Re: Rewrite md raid1 member

[NeilBrown <nfbrown@novell.com>]

[-- Attachment #1: Type: text/plain, Size: 3011 bytes --]

On Sat, Aug 20 2016, Chris Dunlop wrote:

> Hi Neil,
>
> Nice work on the Bus1 article!

Thanks :-)

>
> On Sat, Aug 20, 2016 at 07:26:27AM +1000, NeilBrown wrote:
>> On Fri, Aug 19 2016, Chris Dunlop wrote:
>>> In my case, I want it to write everything.
>>>
>>> If I do my 'dd' to write everything as previously described, with the window
>>> of opportunity for stale data to end up on the written disk, one option
>>> would to run a scrub / repair to check the data is the same - but if I'm
>>> unlucky with my dd and the data isn't the same for some sector[s], I want to
>>> ensure the correct data is copied over the stale data and not the other way
>>> around, e.g. to specify "in the event of a mismatch, use the data from sda
>>> and overwrite the data on sdb".
>>>
>>> Unfortunately I don't know how that can be done.
>>>
>>> Does anyone know?
>> 
>> If it is the second device in the array (as listed by mdadm --detail)
>> then you can stop the array and re-assemble with --update=resync.
>
> That's nearly there - except in this specific case it's my root filesystem
> so I can't stop the array without booting into a recovery disk etc. Of
> course I could do that, but the point of the exercise is to see if it can
> be done live, safely.

Well... you could
  cd /sys/block/mdXX/md
  echo frozen > sync_action
  echo 0 > resync_start
  echo idle > sync_action

that should start a resync on a live array.
Still, only works for non-first device in RAID1

>
>> If it is the first device I can only suggest that you
>> fail the device and add it again:
>> 
>>  mdadm /dev/mdXX --fail /dev/sdYY
>>  mdadm /dev/mdXX --remove /dev/sdYY
>>  mdadm /dev/mdYY --add /dev/sdYY
>> 
>> If the "good" drive fails during the rewrite it might be a little bit
>> fiddley getting the array working again, but all the data will certainly
>> be there on the device you are re-writing, so you won't lose anything.
>
> OK, that sounds good. What would the process be if the good drive fails,
> either completely, or a few specific sectors?

If you think there is a serious risk of that happening, then it's best
to skip this option.
You would need to boot from a rescue disk and re-create the array using
just the working device - and make sure the same data-offset and size
are used.  Certainly possible, but not at all straightforward.

Another thing you could do, particularly if you know what region of the
device needs to be over-written, is to write sector numbers to
suspend_lo and suspend_hi.  This will suspend all IO through the
/dev/mdXX device to that range of array sectors.
Then you could read from/write to the raw device with dd or whatever.

raid6check.c does this on a raid6 to correct errors that can be detected
with the raid6 syndrome, even while the array is online.  A similar
thing could be done to allow individual blocks to be rewritten.
Care is needed to map between array addresses and device addresses.

NeilBrown



[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 818 bytes --]