Re: RFC: QEMU bumping memory limit and domain restore

[Ian Campbell <ian.campbell@citrix.com>]

On Thu, 2015-06-04 at 10:32 +0100, Andrew Cooper wrote:
> On 04/06/15 10:26, Ian Campbell wrote:
> > On Thu, 2015-06-04 at 10:14 +0100, Wei Liu wrote:
> >> The main objection is that we shouldn't call xc_domain_setmaxmem in the
> >> middle of a migration stream.
> > In the middle of an _xc_ migration stream.
> >
> > This seems like the sort of thing it would be OK to have in a (to be
> > introduced) libxl stream (which would itself contain the xc stream as a
> > data item).
> >
> > I think we are expecting such a thing to be introduced as part of the
> > libxl side of migration v2, aren't we?
> 
> No.  libxl migration v2 will not be affecting the behaviour here.

By "such a thing" I was referring to "a libxl stream format", not this
piece of data specifically.

This stream format however will be extensible (I hope) and therefore
could accommodate state information which differs from the domain
configuration.

> The only reasonable way I see this being fixed is for the libxl json
> blob

For _a_ libxl json blob, not necessarily the existing one, which is the
user's domain configuration information.

IOW it doesn't have to be the existing libxl_domain_config blob. (Nor
does it really need to be json, but whatever)

>  to contain the correct size of the VM, and for
> libxl_domain_create() to make an appropriately sized VM in the first place.
> 
> One extension which libxl migration v2 will bring is the ability to send
> a new json blob later in the stream, but such a fix still requires the
> json blob to have the correct size in it.
> 
> ~Andrew