Re: mismatch_cnt and Raid6

[Andrew Falgout <andrew.falgout@gmail.com>]

It takes about 23 hours per check if I don't do anything to the array, 
so it will be a while before I can get back to you with results.  But 
thanks for the quick response.

./Andrew

On 4/21/2011 8:14 AM, NeilBrown wrote:
> On Thu, 21 Apr 2011 08:00:39 -0500 Andrew Falgout<andrew.falgout@gmail.com>
> wrote:
>
>> I got an error last week from a new raid6 array about a mismatch_cnt.  I
>> did some reading online, performed a repair action on the array,
>> performed a check action, and checked for the mismatch_cnt again.  The
>> number was greatly reduced, but it was still there.  According to mdadm,
>> everything appears to be working fine.  All the drives are passing short
>> tests on smartctl.
>>
>> What is mismatch_cnt really?  Should I even be concerned about this?
> Yes, you should be concerned.
> mismatch_cnt is a count of sectors where the parity blocks don't match the
> data blocks.
>
> The code doesn't check every sector individually.  For raid5/6 it checks 4K
> at a time, so divide by 8, and that many 4K blocks are in doubt.
>
> So something if going wrong somewhere.
>
> I would run 'check' a few time and see if the number changes.
> If it goes down at all, then it looks like you occasionally get bad reads
> from a device.
> If it only ever increases, then you are presumably getting bad writes
> sometimes.
>
> You could:
>   - stop the array
>   - run sha1sum on each member disk, several times.
>   - if any one disk has an unstable result - check cabling, or replace the disk
>   - if more than one disk has an unstable result, replace the controller maybe.
>   - if all results are stable it must be a write-only problem - much harder
>     to work with.
>
> NeilBrown
>
>
>
>> The array is giving me 25-30MB/sec performance on an sshfs mount over
>> the network.  With a local copy I can see speeds of 50 to 60MB/sec.
>>
>> Thanks,
>> Andrew Falgout