[Kimchi-devel] [RFC] Guest cloning

[Crístian Viana]

On Sex, 2014-10-03 at 08:28 -0500, Brent Baude wrote:
> At one point Adam had really suggested we look at how guests are defined
> and changed.  In particular, straight copies of XML and or straight
> edits of XML could lead (after time) to the XML digressing into
> something libvirt might not understand at some point.  

Do you remember exactly what in the XML file digressed like that? It
would be very nice to know it.

> To prevent this sort of problem, we thought that each time a guest is
> edited, the XML is created entirely anew but of course using the same
> values.

I first thought about doing that as well, but I'm afraid our own new XML
does not have all the features the original VM had. For example, on
Kimchi we don't have support to custom CPU flags yet (i.e. we don't
add/change those values). What would happen if we clone an existing VM
with "unsupported" (as of Kimchi) features like that one? If we don't
know about them and we're creating a new XML from scratch, we're only
going to add what we're aware of. And I think this could make our
cloning feature very unrealiable. Reading an existing XML and redefining
it will always propagate whatever there is in that VM. As long as the
XML doesn't get digressed, of course.

Here's one current scenario for me. I have a few VMs on my laptop which
I use to test Kimchi on them. As they need to be a hypervisor in order
to run Kimchi, I have to add a few CPU flags to their XML descriptors
which allow them to be a hypervisor (i.e. those Intel/AMD virtualization
parameters). I do that manually using "virsh edit". If I clone one of my
VMs using the approach above and it recreates their XML files from the
beginning, AFAIU, the CPU flags will not be copied along. And I'd be
very unsatisfied because the clone feature dropped a few things of my
original VM while cloning.

If we know what are the problems you had with Adam while copying the XML
files over and over, maybe it' easier to fix them than to try to create
a new one with all the libvirt's features, including the ones Kimchi is
still not aware of.

> About a month ago I had a working version of this in draft
> form.
> Would anyone be interested in reviewing it and or adding to it prior to
> implementing the clone function?  If so, ping me on IRC and I'll get you
> a copy of the patch.

I am interested on it! I'll ping you then.

-- 
Best regards,
Crístian.