LWN: Comments on "Positions forming in the Debian init system discussion"

Positions forming in the Debian init system discussion

faheem — Tue, 21 Jan 2014 17:48:43 +0000

> And nobody argues (well, except for Canonical employees) that systemd is really more advanced.

Did you mean upstart here?

unhelpful comments

neilbrown — Mon, 20 Jan 2014 21:01:15 +0000

I agree that if the core developers appear uninterested in any problems but their own, then that is a cause for concern.

On the other hand, looking up binaries in $PATH has a well-known solution. Just run

    /usr/bin/env some-binary some-arguments

and /usr/bin/env does the path lookup. i.e. it is a feature that is trivially added after-market. If lots of distros start doing this, that might help convince systemd upstream to make it a native feature.

unhelpful comments

rodgerd — Mon, 20 Jan 2014 20:54:20 +0000

As opposed to the landgrab of making a free software distribution depend on the CLA-ed software of a for-profit company as a core component?

unhelpful comments

amck — Mon, 20 Jan 2014 19:46:57 +0000

In fact, I think that Ian completely misses the big (ecosystem) picture in his email, instead resorting to handwaving and pseudo technical arguments (look up binaries in $PATH, seriously?). Russ otoh clearly understands what's going here and what opportunity systemd provides. I can only hope that the remaining CTTE members also do.

I think that Ian here is seeing the big picture, and this example highlights it. Looking up binaries in $PATH is done in Debian because of things breaking during upgrades, etc. where binaries move. If systemd doesn't work this way, then upgrading Debian breaks hard. You need to look beyond the immediate technical solution to overall maintainability. One of Debian's strengths is its large technical developer base, with an institutional history of previous problems solved. Here systemd developers ignore a problem outside their experience but which will break other code. This is a worrying feature and a real fear.

Positions forming in the Debian init system discussion

oldtomas — Fri, 17 Jan 2014 10:34:17 +0000

> The most obvious problems with this are that starting daemons via /etc/inittab doesn't give you service ordering even in the clumsy way SysV init does [...]

Yep. That's why I talked about the mid-90ies (no, I wasn't trying to show off what an old fart I am ;-) and hand-waved about evolution path. I do know that when I first saw this SysV bunch-of-scripts I thought "oh, my! what happened to respawn?" (there were other thoughts about those brutal shell epics in HP/UX, but I disgress).

Anyway, this has taken us astray from the original intention of my post, and that was that beyond "technical excellence", which is, of itself not a well-defined thing anyway, there are other factors *for me* on which to judge whether I want to use/take part in some software. And systemd and its "environment" fail on those other accounts so much (I repeat: *for me*, probably not for you) that I refuse to even have a deeper look at it.

Positions forming in the Debian init system discussion

smurf — Thu, 16 Jan 2014 19:44:58 +0000

> as long as the community around systemd behaves the way it does

Sorry, but I don't see a problem here. All I see are people who know what they're doing, whose software _works_, and has a ton of features certain other software does not have. In my book, that gives them some leniency in the "how much arrogance willl I put up with" department -- of which I don't see all that much in the first place.

I _have_ worked with people (and/or their software) who were so sure of themselves that there was absolutely no way to convince them to change a single bit in their source code.

The authors of systemd are not members of that set.

IMHO.

Positions forming in the Debian init system discussion

anselm — Thu, 16 Jan 2014 18:24:43 +0000

Of course, one could use inittab for daemons under Linux too, but the point is that it was customary in the mid-90's to shepherd system daemons this way, thus potentially controlling their run status and their stdin/stdout/stderr and reacting to their termination via signals.

The most obvious problems with this are that starting daemons via /etc/inittab doesn't give you service ordering even in the clumsy way SysV init does (so no making sure Apache only starts when syslogd and MySQL are already running), and that there is no provision on the part of init for actually dealing usefully with a daemon's stdin/stdout/stderr, which Unix-style daemons customarily close when they start, anyway (and that convention is older than the mid-90s). There's a reason why SysV init works like it does.

This of course completely disregards the fact that systemd does many other useful things even when it is simply starting service processes based on a »runlevel-like« target (leaving aside obvious nifty things like socket activation). All of this would have to be retrofitted into the SysV init infrastructure and would only make that more convoluted and non-standard.

Positions forming in the Debian init system discussion

oldtomas — Thu, 16 Jan 2014 17:42:07 +0000

> The process described in FreeBSD's init(8) sounds almost identical to Linux's sysvinit scheme (/etc/ttys : /etc/inittab and /etc/rc.d : /etc/rc?.d). Is this what you mean, or are you talking about something else?

