Re: Bringing broken raid 5 back on-line.

I didn't see any other answers, so I'll take a shot...

On Mon, Feb 02, 2004 at 03:44:57PM -0800, Louis Erickson wrote:
[snip]

We've got a raid 5 with three drives.  I'll spare you the comedy of errors
and/or sob story that you've all probably heard before, but the upshot is
that we have two failed drives and no current backup.

When the second drive failed, the raid went offline right away.  The
second failure was apparently due to a write fail on the disk.  I presume
that the data on the second failed drive and the main drive is mostly in
synch, except for the last change made.  My conclusion is that other than
the file being written, the rest of the data should be okay.  Please tell
me if I'm wrong there.

That's a reasonable guess, I think.  It depends to some extent what
went bad... once I had a drive that developed a few bad sectors, but
the bad sectors were in the middle of the inodes for my /home
partition, so it did some pretty serious damage.

We got identical drives to the one good drive and the second
failure, and have used dd to copy the data on to those volumes, so
we can attempt to recover and not hurt our only copies of this data.
I've seen another message on this list which suggests this should
work.  Copies were made with no errors.

At the moment, my system is a little old and does not have mdadm on it.
Is it worth getting and installing this, or am I best to stick with the
original raidtools?

I advise you to get and install mdadm -- it has worked fine for me and
mdadm seems to be the overwhelming preference of the posters to this
list and thus easier to get support for.

The question of the hour is: How do I get the two drives to come back
on-line in degraded mode?  They have different event counters and will not
come up.

Any advice or direction you can provide will be very much appreciated.

Once you have mdadm, 

mdadm --assemble /dev/md<whatever> --force <good drive #1> <good drive
#2> missing

should work for you.  Note that "missing" is a literal string which
you supply instead of a third device name, indicating that you want to
assemble a three-drive array in degraded mode without one of its
drives.  You should see some message in the logs about forcing the
event count of some drive or another.  "--force" is necessary because
the remaining drives' event counts are different.

-Steve