Comment 13 for bug 1512857

Using the reproduce from #12, I've tested the following sequence:

udevadm settle
udevadm control --stop-exec-queue
<mdadm create>
udevadm control --start-exec-queue
udevadm settle

The above effectively "pauses" udev events on the block device while we create our /dev/mdX device
then we unpause and let any queued events take place.

With that in place, I've successfully installed this configuration 10/10 times in a row. I'm preparing a branch
with this fix including a vmtest which executes this configuration. If we're still racy here then we will see
it there, albeit intermittently.

It's not clear to me yet if mdadm itself needs a bug w.r.t learning to "wait" on the creation of the device node
under /dev or if the race is somewhere else.