Re: Disk I/O error while rebuilding an md raid-5 array

Thanks for the good advice.  ddrescue on sdb returned an error of 4kB.
 I do still have my old sde.  But one thing I did, which was stupid,
was trying to rebuild the raid-5 when it was mounted.  So I don't know
the old sde is still consistent with the rest 3 disks, since some
files would have been modified between the times when I took the old
sde offline and when the rebuild failed.

So at this point, I guess I'll get 4 new drives and set up a brand new
raid-6 and try to restore my data from my backup in an external drive
and hope for the best.  I'll keep the 4 drives from my old raid-5 just
in case if I need to recover something from them.

I guess I learned my lesson.  I should have ddrescued all the disks I
want to replace, instead of using md's rebuild mechanism.

DS


On Mon, Feb 8, 2010 at 10:57 PM, Stefan Hübner
[off-list ref] wrote:

Hi!

I do RAID-recoveries at least once a month and get paid for it.  Rule of
thumb: if your have one drive dropped and another one with pending
sectors, your rebuild will fail - no need for calculations there.

ddrescue on a clean disk is about half as fast as dd with a blocksize
beyond 1M.  ddrescue on a disk with pending sectors is just no pita as
dd or sg_dd would be, because it adds the neccesary intelligence.

Do you have the original sde still around?  If yes, ddrescue both: sdb
and sde.  My experience says: there will only be a few KB lost.  Then
re-create your raid (it will only write the superblock new) with
"--assume-clean".  After that worked, you might make another (big)
backup first, then run fsck and see what happens.  If the lost bytes
have screwed the filesystem, you might want to re-create the raid with
another (personally I prefer xfs) fs and replay your backup into it.

A few commands to make the intentions cleaner:
ddrescue -dv -r5 /dev/oldsdb1 /dev/newsdb1 /root/sdblog
ddrescue -dv -r5 /dev/oldsde1 /dev/newsde1 /root/sdelog
... find out which drive is which raid-device -> mdadm -E /dev/sdX1
mdadm --create /dev/md0 --raid-devices=4 --level=5
--chunk=${your_chunk_size_in_kb} --assume-clean
${ordered_list_of_raid_devices}

Hope this helps,
Stefan Hübner


Am 09.02.2010 05:20, schrieb Dawning Sky:

quoted

On Mon, Feb 8, 2010 at 3:23 PM, Dawning Sky [off-list ref] wrote:

quoted

Hi,

Now I have two faulty drives and things don't look good.  However, I
was able to add the sdb back to the array and md seemed not mind and
still reported "active sync".  At this point I shut down computer and
decided to clone sdb with clonezilla so that I can have a good sdb to
finish rebuilding sde.  Not sure if it will complete without I/O
errors.  It appears clonezilla is using dd and the speed is extremely
slow (~5MB/sec) and it says it's gonna take 1 day to clone the 500GB.

As expected, dd encountered the same UNC error.  Now I'm trying to
ddrescue the drive to see what happens.  My question is whether this
is worth doing.  Assuming ddrescue cannot read the bad sector either
and writes 0's to the new drive, will I be able to rebuild the raid-5,
from 2 good disks and this disk with a bad sector?  I can assume there
will be a bad file but will the array still function?

Or I'm better off just build a new array from scratch.

Any suggestions are appreciated.

Regards,

DS
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html