Skip to contentSkip to conversation

systemd timeouts started daemons

DISCUSSION Jukka Rissanen 3 participants 6 messages Mar 12, 2013 – Mar 21, 2013

lore
Share & tools
lore
#0 lore raw

Hi,

I have a weird problem with systemd. It starts a daemon then kills it after a 30 second timeout.

Mar 12 08:50:19 eca connmand[970]: Connection Manager version 1.12
...
Mar 12 08:51:49 eca systemd[1]: connman.service operation timed out. Terminating.
Mar 12 08:51:49 eca connmand[970]: Terminating

I am seeing this behaviour at least with connman, bluez and ofono.
I can start the daemons from command line (as a root) and they work just fine. I can connect to connman via dbus and it can properly connect to ethernet etc.

When connman tries to activate bluez, following is printed

Mar 12 08:50:19 eca dbus[941]: [system] Activating service name='org.bluez' (using servicehelper)
Mar 12 08:50:19 eca dbus[941]: [system] Activated service 'org.bluez' failed: Launch helper exited with unknown return code 1

Usually also connman-vpnd cannot be started, but in this run it was activated successfully

Mar 12 08:50:19 eca dbus[941]: [system] Activating service name='net.connman.vpn' (using servicehelper)
Mar 12 08:50:19 eca dbus[941]: [system] Successfully activated service 'net.connman.vpn'

So it looks there is something fishy with dbus.

I am pretty sure this was working something like a month ago, but cannot really point out what version I had at that time.

Currently I am using latest yocto + oe from git.

Any pointers / ideas what is wrong here?

Cheers,
Jukka

#1 lore raw

Hi all,

some update to the problem I am seeing with systemd.

So the symptom of the problem is that after systemd has started a daemon like connmand, it kills it after 30 secs.

The reason for this was that systemd did not receive NameOwnerChanged dbus signal telling that the daemon managed to register its name to the system bus.

I checked with dbus-monitor that these NameOwnerChanged signals are indeed sent by dbus-daemon but they are not seen in debug prints from systemd.

The systemd version I have is 197, dbus version is 1.6.8. I tried with kernel 3.4 and 3.8, the same result. The distro I am using is very minimal one without X11 or any other graphics, so it might be that I have accidentally removed some vital component. The distro is using packagegroup-base + some additional packages, polkit is not installed.

What is even weirder that in qemu image, I can start connman successfully via systemd but not ofono or bluez. In Intel NUC device, I am not able to start either connman, bluez or ofono from systemd. All these daemons work just fine when started manually from command line.

This paste contains qemu log where connman starts successfully but ofono and bluez do not. http://pastebin.com/UTUVBjqu

I am using latest yocto + oe from git.

Any ideas what is wrong here?

Cheers,
Jukka

#2 lore raw
Quoted 6 lines · Jukka Rissanen

The reason for this was that systemd did not receive NameOwnerChanged dbus signal telling that the daemon managed to register its name to the system bus.

I checked with dbus-monitor that these NameOwnerChanged signals are indeed sent by dbus-daemon but they are not seen in debug prints from systemd.

Andreas sent a patch earlier that masks the dbus init script, because the names don't match so you'll end up with two system buses attempting to run. I'm not sure how this will fail, but it could mean that systemd isn't listening to the bus that connman is starting on, or something.

The easy hack to test this is to delete /etc/init.d/dbus-1 and the symlinks to it in /etc/rc*.

Ross

#3 lore raw

Hi Ross,

Quoted 8 lines · Ross
Quoted 6 lines · Jukka Rissanen

The reason for this was that systemd did not receive NameOwnerChanged dbus signal telling that the daemon managed to register its name to the system bus.

I checked with dbus-monitor that these NameOwnerChanged signals are indeed sent by dbus-daemon but they are not seen in debug prints from systemd.

Quoted 9 lines

Andreas sent a patch earlier that masks the dbus init script, because the names don't match so you'll end up with two system buses attempting to run. I'm not sure how this will fail, but it could mean that systemd isn't listening to the bus that connman is starting on, or something.

The easy hack to test this is to delete /etc/init.d/dbus-1 and the symlinks to it in /etc/rc*.

Removed the dbus from init.d and rc* and everything started to work. This was really good news, thanks a lot for your help!

Cheers,
Jukka

#4 lore raw
In reply to Burton, Ross

Does this resolve also update-rc.d postinst trying to start /etc/init.d/dbus-1 and failing to find it?

Quoted 26 lines · Ross
Quoted 1 line · Jukka Rissanen

The reason for this was that systemd did not receive NameOwnerChanged

dbus

Quoted 4 lines

signal telling that the daemon managed to register its name to the system bus.

I checked with dbus-monitor that these NameOwnerChanged signals are

indeed

Quoted 1 line

sent by dbus-daemon but they are not seen in debug prints from systemd.

Andreas sent a patch earlier that masks the dbus init script, because the names don't match so you'll end up with two system buses attempting to run. I'm not sure how this will fail, but it could mean that systemd isn't listening to the bus that connman is starting on, or something.

The easy hack to test this is to delete /etc/init.d/dbus-1 and the symlinks to it in /etc/rc*.

Ross
_______________________________________________
yocto mailing list
[email protected]
https://lists.yoctoproject.org/listinfo/yocto

#5 lore raw
Quoted 2 lines · Martin Jansa

Does this resolve also update-rc.d postinst trying to start /etc/init.d/dbus-1 and failing to find it?

That's something else to be fixed.

Ross