[dm-devel] A question about dm-raid1 on CentOS5.8

[Brassow Jonathan]

On May 14, 2013, at 2:36 AM, Striver/sj wrote:

>   Hi guys
> I create a mirror device on CentOS5.8, the table is 
> "0 8372224 mirror core 3 4096 nosync block_on_error 2 8:16 8192 8:32 8192"
> But I found when I offline the first leg of the mirror device. All write IO is hung up.
> But it will not occur on CentOS5.3/5.4, in my opinion, if one leg returns IO error, 
> IO will continue if there is still other good leg.
> So does this should be deal with by user space (dmeventd)? Or I make something error. 
> 

Yes, it should be handled by dmeventd.  (dmeventd will notice the event raised by the kernel and attempt a 'lvconvert --repair vg/lv' or 'vgreduce --removemissing vg if old enough.)

You should find that a 2-way mirror is converted to a linear device when one of the legs fails.

 brassow