[systemd-devel] udevadm settle takes too long to finish

[Chris Murphy]

On Dec 7, 2013, at 2:39 PM, Robert Milasan <rmilasan at suse.com> wrote:
> 
> Well that didn't help, but this did:
> 
> systemctl mask lvm2-activation.service
> systemctl mask lvm2-activation-early.service
> 
> If I only mask lvm2-activation.service, then it doesn't make much
> difference, but masking both it works better.
> 
> My question is: why does this happen, if lvm is not in use at all?

Right. And on Fedora 20 I don't have those two lvm2 services, instead I have lvm2-monitor.service being pulled in even though LVM isn't being used at all.

          1.270s lvm2-monitor.service
          1.267s systemd-udev-settle.service

[root at f20s ~]# systemctl status lvm2-monitor.service
lvm2-monitor.service - Monitoring of LVM2 mirrors, snapshots etc. using dmeventd or progress polling
   Loaded: loaded (/usr/lib/systemd/system/lvm2-monitor.service; enabled)
   Active: active (exited) since Sat 2013-12-07 12:25:57 MST; 2h 23min ago
     Docs: man:dmeventd(8)
           man:lvcreate(8)
           man:lvchange(8)
           man:vgchange(8)
  Process: 226 ExecStart=/usr/sbin/lvm vgchange --monitor y (code=exited, status=0/SUCCESS)
 Main PID: 226 (code=exited, status=0/SUCCESS)

Dec 07 12:25:56 f20s.localdomain lvm[226]: /dev/sr0: open failed: No medium found
Dec 07 12:25:57 f20s.localdomain lvm[226]: No volume groups found
Dec 07 12:25:57 f20s.localdomain systemd[1]: Started Monitoring of LVM2 mirrors, snapshots etc. using dmeventd or progress polling.


So is the correct question, why is this service enabled by default? And what enables such things? Is this an installer just doing the easy thing and setting it to be enabled just in case the user chose an LVM install (which is the Fedora 20 default) rather than making it a conditional monitoring? Or is the correct question why is this one command to monitor taking even 1.2 seconds to do what it needs to do? That's a long time.

The other thing is that on baremetal I get 30 second boot times according to blame. However the exact same install parameters in qemu/kvm on that same baremetal machine and I'm getting 4 second boot times. Why is the baremetal sooo much slower in user space? I can understand optimization of the kernel and initrd being loaded but I don't understand the performance of userspace going from 22 seconds to 3 seconds.


Chris Murphy