Comment # 15 on bug 1058028 from Shivasharan Srikanteshwara

(In reply to Thomas Blume from comment #14)
> (In reply to Shivasharan Srikanteshwara from comment #13)
> > I have attached the requested logs, but there is nothing related to
> > udev-settle service in the logs.
> > Do I need to enable the service explicitly after OS install?
> > How does this service get triggered? 
> > I can see multipathd.service and dmraid-activation.service have lines that
> > refer to systemd-udev-settle.service. But both multipathd and dmraid are not
> > enabled.
> > 
> > linux-so07:/usr/lib/systemd # grep -R "udev-settle" *
> > system/multipathd.service:Wants=systemd-udev-trigger.service
> > systemd-udev-settle.service
> > system/multipathd.service:After=multipathd.socket
> > systemd-udev-trigger.service systemd-udev-settle.service
> > system/dmraid-activation.service:After=systemd-udev-settle.service
> > system/dmraid-activation.service:Wants=systemd-udev-settle.service
> > 
> > Thanks.
> 
> Indeed systemd-udev-settle.service is only triggered at certain conditions.
> That doesn't look correct, it should be triggered by
> systemd-udev-trigger.service.
> Could you please do the following changes in systemd-udev-trigger.service:
> 
> -->
> --- /usr/lib/systemd/system/systemd-udev-trigger.service.orig	2017-09-29
> 10:10:08.423500663 +0200
> +++ /usr/lib/systemd/system/systemd-udev-trigger.service	2017-09-29
> 10:10:24.143500663 +0200
> @@ -9,7 +9,7 @@
>  Description=udev Coldplug all Devices
>  Documentation=man:udev(7) man:systemd-udevd.service(8)
>  DefaultDependencies=no
> -Wants=systemd-udevd.service
> +Wants=systemd-udevd.service systemd-udev-settle.service
>  After=systemd-udevd-kernel.socket systemd-udevd-control.socket
> systemd-hwdb-update.service
>  Before=sysinit.target
>  ConditionPathIsReadWrite=/sys
> --<
> 
> afterwards run mkinitrd and check whether the issue still appears after
> reboot?

This does alleviate my problem a bit but does not solve completely.
The udev-settle service has a default timeout of 120secs.
So with this the tolerance increases from 30secs to about 150seconds (120 secs
for udev-settle to timeout + earlier 30secs for systemd-udevd to kill all child
processes).
Let me know if you require the logs with these changes.

>From the various experiments I have been running, if "KillMode=mixed" option in
systemd-udevd.service file is removed the problem goes away completely... Even
without the udev-settle changes.
With KillMode=mixed, systemd sends a SIGKILL after 30secs timeout.
But the default behaviour is to send a SIGTERM after 30seconds.
Looks like the SIGKILL is causing all the trouble here.
Is the KillMode option required for systemd-udevd?