Error checking on migration?

Error checking on migration?

[John Byrne]

I've noticed that migration doesn't fail if the VM cannot access the 
disk image on the new host. (xen-unstable 10273)  The VM ends up on the 
new host and hangs when it does a disk access. This is a somewhat evil 
behavior.

Does anyone have a fix for this in-hand, somewhere? If not, any advice 
where to look to fix the problem?

Thanks,

John Byrne

Re: Error checking on migration?

[Anthony Liguori]

John Byrne wrote:
>
> I've noticed that migration doesn't fail if the VM cannot access the 
> disk image on the new host. (xen-unstable 10273)  The VM ends up on 
> the new host and hangs when it does a disk access. This is a somewhat 
> evil behavior.

You could add some basic sanity checking in XendRestore.py:restore() 
before xd.restore_() is invoked.  A consequence of doing device setup 
through scripts is that the only proper way to know if a device creation 
will succeed is by actually creating it.  It's further complicated by 
the race condition that results by doing the checking before actually 
grabbing the devices (what if another migration is in progress that 
happens to use the same devices).

At any rate, having some sort of error checking is probably better than 
having none at all.

Regards,

Anthony Liguori

>
> Does anyone have a fix for this in-hand, somewhere? If not, any advice 
> where to look to fix the problem?
>
> Thanks,
>
> John Byrne
>
>
>
>
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel

Re: Error checking on migration?

[Ewan Mellor]

On Mon, Jun 05, 2006 at 09:50:06PM -0500, Anthony Liguori wrote:

> John Byrne wrote:
> >
> >I've noticed that migration doesn't fail if the VM cannot access the 
> >disk image on the new host. (xen-unstable 10273)  The VM ends up on 
> >the new host and hangs when it does a disk access. This is a somewhat 
> >evil behavior.
> 
> You could add some basic sanity checking in XendRestore.py:restore() 
> before xd.restore_() is invoked.  A consequence of doing device setup 
> through scripts is that the only proper way to know if a device creation 
> will succeed is by actually creating it.  It's further complicated by 
> the race condition that results by doing the checking before actually 
> grabbing the devices (what if another migration is in progress that 
> happens to use the same devices).

Also, migration to localhost is very useful for testing, so it's also 
complicated by the fact that the outgoing domain is using the devices
when the process starts, so you have to be careful to do the checks with
that in mind.

Ewan.