Re: [Qemu-devel] [RFC PATCH 0/4] Fix subsection ambiguity in the migration format

[Anthony Liguori]

On 07/26/2011 07:51 AM, Stefan Hajnoczi wrote:
> On Tue, Jul 26, 2011 at 10:48 AM, Stefan Hajnoczi
> <address@hidden>  wrote:
> > On Mon, Jul 25, 2011 at 06:23:17PM -0500, Anthony Liguori wrote:
> > However, doing so imposes extra work on management tools - they need to
> > understand and drive negotiation.  If QEMU adds a new capability we
> > might even need to update management tools!
> >
> > As a management tool author I would prefer the source and destination to
> > work it out amongst themselves so that I just issue the 'migrate'
> > command.  Negotiation can be done without the management tool's
> > involvement: fail migration if the initial negotation phase fails.
>
> An advantage I didn't think of was that management tools handling
> negotiation makes negotiation out-of-band and the migration protocol
> doesn't need to be changed.

Not quite that, but that you can detect when the migration changes.  For 
instance, this feature would allow the following behavior:

1) src doesn't know the new protocol, dst still supports the old 
protocol and the new protocol, migration uses old protocol.

2) src knows the new protocol, dst doesn't know the new protocol, old 
protocol is used.

3) src knows the new protocol, dst knows the new protocol, new protocol 
is used

4) src doesn't know the new protocol, dst chooses to only support the 
new protocol, migration fails gracefully.

Even if we only ever introduce a single feature, having the mechanism 
means that we can gracefully fail with a new format and have a 
transition period.

Regards,

Anthony Liguori

> It seems like the migration protocol needs an overhaul sooner or later
> anyway, so perhaps it's not work making the negotiation external.
>
> Stefan