systemd: Yet Another Init Replacement

litemotiv · 2011-05-10 21:58:50

Glaucous wrote:

systemd-1 mounts? Not quite sure what this means, and how can I check it?

If you look in /proc/mounts/, do they look like this:

systemd-1 on /media/mountpoint type autofs (rw,relatime,fd=14,pgrp=1,timeout=300,minproto=5,maxproto=5,direct)

Glaucous · 2011-05-10 22:03:14

Similar.
http://codepad.org/7wQP5mJK

litemotiv · 2011-05-10 22:12:24

Glaucous wrote:

Similar.
http://codepad.org/7wQP5mJK

Right, if that is the same after boot then they are mounted...

ethail · 2011-05-11 07:54:27

Why is in that file / showed as mounted twice?

rootfs / rootfs rw 0 0
/dev/disk/by-uuid/69c7a380-dcf0-45bf-acdf-4c8b1a059d75 / ext4 rw,noatime,barrier=1,data=ordered,discard 0 0

I'm in quite a similar situation and I don't know if it is normal

EDIT: Nevermind, found this 2 minutes later:
https://bbs.archlinux.org/viewtopic.php?id=99417

Last edited by ethail (2011-05-11 08:42:59)

WorMzy · 2011-05-11 14:08:15

How is everybody running httpd with systemd? Have you written your own .service file?

All I have is a http-daemon.target file, which is apparently part of systemd.special, although I don't understand why, or how I'm supposed to use it. At the moment I assume systemd is running httpd due to it being included in the daemons array in rc.conf, but I'm aiming to cut rc.conf out of the picture completely and have a pure systemd setup.

It shouldn't take long to write up a httpd.service, but I'm curious as to why there isn't already one in the systemd or systemd-arch-units packages.

hdhoang · 2011-05-11 17:05:42

I have a lighttpd.service at https://github.com/hdhoang/systemd-arch … e/lighttpd (with a tmpfiles conf). It works for my simple usage, but as discussed in https://github.com/falconindy/systemd-a … ts/pull/15 it doesn't always do the right thing.

Possibly the debian and red hat packagers have got an apache service, ask someone on #systemd.

daasdingo · 2011-05-11 17:34:56

hi,

I am booting with systemd, but since I switched I do not have virtual terminals anymore, which is odd since there are some agetty processes running.
When want to switch to another virtual terminal, there is a black scree with two tiny white dots at the top, very odd.

FYI, I am using ordinary nvidia binary drivers and Xorg, no special repos or testing.
Has anyone else had this problem? can't seem to find this in the wiki

EDIT: I fixed it, it was a problem with grub2 automatically passing the kernel display size parameters which seems to have messed up the display of virtual consoles.

PS: what I think is weird when booting with my SSD, is that it is really fast to boot to GDM(about 6 seconds), but initializing my keyboard takes an additional 10 seconds or what

Last edited by daasdingo (2011-05-13 19:51:36)

sidneyk · 2011-05-12 14:59:28

I think I've finally got systemd working properly on my system. It does seem to boot a little faster. I suffered a little problem with shutdown always setting my RTC to local time even though I previously had Win 7 and Arch adhering to UTC time. I made a guess from the wiki about where to put the UTC setting mentioned and now all is OK with time again. I tried to follow along with the wiki and various other posts here and other places to set things up as "pure" systemd as possible. I had plymouth installed prior to installing systemd and the init-scripts pkg, but since it wasn't working anyway I uninstalled and disabled the services and purged any other settings related to plymouth that I could remember making. But after many, many reboots since then I still see plymouth related stuff showing up with systemadm. It shows them as installed but inactive. I've tried several times disabling them one by one but they keep showing up. Is this the way it's supposed to be? Also, I notice that systemd-vconsole-setup.service is taking the most time from my boot usually. Is this service necessary and what exactly does it provide? I'm running Gnome 3 / Gnome Shell. Overall, it looks pretty promising and I agree that it would be nice to have some cross-distro standards in place. Keep up the good work!

Last edited by sidneyk (2011-05-13 13:21:06)

eDio · 2011-05-15 07:58:14

I've faced problems with shutdown, suspending to ram and disk while using systemd.

EDIT: suspend problem is caused by mouse Genius NAVIGATOR 535. Will create separate thread for this

Shutdown
computer will always stuck on shutdown, displaying login prompt on the tty. After few seconds appears message

USB port 5 disabled by hub (EMI?)

