Comment # 14 on bug 1202589 from Neil Brown

I think the problem relates to md2.  It shouldn't get restarted as seems to be
happening.

Can you (temporarily) remove md2 from md4 and see if that avoids the problem?

Is it possible to enable some udev tracing during the shutdown to see if
something is happening there.

I don't think the problem is with dracut.
systemd runs systemd-shutdown and tries to shut down devices until it cannot
make any more progress, and only then does it switch to dracut in the initrd.
As it does seem to make progress on every loop - always stopping md2 - it just
keeps going.