Raid 6 rebuild

Raid 6 rebuild

[Bryan Bush]

If you have 2 failed drives in a raid 6 array, will it rebuild them
both at the same time or in serial?
Thanks
-Bryan

Re: Raid 6 rebuild

[NeilBrown]

[-- Attachment #1: Type: text/plain, Size: 972 bytes --]

On Fri, 6 Apr 2012 03:01:26 -0400 Bryan Bush <bbushvt@gmail.com> wrote:

> If you have 2 failed drives in a raid 6 array, will it rebuild them
> both at the same time or in serial?

Maybe.

Probably not, but it can be coaxed to.

The next release of mdadm will contain

http://neil.brown.name/git?p=mdadm;a=commitdiff;h=9f58469128c99c0d7f434d28657f86789334f253

which will ensure that if you:

  mdadm /dev/md0 --add /dev/foo /dev/bar

then both will be added before the recovery starts, so both will be recovered
at the same time.

Alternately you could:

 echo freeze > /sys/block/md0/md/sync_action
 mdadm /dev/md0 --add /dev/foo /dev/bar
 echo idle > /sys/blcok/md0/md/sync_action

you might even be able to leave of the first step.
The "add" will cause a recovery to start on just one device, but then when
you write 'idle', it will abort  the recovery, automatically restart, but
this time will use all available spares.

NeilBrown


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 828 bytes --]

Raid 6 rebuild

[Gallas, William N]

Hi,

I have a raid 6 unit that takes over 40 hours to rebuild from an HDD fail condition. I have ample CPU cycles available and was wondering
if MDRAID throttles the rebuild speed. 

Regards,
Bill Gallas

Re: Raid 6 rebuild

[Chris Murphy]

On Feb 11, 2014, at 11:11 AM, "Gallas, William N" <william.n.gallas@intel.com> wrote:

> 
> Hi,
> 
> I have a raid 6 unit that takes over 40 hours to rebuild from an HDD fail condition. I have ample CPU cycles available and was wondering
> if MDRAID throttles the rebuild speed. 

Yes. It's probably a combination of too small md/stripe_cache_size, and /proc/sys/dev/raid/speed_limit_max and maybe /proc/sys/dev/raid/speed_limit_min. Another possibility is misalignment due to incorrect partitioning of a disk with 4KB physical sector sizes.


Chris Murphy