Input devices wont work and the only way to shutdown is hard shutdown by power button.
The worst thing in this, is that partitions remain mounted and hard shutdown causes non-clean unmount.
I use native systemd configuration files already (for modules and daemons) and don't use arch-persistent-settings

I had no problems with shutdown before.

What would you suggest to solve this?

Thanks in advance

Last edited by eDio (2011-05-15 09:16:26)

Zom · 2011-05-15 10:52:34

eDio wrote:

I've faced problems with shutdown, suspending to ram and disk while using systemd.
EDIT: suspend problem is caused by mouse Genius NAVIGATOR 535. Will create separate thread for this
Shutdown
computer will always stuck on shutdown, displaying login prompt on the tty. After few seconds appears message
USB port 5 disabled by hub (EMI?)
Input devices wont work and the only way to shutdown is hard shutdown by power button.
The worst thing in this, is that partitions remain mounted and hard shutdown causes non-clean unmount.
I use native systemd configuration files already (for modules and daemons) and don't use arch-persistent-settings
I had no problems with shutdown before.
What would you suggest to solve this?
Thanks in advance

If this is the same issue that others had in this thread, the cause is syslog-ng.service. The suggested fix is to use rsyslog instead. The package also installs its own .service-file.

litemotiv · 2011-05-15 12:56:45

I'm having a go at creating a privoxy.service file:

[Unit]
Description=Privoxy

[Service]
ExecStart=/usr/sbin/privoxy --user privoxy.privoxy --pidfile /var/run/privoxy.pid /etc/privoxy/config

[Install]
WantedBy=multi-user.target

If i run the command manually it starts up okay, but as a service it just dies:

privoxy.service - Privoxy
          Loaded: loaded (/etc/systemd/system/privoxy.service)
          Active: inactive (dead) since Sun, 15 May 2011 14:50:43 +0200; 1min 1s ago
         Process: 1951 ExecStart=/usr/sbin/privoxy --user privoxy.privoxy --pidfile /var/run/privoxy.pid /etc/privoxy/config (code=exited, status=0/SUCCESS)
          CGroup: name=systemd:/system/privoxy.service

I don't see it running with ps as well.

Any idea why this happens? I have a similar problem with a net-auto-wireless service file. I looked at existing service files for inspiration but i don't see anything really different there..

falconindy · 2011-05-15 13:15:16

You've declared privoxy to be a 'simple' service, but looking at the command line, you're telling privoxy to daemonize itself, so systemd sees the privoxy service detach and go away on exec.

Two options:

1) Add 'Type=forking' to the unit file along with PIDFile=/var/run/privoxy.pid (you though you really should be using /run/privoxy.pid)
2) Add --no-daemon to the ExecStart, ditch the --pidfile, and add 'StandardOutput=syslog'

I prefer option 2.

In both cases, you can probably also add 'User=privoxy' and 'Group=privoxy' in exchange for the --user flag.

litemotiv · 2011-05-15 13:20:55

Excellent, thanks for the explanation.

falconindy · 2011-05-16 13:04:52

I've mentioned it on the wiki, but I'll mention it here as well in case anyone is currently in this position. I discovered over this past weekend that using BFS leads to some subtle issues with systemd -- in particular, my expired SSH sessions (socket-activated) were not being properly garbage collected. They did eventually expire with an error, but dead units would pile up for several hours, or until I manually called a GC/reload. I can only assume that this is due to BFS not interfacing with cgroups as the default CFS scheduler does. I suggest using the autogrouper heuristic of CFS (included with the stock kernel26), which will illicit similar behavior as BFS under heavy desktop load. I cannot speak for the remainder of the CK patchset as I don't use it.

edit: CFS, not CFQ

Last edited by falconindy (2011-05-16 14:38:14)

loonyphoenix · 2011-05-16 13:07:38

I think you mean CFS (cpu scheduler), not CFQ (io scheduler)? Or is the problem with BFQ (io scheduler), not BFS (cpu scheduler)?

falconindy · 2011-05-16 14:39:40

I did mean CFS, thanks. I use BFQ as well, but haven't noticed any issues with it that I can directly tie back to systemd. I'm suspicious that it doesn't play well with virtio_blk, though.

sidneyk · 2011-05-16 18:07:47

To falconindy or anyone that may know - If systemd is installed and appears to be working fine and I have used the systemd native .conf files to set things up, can the initscripts and sysvinit packages be completely removed or not? I know that the initscripts-systemd package can be removed as I have already removed that one. I think I'm pretty close to understanding systemd, but I need to make sure. Thanks in advance.

