Re: AW: Raid 5 and Power Failure

On Saturday July 13, martin.bene@icomedias.com wrote:

Hi Neil, Jason

quoted

Get mdadm
  http://www.cse.unsw.edu.au/~neilb/source/mdadm/

and use
  mdadm --assemble --force /dev/md0 /dev/hd......

shouldn't /dev/hda be left out of the list here? since it looks like hda
really is out out date I think you'd get best results by using only the most
up-to-date disks.

hda certainly could be left out, but the point of "mdadm -Af" is that
it tried to do the "best" thing.  You don't *need* to figure out in
too much detail what has gone wrong.  It will pick the best available
drives and use them.

Also, when using the mkraid -force method that had to be used before mdadm, I
always recommended leaving out all spare disks and one of the data disks,
forcing the array to be (re)created in degraded mode. that way, write
operations to the disks are kept to a minimum and it's possible to retry if
you made an error (wrong disk left out etc).

would it make sense to treat mdadm similarely, i.e only provide enough
devices to assemble a degraded array and hotadd the final device after the
array checks out ok? I don't like the idea of background parity rebuild
kicking in before I've had a chance to check things out.

In general I would agree with that.
I would actually like to be able to start an array in "read-only" mode
and have no rebuild/resync happen until it got switched to write
only.  This may well happen in 2.6.

NeilBrown