Re: Raid reconstructing and process hanging in D state

[david]

On Fri, 16 Oct 2009, Holger Kiehl wrote:

> On Fri, 16 Oct 2009, Justin Piszcz wrote:
> > Hi,
> >
> > I have the same problem with mdadm/raid-1, if you do not limit the speed in 
> > the speed_limit_min paramter, it will starve the I/O from all other 
> > processes and result in the same problem you are having.
> But in my case speed_limit_min was set to 1000 and speed_limit_max was 200000
> (ie. the default) and this still caused all process to hang in D-state. Only
> lowering speed_limit_max helped to make the system responsive again.

yep, MD raid will use every bit of disk bandwidth there is.

even worse, the speed_limit_* appears to be per-drive, not per array 
(although the /proc/mdstat info is per array), so if you have lots of 
drives on one controller card it will kill the system and still think that 
it's not maxed out.

I did a 45 disk array hooked into two PCI slots and found that I needed to 
set speed_limit_max _very_ low to avoid this sort of problem.

David Lang

> > But for me I'm not sure about the reconstruction-- this happens for me 
> > during a raid verify/check.
> This was also the case for me, during a raid verify/check.
>
> Holger
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/