WorMzy · 2011-05-16 18:21:12

Zom wrote:

eDio wrote:
...
If this is the same issue that others had in this thread, the cause is syslog-ng.service. The suggested fix is to use rsyslog instead. The package also installs its own .service-file.

Thanks for this; I've been having a similar problem to Zom lately, although SysRq+REI seems to break out of whatever loop systemd gets itself into. I'll try swapping out the syslog-ng service and see if it helps.

falconindy · 2011-05-16 18:44:48

sidneyk wrote:

To falconindy or anyone that may know - If systemd is installed and appears to be working fine and I have used the systemd native .conf files to set things up, can the initscripts and sysvinit packages be completely removed or not? I know that the initscripts-systemd package can be removed as I have already removed that one. I think I'm pretty close to understanding systemd, but I need to make sure. Thanks in advance.

You're welcome to build your own systemd package that symlinks all the sysvinit stuff to systemctl, i.e.: halt, init, killall5, telinit, poweroff, reboot, and shutdown, and then conflicts with sysvinit. I won't be doing this. Note that you'll also lose the pidof and mountpoint binaries.

Zom · 2011-05-16 19:55:07

WorMzy wrote:

Thanks for this; I've been having a similar problem to Zom lately, although SysRq+REI seems to break out of whatever loop systemd gets itself into. I'll try swapping out the syslog-ng service and see if it helps.

Just a small footnote, I don't have this problem on neither of my machines. I just happened to remember the issue posted in this thread.

WorMzy · 2011-05-17 02:26:57

Zom wrote:

WorMzy wrote:
Thanks for this; I've been having a similar problem to Zom lately, although SysRq+REI seems to break out of whatever loop systemd gets itself into. I'll try swapping out the syslog-ng service and see if it helps.
Just a small footnote, I don't have this problem on neither of my machines. I just happened to remember the issue posted in this thread.

Well, it did the trick in any case. I can shutdown again~

sidneyk · 2011-05-17 02:43:45

falconindy wrote:

sidneyk wrote:
To falconindy or anyone that may know - If systemd is installed and appears to be working fine and I have used the systemd native .conf files to set things up, can the initscripts and sysvinit packages be completely removed or not? I know that the initscripts-systemd package can be removed as I have already removed that one. I think I'm pretty close to understanding systemd, but I need to make sure. Thanks in advance.
You're welcome to build your own systemd package that symlinks all the sysvinit stuff to systemctl, i.e.: halt, init, killall5, telinit, poweroff, reboot, and shutdown, and then conflicts with sysvinit. I won't be doing this. Note that you'll also lose the pidof and mountpoint binaries.

Thanks, that's what I needed to know as I didn't want to unneccessarily break my system. I kind of figured that there may be some dependencies there, at least with various init scripts and I did see the the file list for sysvinit indicating some files used for mounts or mounting, etc. I understand that systemd is compatible with sysvinit, I was just curious if there was yet a possibility of exploring a 'pure' systemd setup. I also fired up my Fedora 15 Beta Live CD to compare it's systemd setup to what we currently have here. It looks very similar with maybe a few customized targets added for specific purposes. Thanks again.

Last edited by sidneyk (2011-05-17 14:10:54)

sidneyk · 2011-05-18 08:51:34

WorMzy wrote:

How is everybody running httpd with systemd? Have you written your own .service file?
All I have is a http-daemon.target file, which is apparently part of systemd.special, although I don't understand why, or how I'm supposed to use it. At the moment I assume systemd is running httpd due to it being included in the daemons array in rc.conf, but I'm aiming to cut rc.conf out of the picture completely and have a pure systemd setup.
It shouldn't take long to write up a httpd.service, but I'm curious as to why there isn't already one in the systemd or systemd-arch-units packages.

