On Thu, 2007-11-15 at 14:20 -0500, Hugh O. Brock wrote: > On Thu, Nov 15, 2007 at 07:14:29PM +0000, David Lutterkort wrote: > > > > Another route (though one that takes more effort), is to base install on > > virt-image metadata - the idea is that we would ship stock virt-image > > files for installable OS's; those files would contain the salient bits > > about an OS (such as APIC/ACPI, whether to enable a graphics console, > > kernel cmdline for pv) virt-manager would have to modify those files a > > little before passing them off to virt-image, mostly to put things like > > the ISO location and path to the root disk in. > > > > That seems like the right way to do it. We need an image-packager though, right? Not necessarily .. I was thinking that virt-manager would edit the stock image.xml to include user-provided information about where the install ISO and the target disk file are. The image.xml we ship would be complete except that it doesn't contain a <storage/> section, and that the disk/target mapping is missing; those owuld be filled in by virt-manager before kicking off the install. IOW, we'd use it mostly for expressing arch, features, and whether to enable graphics. There's a few problems still though: (1) virt-image doesn't let you assign a MAC from XML and (2) IIRC it doesn't handle LVM volumes. Both shouldn't be too hard to address. The main point is that the OS-specific install information should move out of the code into some data files - and virt-image seems to be pretty close in terms of what needs to be expressed. David _______________________________________________ et-mgmt-tools mailing list et-mgmt-tools@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/et-mgmt-tools