Periodically scanning RAID arrays for inconsistencies

Periodically scanning RAID arrays for inconsistencies

[Brassow Jonathan]

Is there a way to periodically scan a RAID array to report any discrepancies?  This would be something like the initial synchronization, but would do nothing but report anomalies.  For example, it is possible for tracks to be altered when adjacent tracks are repeatedly written.  It might happen rarely, but without a periodic scan, you would never know about it.  Leaving aside the question of /what/ to do when an anomaly is found, I haven't seen any code or indication in the man page for 'mdadm' that would suggest this is currently possible.  Am I wrong?  Is there a way to do this?

 brassow

Re: Periodically scanning RAID arrays for inconsistencies

[Tregaron Bayly]

On Fri, 2013-03-01 at 15:50 -0600, Brassow Jonathan wrote:
> Is there a way to periodically scan a RAID array to report any discrepancies?

Yes, it's possible - Redhat distributes a script called raid-check that
is a good example of this very thing.

Documentation/md.txt talks about 'check' and 'repair' as valid words for
sync_action in sysfs.  The number of errors found will be reported to
mismatch_cnt.  These can be found at /sys/block/md#/md for your array.

hth,

Tregaron

Re: Periodically scanning RAID arrays for inconsistencies

[Phil Turmel]

On 03/01/2013 05:10 PM, Tregaron Bayly wrote:
> On Fri, 2013-03-01 at 15:50 -0600, Brassow Jonathan wrote:
>> Is there a way to periodically scan a RAID array to report any discrepancies?
> 
> Yes, it's possible - Redhat distributes a script called raid-check that
> is a good example of this very thing.
> 
> Documentation/md.txt talks about 'check' and 'repair' as valid words for
> sync_action in sysfs.  The number of errors found will be reported to
> mismatch_cnt.  These can be found at /sys/block/md#/md for your array.

In general, this is called "scrubbing" your array, and it is *vital* to
do this on a regular basis.  Weekly is good.  You should only use
"check" to find mismatches (and it will fix UREs).  If any mismatches
are found, use "repair" to synchronize the devices, then fsck to fix any
resulting errors.

There's lots of history on this in the archives...

Phil

Re: Periodically scanning RAID arrays for inconsistencies

[NeilBrown]

[-- Attachment #1: Type: text/plain, Size: 943 bytes --]

On Fri, 1 Mar 2013 15:50:05 -0600 Brassow Jonathan <jbrassow@redhat.com>
wrote:

> Is there a way to periodically scan a RAID array to report any discrepancies?  This would be something like the initial synchronization, but would do nothing but report anomalies.  For example, it is possible for tracks to be altered when adjacent tracks are repeatedly written.  It might happen rarely, but without a periodic scan, you would never know about it.  Leaving aside the question of /what/ to do when an anomaly is found, I haven't seen any code or indication in the man page for 'mdadm' that would suggest this is currently possible.  Am I wrong?  Is there a way to do this?
> 
>  brassow

echo check > /sys/block/mdX/md/sync_action

look for results in /sys/block/mdX/md/mismatch_cnt

See "SCRUBBING AND MISMATCHES" in "man 4 md".

Yes it would be nice if mdadm provided a nicer interface to this.  Patches
welcome:-)

NeilBrown

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