Yes and no. I was talking about this (cf. your reference):

> The init utility can also be used to keep arbitrary daemons running,
> automatically restarting them if they die. In this case, the first field
> in the ttys(5) file must not reference the path to a configured device
> node and will be passed to the daemon as the final argument on its com-
> mand line. This is similar to the facility offered in the AT&T System V
> UNIX /etc/inittab.

Of course, one could use inittab for daemons under Linux too, but the point is that it was customary in the mid-90's to shepherd system daemons this way, thus potentially controlling their run status and their stdin/stdout/stderr and reacting to their termination via signals.

This would have been a nice evolution path towards something interesting (and those are the things systemd gets right, IMHO)

Positions forming in the Debian init system discussion

cortana — Thu, 16 Jan 2014 17:08:13 +0000

yes, the PID file stuff was clumsy in SysV init from the very beginning (remember: BSDinit had it already right then)

The process described in FreeBSD's init(8) sounds almost identical to Linux's sysvinit scheme (/etc/ttys : /etc/inittab and /etc/rc.d : /etc/rc?.d). Is this what you mean, or are you talking about something else?

Positions forming in the Debian init system discussion

oldtomas — Thu, 16 Jan 2014 16:18:32 +0000

> Which point would that be?

That (for me, at least) there are several criteria beyond technical on which to decide which software I use and associate with. That while systemd has a couple of nifty ideas -- (x)inetd could rip off some of that socket association thing; yes, the PID file stuff was clumsy in SysV init from the very beginning (remember: BSDinit had it already right then). CGroups, DBus -- meh, for me (the CGroup thing becoming negative with the last developments in the kernel front). The mixing of lots of policy into a bunch of binary: negative.

But: as long as the community around systemd behaves the way it does, I won't touch it with a ten foot pole. That's the deal breaker for me.

Positions forming in the Debian init system discussion

Cyberax — Thu, 16 Jan 2014 08:35:21 +0000

The only problem is that they _are_ right.

I was initially against systemd, but then I actually tried to use it... It is really a brilliant piece of software.

Positions forming in the Debian init system discussion

smurf — Thu, 16 Jan 2014 08:33:44 +0000

Which point would that be? This discussion is split about some minor advantages of using upstart, some major features of systemd which upstart doesn't have and won't get any time soon, and some non-technical problems which some upstart proponents have with systemd and/or one of its main authors.

One can only respond calmly to the latter sort of 'argument' a finite number of times before losing one's patience. That is not the problem, that's basic human nature.

The grandstanding of some upstart proponents on a "The systemd repository is taking over Linux! Run away!" platform is the problem. (I know you don't see it that way. But I do.)

Positions forming in the Debian init system discussion

oldtomas — Thu, 16 Jan 2014 07:44:23 +0000

> [Wol] it is clearly the best init system currently available on linux

> [raven667] some people would prefer an apology or contrition and are skeptical of confidence.

> [anselm] Lennart Poettering does not appear to suffer fools gladly

> [jezuch] "confidence in one's work" [...] characterizing this as "objectionable" is laughable

> [anselm] if you can't attack the work, attack the author

And so on and so forth. In a nutshell "if you don't share our opinion, you're an idiot, because We Are Right (TM)"

Thanks for making my point even more clearly that I could ever have made it.

various options

nye — Mon, 13 Jan 2014 14:24:25 +0000

>In order to take away rights on Windows, you need special contortions.

Or you could set it to 'deny', rather than leaving it at the default of 'inherit'. If you're already setting a user ACL, this adds no extra work. No contortions required, and it allows a greater flexibility in how permissions are assigned.

Positions forming in the Debian init system discussion

smurf — Mon, 13 Jan 2014 09:56:38 +0000

Well, Lennart did voice his opinion in language which can't in good conscience be described as "polite".

However: (a) systemd is not just Lennart's work, some major ideas (and code) are from other people, (b) this perception is from the early days of pulseaudio and, in all fairness, those days are long gone.

Of course, if somebody has a history of arrogant comments, pretty much anything they say will be perceived as more of the same even though the same comments would be regarded as perfectly harmless if voiced by somebody else.

To me, the attitude of the upstart proponents as I perceive it of "we can do everything systemd can do, it's just a Small Matter Of Programming … which we haven't done yet and don't have plans to do either, esp. since we have no idea how much work it'd actually be, but you STILL should choose our model" is far more grating.

