
configuration information and events. Why yell at the systemd folks
over this when there's nobody else working on the sort of stuff that
they need? The people that fear lock-in to systemd for distros using
Gnome, or who fear the lack of portability of systemd, need to step up
and work on providing the system-level stuff Gnome and similar
projects need instead of complaining that people are making pragmatic
choices with the development manpower they've got.

> TL;DR  Not a good interview, mostly a puff piece that lets Lennart
> rehash stuff he has already said in blogs, Google+ posts, etc.   A few
> interesting points, but not very good overall.

It seems you didn't like it because it didn't really address your
concerns, and you felt that the interviewer ought to have shared your
concerns. That's fine, but not really an objective criticism of the
interview.  I hope I've been able to address some of your concerns, or
at least draw out a more descriptive statement of what they are.

     --Levi

/*
PLUG: http://plug.org, #utah on irc.freenode.net
Unsubscribe: http://plug.org/mailman/options/plug
Don't fear the penguin.
*/

Reply via email to