I can't really answer your question about the httpd situation as I see the same thing as you, but I don't believe I'm running it anyway. That is my understanding, so far, that if there is not a pre-written *.service file available, or you don't write one yourself, then rc.conf will still be scanned for whatever is in the DAEMONS array and systemd will load any that it can't find a *.service file for. I did an experiment today though as I was also wondering if the rc.conf could be obsoleted by systemd. I set up an Arch install in Virtual Box and got my normal Gnome setup going minus Gnome-Shell, of course because of the 3d, and then I installed ONLY systemd and systemd-arch-units and enabled the services and targets I needed and all was working fine and booting up even faster in the virtual machine than on the real machine. I then setup the systemd native config files, as per the wiki, rebooted and all was still fine. I then tried to comment out every line that was uncommented in /etc/rc.conf and rebooted. I booted up to my gdm login screen but had no keyboard or mouse control. I then rebooted in fallback mode and uncommented the line about auto loading modules and rebooted and had keyboard and mouse control back. I think, even with a few systemd conf files to setup, all combined still amounts to fewer lines than rc.conf, not to mention all the other scripts that are probably superceded by systemd. It seems pretty simple to me, once I understood it better. Of course, I had *.service files for anything I was starting in the DAEMONS array, so in my case, I was able to get rid of everything in rc.conf except for autoloading modules. I just went through the same process on the real machine with the same results. I was considering as you can see in another post trying to see if sysvinit could be eliminated entirely, but systemd has a dependency on it still at the moment, at least the Arch iteration. I think I now have about as 'pure' a systemd setup as I can get, but I'm going to keep drilling through the bootup stuff to see what I might figure out, though it may be above my current skillset.

Last edited by sidneyk (2011-05-18 08:54:10)

Zom · 2011-05-18 09:45:45

That's pretty much wrong. AFAIK, systemd just uses some variables from rc.conf, things like hostname, while the DAEMONs part is completely skipped. It doesn't automatically use the legacy rc.d-files if no corresponding .service-file is found and it doesn't automatically load .service files from the DAEMONs section in rc.conf.

litemotiv · 2011-05-18 10:41:40

Zom wrote:

That's pretty much wrong. AFAIK, systemd just uses some variables from rc.conf, things like hostname, while the DAEMONs part is completely skipped. It doesn't automatically use the legacy rc.d-files if no corresponding .service-file is found and it doesn't automatically load .service files from the DAEMONs section in rc.conf.

Actually, that's exacty what initscripts-systemd does, see https://wiki.archlinux.org/index.php/Sy … md_package

Arch Linux

#576 2011-05-10 21:58:50

Re: systemd: Yet Another Init Replacement

#577 2011-05-10 22:03:14

Re: systemd: Yet Another Init Replacement

#578 2011-05-10 22:12:24

Re: systemd: Yet Another Init Replacement

#579 2011-05-11 07:54:27

Re: systemd: Yet Another Init Replacement

#580 2011-05-11 14:08:15

Re: systemd: Yet Another Init Replacement

#581 2011-05-11 17:05:42

Re: systemd: Yet Another Init Replacement

#582 2011-05-11 17:34:56

Re: systemd: Yet Another Init Replacement

#583 2011-05-12 14:59:28

Re: systemd: Yet Another Init Replacement

#584 2011-05-15 07:58:14

Re: systemd: Yet Another Init Replacement

#585 2011-05-15 10:52:34

Re: systemd: Yet Another Init Replacement

#586 2011-05-15 12:56:45

Re: systemd: Yet Another Init Replacement

#587 2011-05-15 13:15:16

Re: systemd: Yet Another Init Replacement

#588 2011-05-15 13:20:55

Re: systemd: Yet Another Init Replacement

#589 2011-05-16 13:04:52

Re: systemd: Yet Another Init Replacement

#590 2011-05-16 13:07:38

Re: systemd: Yet Another Init Replacement

#591 2011-05-16 14:39:40

Re: systemd: Yet Another Init Replacement

#592 2011-05-16 18:07:47

Re: systemd: Yet Another Init Replacement

#593 2011-05-16 18:21:12

Re: systemd: Yet Another Init Replacement

#594 2011-05-16 18:44:48

Re: systemd: Yet Another Init Replacement

#595 2011-05-16 19:55:07

Re: systemd: Yet Another Init Replacement

#596 2011-05-17 02:26:57

Re: systemd: Yet Another Init Replacement

#597 2011-05-17 02:43:45

Re: systemd: Yet Another Init Replacement

#598 2011-05-18 08:51:34

Re: systemd: Yet Another Init Replacement

#599 2011-05-18 09:45:45

Re: systemd: Yet Another Init Replacement

#600 2011-05-18 10:41:40

Re: systemd: Yet Another Init Replacement

Board footer