Re: RAID scrubbing

[Neil Brown <neilb@suse.de>]

On Wed, 14 Apr 2010 17:51:11 -0700
Justin Maggard <jmaggard10@gmail.com> wrote:

> On Fri, Apr 9, 2010 at 7:01 PM, Michael Evans <mjevans1983@gmail.com> wrote:
> > On Fri, Apr 9, 2010 at 6:46 PM, Justin Maggard <jmaggard10@gmail.com> wrote:
> >> On Fri, Apr 9, 2010 at 6:41 PM, Michael Evans <mjevans1983@gmail.com> wrote:
> >>> On Fri, Apr 9, 2010 at 6:28 PM, Justin Maggard <jmaggard10@gmail.com> wrote:
> >>>> Hi all,
> >>>>
> >>>> I've got a system using two RAID5 arrays that share some physical
> >>>> devices, combined using LVM.  Oddly, when I "echo repair >
> >>>> /sys/block/md0/md/sync_action", once it finishes, it automatically
> >>>> starts a repair on md1 also, even though I haven't requested it.
> >>>> Also, if I try to stop it using "echo idle >
> >>>> /sys/block/md0/md/sync_action", a repair starts on md1 within a few
> >>>> seconds.  If I stop that md1 repair immediately, sometimes it will
> >>>> respawn and start doing the repair again on md1.  What should I be
> >>>> expecting here?  If I start a repair on one array, is it supposed to
> >>>> automatically go through and do it on all arrays sharing that
> >>>> personality?
> >>>>
> >>>> Thanks!
> >>>> -Justin
> >>>>
> >>>
> >>> Is md1 degraded with an active spare?  It might be delaying resync on
> >>> it until the other devices are idle.
> >>
> >> No, both arrays are redundant.  I'm just trying to do scrubbing
> >> (repair) on md0; no resync is going on anywhere.
> >>
> >> -Justin
> >>
> >
> > First: Reply to all.
> >
> > Second, if you insist that things are not as I suspect:
> >
> > cat /proc/mdstat
> >
> > mdadm -Dvvs
> >
> > mdadm -Evvs
> >
> 
> I insist it's something different. :)  Just ran into it again on
> another system.  Here's the requested output:

Thanks.  Very thorough!


> Apr 14 17:32:23 JMAGGARD kernel: md: requested-resync of RAID array md2
> Apr 14 17:32:23 JMAGGARD kernel: md: minimum _guaranteed_  speed: 1000
> KB/sec/disk.
> Apr 14 17:32:23 JMAGGARD kernel: md: using maximum available idle IO
> bandwidth (but not more than 200000 KB/sec) for requested-resync.
> Apr 14 17:32:23 JMAGGARD kernel: md: using 128k window, over a total
> of 972041296 blocks.
> Apr 14 17:32:51 JMAGGARD kernel: md: md_do_sync() got signal ... exiting
> Apr 14 17:33:35 JMAGGARD kernel: md: requested-resync of RAID array md3

So we see the requested-resync (repair) of md2 started as you requested,
then finished at 17:32:51 when you write 'idle' to 'sync_action'.

Then 44 seconds later a similar repair started on md3.
44 seconds is too long for it to be a direct consequence of the md2 repair
stopping.  Something *must* have written to md3/md/sync_action.   But what?

Maybe you have "mdadm --monitor" running and it notices when repair on one
array finished and has been told to run a script (--program or PROGRAM in
mdadm.conf) which would then start a repair on the next array???

Seems a bit far-fetched, but I'm quite confident that some program must be
writing to md3/md/sync_action while you're not watching.

NeilBrown


--
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