News for hackers. Stuff that matters.
Re:The Tragedy of Systemd [video] by ilovecaching on Wednesday February 20, 2019 @09:34AM (#19208063)
What problems do people have with systemd? I actually think it’s pretty great. They also made some tools like busctl that are really nice for Linux developers.
Re:The Tragedy of Systemd [video] by isostatic on Wednesday February 20, 2019 @09:52AM (#19208217)
Something that annoyed me 10 minutes ago

Old fashioned way, when I restart apache having made a typo:

  Output of config test was:
  AH00526: Syntax error on line 2 of /etc/apache2/sites-enabled/mysite.conf:
  Invalid command 'blah', perhaps misspelled or defined by a module not included in the server configuration
  Action 'configtest' failed.

Now

   [....] Restarting apache2 (via systemctl): apache2.serviceJob for apache2.service failed because the control process exited with error code.
   See "systemctl status apache2.service" and "journalctl -xe" for details.
    failed!
The former is far more informative
Re:The Tragedy of Systemd [video] by gh02t on Wednesday February 20, 2019 @11:11AM (#19209018)
I mean I agree that it should just show you the output of "systemctl status" in the error message, but your comparison isn't entirely fair. The commands systemd error message give will show you the same thing as the generic sysv output.
Re:The Tragedy of Systemd [video] by isostatic on Wednesday February 20, 2019 @11:18AM (#19209070)
But it's that type of unhelpful behaviour that leads to dislike and accusations of arrogance (how dare you type a command that doesn't have the word "systemd" in it)
Re:The Tragedy of Systemd [video] by kbenson on Wednesday February 20, 2019 @11:45AM (#19209343)
Is it unhelpful, or just different? In the case of systemd, isn't it logging all the errors in the journal so you can see prior occasions it had a problem and errored? What is the error output in the jouirnal? Is it better, worse, or the same?

Systemd makes the process two steps, which you think is worse, but you've failed to address whether the possible benefits of those two steps (error retention, possibly more or different levels of error reporting) have been taken advantage of, and how that looks.

There's a difference between something being worse and just different. So far all I've seen is different, and possibly worse, but you've left out the information that would be required to determine that.

Re:The Tragedy of Systemd [video] by isostatic on Wednesday February 20, 2019 @12:34PM (#19209844)
When you're using an interactive shell to restart a process, and it fails, why wouldn't you want it to tell you why it failed.

  sudo systemctl restart apache2.service || systemctl status apache2.service
does the job, it's just ridiculous to seperate a system that used to be helpful.

Of course if you run sudo systemctl restart apache2.service You get no visual indication it's worked

  sudo /etc/init.d/apache2 restart
  [ ok ] Restarting apache2 (via systemctl): apache2.service.
tells you it worked, at least on this ubuntu 1804 machine.

This regressive change in behaviour (taking away pertinent information for no obvious benefit) leads to assumptions being made that things are getting worse for normal people. This view is then backed up by the volume of "defending systemd" statements that seem to stem from "scripting is hard".

People get turned off before they even begin to understand what systemd is trying to do, not because of core reasons, but because of one of the first interactions the average person has with systemd is to make their life harder.

Re:The Tragedy of Systemd [video] by kbenson on Wednesday February 20, 2019 @01:19PM (#19210388)
> This regressive change in behaviour (taking away pertinent information for no obvious benefit)

There is a benefit. The benefit is that it's logged, always, and through a mechanism that is the same for every process. It's a trade off. We can argue whether it's a worthwhile trade off, or one that even really needs to exist (is there a reason they don't immediately show an error? I don't know), but to say there is no obvious benefit even after I pointed out this specific benefit in my prior comment is odd.

> People get turned off before they even begin to understand what systemd is trying to do, not because of core reasons, but because of one of the first interactions the average person has with systemd is to make their life harder.

Any change from what existed previously that requires a different way of doing something, whether ultimately better or worse, makes the user's life harder in the short run since they need to learn something new.

