On Wed, Jul 5, 2017 at 11:38 PM, Rob Leslie <r...@mars.org> wrote: > Package: udev > Version: 232-25 > Severity: important > File: /lib/systemd/systemd-udevd > > Dear Maintainer, > > After upgrading to stretch I've discovered that some machines are taking a > long time to boot -- more than two minutes -- with no immediately obvious > cause. For example: > > > % systemd-analyze > > Startup finished in 2min 55.491s (kernel) + 13.076s (userspace) = 3min > 8.568s >
This suggests the kernel is the one slowing things down. > > During most of this time, nothing appears on the console except a cursor. > Kernel and system logs show nothing to pinpoint the issue other than long > time gaps, e.g.: > > > [ 3.194336] clocksource: Switched to clocksource tsc > > [ 121.499692] md: linear personality registered for level -1 > > Adding 'debug' to the kernel command line however I got this clue: > > > [ 64.386629] systemd-udevd[88]: seq 905 '/devices/virtual/block/dm-5' > is taking a long time > This is just udev complaining. Udev does not generate the devices, it waits for the kernel to notify it. > > After some further investigation it appears the problem stems from having > an > LVM PV inside another LV. Here is how I am able to recreate the problem on > a > fresh stretch install: > > > # lvcreate -n pvtest -L 1g $vg > > # pvcreate /dev/$vg/pvtest > > (where $vg is some existing volume group name) > > Now rebooting the system will result in the long boot delay. > > This sort of setup did not cause any problems in jessie. An obvious > workaround > is not to put PVs inside other LVs, but this is sometimes a convenient > arrangement and I would hope it could be made to work without an excessive > boot delay. > Can you try booting with the old jessie kernel? This looks like a kernel regression. -- Saludos, Felipe Sateler
