Re: [Qemu-devel] Re: [PATCH 3/3] disk: don't read from disk until the guest starts

[Anthony Liguori <anthony@codemonkey.ws>]

On 09/14/2010 04:47 AM, Avi Kivity wrote:
>  On 09/13/2010 04:34 PM, Anthony Liguori wrote:
>> On 09/13/2010 09:13 AM, Kevin Wolf wrote:
>>>> I think the only real advantage is that we fix NFS migration, right?
>>> That's the one that we know about, yes.
>>>
>>> The rest is not a specific scenario, but a strong feeling that 
>>> having an
>>> image opened twice at the same time feels dangerous.
>>
>> We've never really had clear semantics about live migration and block 
>> driver's life cycles.  At a high level, for live migration to work, 
>> we need the following sequence:
>>
>> 1) src> flush all pending writes to disk
>> 2) <barrier>
>> 3) dst> invalidate any cached data
>> 4) dst> start guest
>
> That's pretty complicated, compared to
>
> 1) src> close

1.5) <barrier>

> 2) dst> open
> 3) dst> start guest

You need to make sure the open happens *after* the close.

You're just using close to flush all pending writes or open to 
invalidate any cached data.

Regards,

Anthony Liguori

> There are two failure scenarios with this model:
>
> 1. dst cannot open the image
>
> We fix that by killing dst and continuing src (which has to re-open 
> its images).
>
> 2. dst cannot open the image, and src cannot as well
>
> In this case, what would be gained by having an image handle open in 
> one of the hosts, but no way to open it again?  As soon as the 
> surviving qemu exited (or crashed), the image would be lost for ever.
>
> To get (1) working correctly we need an event that tells management 
> that all initialization has completed and the guest is ready to run 
> (so management can terminate the source).
>