That's not to say that's all that's happening here. There appears to be a regression in usability in at least one aspect. That said, I sort of understand why they didn't make it output that info. There have been plenty of times in the past for where sysV has reported a service started, but it actually failed very shortly after start. Generally, figuring out why can be painful, because STDOUT/STDERR have been redirected (or likely sent to /dev/null). Keeping a log of the stdout so you can see what happened on that last service action (and not hope it was logged to a log file, which it sometimes isn't...) is very useful. Outputting that the service started successfully is nice. Outputting that it started successfully when it didn't is misleading and more of a problem. Again, that's a trade-off, where it appears they've decided to favor accuracy over convenience in this case.

Re:The Tragedy of Systemd [video] by loeg on Wednesday February 20, 2019 @02:06PM (#19210924)
> The benefit is that it's logged, always, and through a mechanism that is the same for every process. It's a trade off. We can argue whether it's a worthwhile trade off, or one that even really needs to exist (is there a reason they don't immediately show an error? I don't know)

Yeah, it seems feasible (and maybe a good default) for systemctl to dump out the same last few lines it would for 'status' if an administrative request fails. I don't know why this has to be a tradeoff; systemctl actions could easily both log and print (some limited subset of) the log on error.

Re:The Tragedy of Systemd [video] by isostatic on Wednesday February 20, 2019 @02:39PM (#19211235)
Logging it is great, it's the fact the interface says "look in the log" rather than doing it for you.

It's tiny annoying things like that which leave a bad first impression. Rather than going "hey this system is great, it logs what it tells me", instead I'm left thinking "this system isn't for people like me, it doesn't think I need to know why it's broken"

First impressions are so important.

Re:The Tragedy of Systemd [video] by gh02t on Wednesday February 20, 2019 @01:21PM (#19210426)
I mean, I still think it's just a matter of being different. What systemd does is actually more consistent with the old Unix philosophy (in this case at least, systemd is hardly a paragon of Unixy-ness), where a command succeeding without a problem should just return silently. If it fails, you get a short message that says why and reminds you how to check the logs.

I agree that it should just output the log messages, but I don't think there is any serious UI problem here apart from just being different. It's also worth noting that the detailed error message you get from Apache isn't universal and depends on the app and how the init script is written. Plenty of sysv services just give you a cryptic error message on failure, e.g. when I restart postgres on my FreeBSD server with invalid configuration it just says "failed" and I have to then enter another command to see why.

Re:The Tragedy of Systemd [video] by sliken on Wednesday February 20, 2019 @11:55PM (#19214658)
Systemd is pretty good at hiding error messages. Even if you run the commands it claims will show you error message it makes stupid assumption that your window is infinitely wide or that you want lines truncated instead of wrapped. Even with an infinitely wide window you still don't get nearly as much debugging info as previous systems. For anything complex I end up reverse engineering the startup script and running it from the command line. Then if I update the service systemd notices the file changed, but instead of loading it it runs the old version and complains.

It really seems rather sysadmin hostile and replaces tried and true daemons like ISC's named or unbound with a rather poor systemd replacement.

Re:The Tragedy of Systemd [video] by npongratz on Wednesday February 20, 2019 @09:54AM (#19208237)
A litany of arguments against systemd are listed here: http://without-systemd.org/wiki/index.php/Arguments_against_...
Re:The Tragedy of Systemd [video] by nickik on Wednesday February 20, 2019 @11:11AM (#19209015)
Or other wise called a list of dependable design choice that the author doesn't like and a list of incompatibility with the tools the author used before.
Re:The Tragedy of Systemd [video] by evol262 on Wednesday February 20, 2019 @10:00AM (#19208298)
The actual speaker is great, but the vast majority of people who have a problem with systemd get there philosophically.

In a "I never actually had to write any code which had significant interaction with existing solutions (openrc, upstart, sysv, bsd, whatever), but I know they exist, so I'll talk about them like they solve the same problems as systemd" way. Which they do not. Want socket activation, double forking, namespaced children, sane scripts/unit files that can build a good dependency tree? You have one option.

Additionally, people see other tools like systemd-network and say "but systemd is doing everything!" Which it isn't. Almost no distros package those as part of the default.

Even if they did, many of the anti-systemd crowd would rather live with a different, slower system (NetworkManager, for example) which is vastly most complicated than what they need. Ditto for systemd-resolved and local caching dnsmasq.

