On Sun, Jun 24, 2012 at 5:05 PM, Ata Bohra <ata.husain(a)hotmail.com> wrote:
Thanks Doug for your suggestions.
I believe you are correct about the relation between OVA and OVF. But I am
not 100 % possitive about your suggestion: "defining an appropriate domain
in libvirt". To understand better I am sharing more details about my plans:
1. Enhance libvirt interface code (libvirt.c) to provide a
domain-independent routine: virDomainCreateOVA, an alternate API to create
domain.
To make client code real simple, this routine can take ova path as input
and internally strip the OVA to extract required details. (planning to
define a struct to hold all essential
information).
2. Second, to enhance ESX driver to perform ESX specfic calls.
Given OVA is a tar file, the parsing is just another file open/read
operation; it would be simple to perform it inside domain_conf.c (infact I
have written a parser to strip information off OVA already).
Hope to get some comments/suggestions on above steps.
Thanks!
Ata
Right. I'm suggesting you don't go that route and approach the problem
from another angle. I did a little Googling since my last e-mail to at
least make sure I understood the basics. So an OVF looks like the
following:
virtualappliance/package.ovf
virtualappliance/disk1.vmdk
virtualappliance/disk2.vmdk
virtualappliance/cdrom.iso
virtualappliance/en-US-resources.xml
An OVA would simply be a tar of the above and named
virtualappliance.ova package.ovf is an XML file containing the
description of the hardware of the virtual machine, much like the XML
that libvirt stores about domains. While en-US-resources.xml would be
the US English descriptions of the machine and its hardware.
I'm suggesting you write an application that transforms package.ovf
into libvirt's own domain XML format and simply call
virDomainDefineXML() rather than adding API to libvirt itself. You
could then further extend the application to allow you to take a
libvirt domain and export it as a OVA.
Looking at VMWare and Xen, they both treat OVA/OVF as a foreign format
and require a converter application to import them to their native
internals so it wouldn't be much different than their approach.
Just my 2 cents.
--
Doug Goldstein