[libvirt-users] release open_disk error

[Frido Roose]

On Tuesday 8 May 2012 at 18:56, Frido Roose wrote:  
> Hello,  
>  
> I wonder what the "open error -1" / "release open_disk error" messages in sanlock.log actually mean.
>  
> I saw these messages in the log on a KVM host that rebooted, and after running "/usr/sbin/virt-sanlock-cleanup" on that host.
> The resources where disks from 2 guests running on another KVM host.
>  
> So in fact the disks are still in use, bot got cleaned up by the other KVM host because it thought they were not longer in use...
>  
> Several questions:
> 1. Could this indicate a problem with the lease of these resources on the server that host these guests?
>  

Let's try to answer myself from what I've learned so far...

Doing a "sanlock client status" revealed that in fact virt-sanlock-cleanup had removed the resource, so I suspect the device were no longer 'locked' at that point.
The reason for this is still unclear to me, but I think I found a way to re-register the resources.

> 2. Is there a way to reregister these resources, because they are still in use (but not for sanlock it seems)?

I made a script to reregister the resources by looking up the devices that were assigned to the VM and by running some sanlock commands.  The crucial part in this is:

dd if=/dev/zero of=/var/lib/libvirt/sanlock/$MD5 bs=1M count=1
chmod 600 /var/lib/libvirt/sanlock/$MD5
sanlock direct init -r __LIBVIRT__DISKS__:$MD5:/var/lib/libvirt/sanlock/$MD5:0
sanlock client acquire -r __LIBVIRT__DISKS__:$MD5:/var/lib/libvirt/sanlock/$MD5:0 -p $pid



The $pid of the VM was still registered with the sanlock daemon, so running these commands was sufficient.
  
> 3. Overall, is it a bad idea to run virt-sanlock-cleanup?


Running virt-sanlock-cleanup now leaves the resources assigned, so I don't know why the were cleaned up.  Perhaps sanlock was not completely ready with initialization after the reboot… I'm not sure.  I guess time will figure out.

Best regards,
Frido