Positions forming in the Debian init system discussion

anselm — Mon, 13 Jan 2014 08:44:02 +0000

This looks like the time-honoured recipe of »if you can't attack the work, attack the author«. Lennart Poettering could say that the sky is blue and the sun comes up in the morning and there would be people claiming the contrary as a matter of principle.

Positions forming in the Debian init system discussion

jezuch — Mon, 13 Jan 2014 08:37:40 +0000

> behavior that is deemed most objectionable by systemd proponents is the lack of humility and the persistent characterization of systemd as superior

I think that's called "confidence in one's work". Frankly, characterizing this as "objectionable" is laughable.

Positions forming in the Debian init system discussion

rahulsundaram — Fri, 10 Jan 2014 19:31:31 +0000

That may well be but the results are pretty amusing

http://mirror.linux.org.au/pub/linux.conf.au/2014/Monday/...

Positions forming in the Debian init system discussion

anselm — Fri, 10 Jan 2014 16:14:53 +0000

One issue is that Lennart Poettering does not appear to suffer fools gladly. This is a trait that he shares with various other prominent people in our community – including, for example, Linus Torvalds – but also one which, for whatever reason, people are less forgiving of in his case than they are elsewhere, so they tend to be just as obnoxious in return. This does nothing to advance the level of the discussion on both sides.

So far in the Linux community there has been no need to »apologise« or show »contrition« for technical excellence. If systemd is persistently characterised as »superior« this may well be because by most criteria, as an init system for Linux it actually is superior to the alternatives, and by a fairly wide margin. In fact we do not need Lennart Poettering to tell us this (and frankly, what would you expect him to say?) if we have independent assessments such as that of Russ Allbery in the article at the root of this discussion. Well-founded opinions like Russ's are especially important given that many of the people who are very vocally against systemd do not actually appear to have looked at the issue in any kind of detail, and that shows in the type and quality of counterargument one tends to encounter.

Positions forming in the Debian init system discussion

raven667 — Fri, 10 Jan 2014 15:26:33 +0000

My take on it is that the behavior that is deemed most objectionable by systemd proponents is the lack of humility and the persistent characterization of systemd as superior, some people would prefer an apology or contrition and are skeptical of confidence.

Positions forming in the Debian init system discussion

Wol — Fri, 10 Jan 2014 00:28:06 +0000

:-)

WHAT vocal proponents?

All I can see in this thread is people pointing out the FACT that systemd takes advantage of all the facilities provided by linux - which is true!

As a result, it is clearly the best init system currently available on linux. But - as an equally obvious result - it will be a pig to implement *well* on any other kernel.

(Although I have seen a few comments where people have objected to systemd being mis-characterised. But what's wrong about objecting to misleading / false claims?)

Cheers,
Wol

various options

Wol — Thu, 09 Jan 2014 23:47:35 +0000

I've never used linux ACLs, but I have used them on Pr1mos. And the version on Windoze is crap - I hope linux didn't copy that!

Doze *adds* user and group ACLs together. I gather you can get round that, but ...

Pr1mos gives you your user ACL *or* your group ACLs (or the default ACL if neither apply).

Thus making it extremely easy to control who has what access. "any old user" will get $DEFAULT. You can give group acls to departments or whatever, and people then get the access they need by default (Pr1mos adds group acls together so you get the sum of all your group rights).

BUT - if you explicitly set a user ACL, then that is *exactly* what they get! If their group acls give them a bunch of rights that their user acl doesn't, then the group acls are ignored (well, they're ignored anyway :-)

In order to take away rights on Windows, you need special contortions. In Pr1mos, you just set a user acl.

Cheers,
Wol

Positions forming in the Debian init system discussion

raven667 — Thu, 09 Jan 2014 15:16:13 +0000

> After having witnessed (with some disbelief and some horror) this thread, I'm more convinced than ever that I don't want systemd near any of my computers. Why, you ask? The attitude of its more vocal proponents.

Why, do processes start with the wrong color if they are exec'd by systemd? Will the attitude of some person on LWN who you disagree with cause RAM failure?

Positions forming in the Debian init system discussion

anselm — Thu, 09 Jan 2014 10:31:49 +0000

What about the attitude of its more vocal detractors?

Positions forming in the Debian init system discussion

oldtomas — Thu, 09 Jan 2014 09:52:06 +0000

After having witnessed (with some disbelief and some horror) this thread, I'm more convinced than ever that I don't want systemd near any of my computers.

Why, you ask?

The attitude of its more vocal proponents.

