Q: cat /proc/mdstat blocking?

Q: cat /proc/mdstat blocking?

[Ulrich Windl]

Hi,

I thought that /proc/mdstat just displays the current state of the MD RAIDs, not querying them, thus never blocking. Unfortunately I could observer a "cat /proc/mdstat" hanging for several seconds when one of the RAID1 legs was unavailable and multipathd was queueing requests to the leg.

Practically this makes monitoring of the other RAIDs difficult when one bad device is enough to block /proc/mdstat.

Now my question: Does it work as designed (i.e. as intended), or is this not as intended?

As I'm not subcribed to the kernel list, maybe CC: your replies.

Effect was observed in SLES11 SP1 x86_64 with multipathed FC SAN disks (2.6.32.36) on a machine with several CPUs.

Regards,
Ulrich Windl

Re: Q: cat /proc/mdstat blocking?

[Al Viro]

On Fri, Aug 05, 2011 at 02:02:49PM +0200, Ulrich Windl wrote:
> Hi,
> 
> I thought that /proc/mdstat just displays the current state of the MD RAIDs, not querying them, thus never blocking. Unfortunately I could observer a "cat /proc/mdstat" hanging for several seconds when one of the RAID1 legs was unavailable and multipathd was queueing requests to the leg.
> 
> Practically this makes monitoring of the other RAIDs difficult when one bad device is enough to block /proc/mdstat.
> 
> Now my question: Does it work as designed (i.e. as intended), or is this not as intended?

I suspect that something is holding all_mddevs_lock too long; that, or one
of the subdrivers' spinlocks.  Try to find out what it blocks on and what
holds the lock in question...