OK,following szenario: /dev/md0 consists of 3 disks lets say hd{a,b,c}1. and /dev/md0 is part of an LVM volume group and is therefore a physical volume. LVM now scans all devices for LVM Metadata. and for some strange reasons (RAID0 here) it finds it on hda1. so it attaches hda1 to a volume group (activating the actual logical volumes fails of cause), but the device is marked as in-use. next mdadm tries to start the raid array,which fails because hda1 is in use. editing /etc/init.d/checkfs to first perform raid initialization and then activate lvm volumes fixed it nicely and the raid array and the lvm volume group on the raid array are both active. I think the proposed order is better as the current one,because raid devices can be physical volumes, but lvm logical volumes can't be part of raid devices (AFAIK).
*** This bug has been marked as a duplicate of 37418 ***