Positions forming in the Debian init system discussion

torquay — Thu, 09 Jan 2014 05:45:46 +0000

That's not really relevant or important.

I beg to differ. It is directly relevant. If kFreeBSD doesn't provide the necessary features now, who is going to write another kernel from scratch which will provide the features in the future? The Hurd kernel has been around for a long while, and yet its development pace is slower than the proverbial snail. And if more choice is good, why not modify Debian to allow the option of using the ReactOS kernel?

By leaving all the options open, all the manpower and effort will be put into maintaining the resultant mess, instead of being directed towards something more useful.

The vote isn't over yet

rahulsundaram — Thu, 09 Jan 2014 04:26:20 +0000

If that is your objection, what are you recommending for Debian? None of the other leading contenders are going to be adopted for anything else either. They might make a port available but other operating systems aren't going to use it by default. They all have their own systems.

The vote isn't over yet

The_Barbarian — Thu, 09 Jan 2014 04:14:37 +0000

>systemd is fairly low level and no other operating system is going to adopt it.

That may well be, if so, that is exactly why it should not be used for Debian. I have no objection to other distros using it.

The vote isn't over yet

rahulsundaram — Thu, 09 Jan 2014 03:23:04 +0000

All things being equal, portable code is indeed better but I don't really see much of a problem here. Characterizing systemd as hostile to portability seems off the mark to me since they are just being honest about the status. systemd is fairly low level and no other operating system is going to adopt it. It takes advantage of a ton of Linux specific features and porting in this case either means reimplementing it with the same interfaces or implementing the specific features that Linux has that other kernels lack.

The vote isn't over yet

The_Barbarian — Thu, 09 Jan 2014 01:31:13 +0000

I use Debian/Linux, and I currently have no plans to use hurd or kfreebsd. But I am against systemd due to hostility to portability (though I acknowledge that other than that it is clearly superior). I wish it was otherwise.

Positions forming in the Debian init system discussion

rahulsundaram — Wed, 08 Jan 2014 18:46:55 +0000

TC clearly believes (and references the constitution) that this is within their purview. Ian wrote key parts of that governance document and is in TC. So I see that as authoritative.

The vote isn't over yet

HelloWorld — Wed, 08 Jan 2014 18:17:13 +0000

sysvinit is terrible, obviously.

Positions forming in the Debian init system discussion

jwarnica — Wed, 08 Jan 2014 17:15:05 +0000

Knowing as little as possible about internal Debian politics, I would imagine it is still entirely reasonable for the TC to decide "This isn't a technical question, and/or also a strategic question which we are not qualified to even make a recommendation on"

The vote isn't over yet

hummassa — Wed, 08 Jan 2014 15:59:35 +0000

In your reference, what would be "terrible"?

The vote isn't over yet

dlang — Tue, 07 Jan 2014 18:01:31 +0000

it's a tool name

http://www.reference.com/motif/reference/why-is-it-called...

The vote isn't over yet

HelloWorld — Tue, 07 Jan 2014 17:28:55 +0000

> What I am saying (in a rather unpleasant way, I agree) is that no matter which way the decision goes, Debian will end with a solution that's technically superior to the status quo.
I just read a nice quote from some blog[1]: “improving on ‘terrible’ doesn’t necessarily mean the result is anywhere near ‘good’”.

[1] http://grundlefleck.github.io/2013/06/23/using-scala-will...

The vote isn't over yet

smurf — Tue, 07 Jan 2014 17:06:28 +0000

… except that the number of Debian/non-Linux users is … somewhat low … compared to those of /Linux.

The vote isn't over yet

mathstuf — Tue, 07 Jan 2014 16:33:23 +0000

> I won't scratch my disk drives with a bastard file

Is this a saying from somewhere? I'm not familiar with the idiom (if it is one); I can guess the connotation, but if there's some subtext… :) .

The vote isn't over yet

jubal — Tue, 07 Jan 2014 12:58:48 +0000

What I am saying (in a rather unpleasant way, I agree) is that no matter which way the decision goes, Debian will end with a solution that's technically superior to the status quo.

Personally, I prefer upstart – but I won't scratch my disk drives with a bastard file if the tech-ctte decides to choose systemd.

The vote isn't over yet

jezuch — Tue, 07 Jan 2014 11:02:44 +0000

> So you're saying that Debian shouldn't care what its users want and do?

Which users? It's you against the users of Debian/kfreebsd. They are users too, you know. So you're saying that Debian shouldn't care what its users want and do?

(Disclosure: I'm a Debian user and I'm in favor of systemd. But your position is... dishonest.)