Comment # 6 on bug 1027846 from Neil Brown

sorry for the delay - various distractions....

> "/dev/md0: /dev/md0: device 0 in 2 device undetected raid1 /dev/md3."

This means that /dev/md0 appear to be the first leg of a two-device RAID1 which
is not currently active (i.e. is "undetected"), but was at one time known as
"/dev/md3". 

> OK, let's formulate it differently: Why doesn't activation of a RAID "undo any incomplete activations"? Or maybe why not defaulting to a non-recursive RAID activation?

sorry, but I don't know what you are asking.
If a give RAID is activated, it obviously isn't "incompletely activated" any
more.  The activation one RAID might cause some other RAID to be activated if
the first RAID appears to be one member of the second RAID.  Otherwise it will
have no other effect on any other RAID.

> If you assemble a RAID6 without "--assume-clean", isn't the array zeroed, recalculating all the parity information?

The array isn't zeroed.  The data block are left untouched. The two parity
blocks per stripe are checked and, if necessary, corrected.