Bug#818344: mdadm: device rescan missing in initrd root device loop

[Martin Godisch]

Package: mdadm
Version: 3.3.2-5+deb8u1
Severity: important

I'm running a bunch of Dell PowerEdge FC430 here which contain each two
SSD drives as raid 1. Since there is no hardware raid controller option,
an mdadm raid 1 is configured for a lvm physical volume, the server's
root device is a logical volume in that volume group.

Here is the problem: In about one of two times the server boots into the
initrd prompt since it does not find its root device, probably because
the SSDs are slower initialized at kernel bootup. local_device_setup()
in /usr/share/initramfs-tools/scripts/local takes care of that and tries
again if the root device hasn't shown up yet, but the process doesn't
involve rescanning for raid devices as it probably should in
/usr/share/initramfs-tools/scripts/local-block like
/usr/share/initramfs-tools/scripts/local-block/lvm2. Therefore the raid
device stays absent and rescanning for logical volumes fails again and
again.

Workaround: Copying /usr/share/initramfs-tools/scripts/local-top/mdadm
to /etc/initramfs-tools/scripts/local-block/mdadm solves the problem,
but I'm not sure whether that script is written with repeated execution
in mind.

I tagged this bug important because it makes server reboots
unpredictable. The reboot may be successfull or it may run into the
initrd prompt which would be bad for administrators who try to reboot
without console access.

Please let me know if you need further information. I can test proposed
solutions if necessary.

Kind regards,

Martin