Systemd is not trying to replace NM or BIND or anything. It's basically saying "it's 2019. If you want to be able to work with a system without knowing the ins and outs of 30 different configuration files, and speed up boot time/auditability of your system significantly, you can use these services"

Some things are bad for the community in general. The dependence on dbus (and kdbus), systemd-logind, and other tools aren't very friendly to BSDs to anti-systemd distros. Frankly, from my POV, this isn't systemd's problem. The anti-distros are free to do what they want, knowing the limitations. The BSD guys have written compatibility layers, which you can do, because they're open standards.

Broadly, I'd say that the anti-systemd crowd holds to an idea of "UNIX" purity, as if HP-UX, Tru64, AIX, Solaris, IRIX, and others were uniform. The reality is that they were more different than they were alike, in terms of administrative tools, filesystem layout (is this /sbin/init.d, /etc/rc.d, /sbin/rc.d, etc?) and even software portability. A large reason that GNU autoconf exists is to detect all of the quirks on all of these platforms.

Additionally, anti-systemd people either never used X11R6/OSS/lilo and other old tools, or don't remember how terrible they were. Any large change in the community (X.org fork, ALSA->PulseAudio, grub) draws ire. From the same people who are trying to use syslinux and efibootmgr instead of just using grub2.

Some people would rather debate philosophy than get actual work done, and that's great for open source, even if it's frustrating. But those people never dug into the guts of upstart to make something work, or had 250 lines of sysvinit boilerplate+supervisord saved off somewhere to speed up a process that you can do in 4 lines in a systemd unit file.

Re:The Tragedy of Systemd [video] by nickik on Wednesday February 20, 2019 @11:13AM (#19209030)
Solaris and other Unix had things like systemd already.
Re:The Tragedy of Systemd [video] by evol262 on Wednesday February 20, 2019 @01:58PM (#19210844)
SMF wasn't added until later in Solaris 10. Other unixes (Tru64, HP-UX, Ultrix, AIX, IRIX, etc) did not have anything like it. They were either BSD init or sysvinit
Re:The Tragedy of Systemd [video] by JdeBP on Wednesday February 20, 2019 @02:53PM (#19211364)
Rubbish. AIX had the SRC doing service management since 1992. And the SAF, also in Solaris, had been doing management of several classes of services since 1988.

* http://jdebp.eu./FGA/inittab-getty-is-history.html

* http://jdebp.eu./FGA/run-levels-are-history.html

* http://jdebp.eu./FGA/unix-service-access-facility.html

Re:The Tragedy of Systemd [video] by evol262 on Wednesday February 20, 2019 @07:49PM (#19213544)
This is _exactly_ what I'm talking about. Neither SAF or srcmstr are equivalent to systemd, and you know it. Making the argument that they're even in the same ballpark is fatuous, except insofar as they have some abstract concept of service groups and replacing sysvinit with something better.
[dead]
Re:The Tragedy of Systemd [video] by evol262 on Thursday February 21, 2019 @08:31AM (#19216702)
I honestly have no idea what you're talking about. The idea that UNIXes diverge from either BSD or SysV is not controversial. The fact that virtually every commercial UNIX had a completely different set of administration tools (other than vxfs as a add-on product to many) for any and every administrative task and init scripts in different locations are not orthogonal. "This system derives from SysV and that system derives from SysV, so I can administer them the same way" is complete bollocks and always was, even down to the location of the init scripts. That's the point.

At no point did I make any claim that inittab+runlevels was the "only thing going". I claimed that many systemd detractors hold to some abstract idea of a "pure" UNIX which never existed in the first place. As mentioned in my last comment (or my first comment) recognizing that there are fundamental limitations to traditional init and replacing or supplementing it with whatever (SAF, srcmstr, openrc, initng, you name it) and "being like systemd" are not even remotely the same thing except that they both manage service initialization.

This kind of false dichotomy is the point of the original post, which you missed in your fervor to turn it into a strawman. launchd, upstart, smf, runit, and other modern replacements are also not "like systemd" except that they manage services. SRC didn't replace init _at all_. It was a system run in parallel, as you well know. It was not a replacement for sysvinit, and AIX still has sysvinit. Again, as you well know, but it doesn't fit your narrative.

Re:The Tragedy of Systemd [video] by seba_dos1 on Wednesday February 20, 2019 @09:53AM (#19208232)
busctl looks simply like mdbus made 10 years later