[Qemu-devel] Re: [PATCH 6/6] Continue a migrated vm is a bad idea

[Avi Kivity <avi@redhat.com>]

On 08/20/2009 12:10 PM, Juan Quintela wrote:
> Avi Kivity<avi@redhat.com>  wrote:
>    
>> On 08/19/2009 05:07 AM, Juan Quintela wrote:
>>      
>>> index ea5c33a..762a9a5 100644
>>> --- a/monitor.c
>>> +++ b/monitor.c
>>> @@ -567,6 +567,11 @@ static void do_cont(Monitor *mon)
>>>    {
>>>        struct bdrv_iterate_context context = { mon, 0 };
>>>
>>> +    if (unlikely(vm_has_migrated)) {
>>> +        monitor_printf(mon, "this vm has migrated to other machine\n");
>>> +        monitor_printf(mon, "You have to load other vm\n");
>>> +        return;
>>> +    }
>>>
>>>        
>> You don't know whether the guest is running on the other side.  It may
>> be that the control program started the other side stopped, and after
>> migration completes decides which side to continue running.
>>      
> Only works if migration suceeded and was done in _non_ autostart node.
>    

Well, you're disabling this method.

> Have to debug this very bug.  User thought that he had done _nothing_
> on the destination side, and that then stoping it, and continue in the
> source side was a good idea.
>
> He didn't understand why he had disk corruption, obviosly qemu had a
> bug.  I haven't found a better approach to fix this bug.  If you have a
> better idea of what to do in this case, I am all ears.
>
> The reason that I created the patch is that if the user didn't got it
> right, it gets disk corruption.
>    

There are tons of ways to cause disk corruption running multiple qemus.  
This only solves one of them.

I suggest taking locks on the disk images instead (and adding an option 
not to lock for raw images).

-- 
error compiling committee.c: too many arguments to function