On Tue, Apr 24, 2007 at 10:57:09PM +0900, Kazuki Mizushima wrote: Hi Kazuki, Hum, interesting usage, of course it sounds difficult to really fully automate it in a completely generic way. Okay, so let's look first at how it would integrate in virsh ... So the command take the domain reference and a comma separated list of device. I assume it expect those device to be already available (for example if using LVM they were already created) and already initialized. I'm surprized the name for the new clone is not an argument. I wonder how we could simplify the device creation, and also how we could preinitialize the disks. It sounds to me a bit more on the area of Cobbler than libvirt/virsh though. The problem of adding a Clone operation at the libvirt level is that it would be very hard to make it solid, I mean just modifying an XML file is not hard, it's the checking of the state, the provisionning and making sure that no disaster will happen if we try to start the domain which are the hard part. I'm not sure we can do that in a relatively generic fashion, it probably depends a lot on the scenario. To get down to something concrete, suppose I have a domain with an Apache server running a RHEL4 paravirt. The new API may sound like it could create a new domain with a similar Apache server, but: - how do we provision it ? - how do we avoid name clashes, i.e. IP, machine name, ... in practice it would mean changing settings in the server - are we using resource which may not be shared (or shareable) We would need to be very precise about what the API call does, and warn about what it does not do, before I would start feeling comfortable about adding it in libvirt, but it's definitely something which could have some value, at least in some scenario ! Daniel -- Red Hat Virtualization group http://redhat.com/virtualization/ Daniel Veillard | virtualization library http://libvirt.org/ veillard(a)redhat.com | libxml GNOME XML XSLT toolkit http://xmlsoft.org/ http://veillard.com/ | Rpmfind RPM search engine http://rpmfind.net/

On Wed, Apr 25, 2007 at 04:32:06PM +0900, Kazuki Mizushima wrote: Hi Kazuki, Look this is all arguments already presents in some form at the virt-manager level. There is just too many similarities to ignore it. Remote support will need plugging at the API level. I think prototyping within virt-manager will get you to more feedback and features faster. Then in turn this will make the analysis required to check feasibility and do the API design faster too. IMHO you will get where you want faster if you do the first tool step in virt-manager, really. Daniel -- Red Hat Virtualization group http://redhat.com/virtualization/ Daniel Veillard | virtualization library http://libvirt.org/ veillard(a)redhat.com | libxml GNOME XML XSLT toolkit http://xmlsoft.org/ http://veillard.com/ | Rpmfind RPM search engine http://rpmfind.net/

Kazuki Mizushima wrote: [...] Right - didn't see the patch the first time around ... I think the clone should be done as a libvirt call, rather than in virsh. Adding stuff to virsh doesn't really help other libvirt users, nor would it work in the remote case. If it goes into libvirt on the other hand, then you get remote "for free" (in this case) and other users of libvirt can implement it, in particular virt-manager. Rich. -- Emerging Technologies, Red Hat http://et.redhat.com/~rjones/ 64 Baker Street, London, W1U 7DF Mobile: +44 7866 314 421 Registered Address: Red Hat UK Ltd, Amberley Place, 107-111 Peascod Street, Windsor, Berkshire, SL4 1TE, United Kingdom. Registered in England and Wales under Company Registration No. 3798903 Directors: Michael Cunningham (USA), Charlie Peters (USA) and David Owens (Ireland)

On Tue, Apr 24, 2007 at 10:57:09PM +0900, Kazuki Mizushima wrote: While this works ok in simple cases, I'm not sure it neccessarily scales upto handle the more complex cases easily enough. With this prototype the user only has the option of specifying new device names. Conceivably the user could want to specify a whole lot more 'new' parameters - eg change the VNC port if it was not previously auto-allocated, specify a new MAC address (for any/all NICs), specify a UUID, etc. If using LVM for the backing store, instead of dd'ing the disk they could simply take a (writable) snapshot of the disk - this could even be done while the source VM was active (assuming use of a journalling FS inside the guest). If doing a flat copy of the disks, it will also be important to provide progress feedback to the user - when initially allocating guests it takes a good 30-60 seconds per GB of disk to be allocated - cloning an existing guest will be longer because instead of just writing zero's you are reading & writing data. For remote support, we'd really need to have some basic storage management APIs to let us allocate new 'chunks' of storage for the virtual disks. I think cloning should be kept separate from the main libvirt API because there are simply too many different ways to approach it, depending on your application use-cases. The way I like to look at it, is that we've just got several different ways to 'install' a guest 1. Traditional installation media (eg CDROM, HTTP kickstart) 2. Boot a LiveCD which can install itself to disk 3. Create from a VM 'template', possibly cloning some base disk 4. Cloning an existing provisioned VM Now virt-install currently deals with the first option, and there is some work going on to experiment with making it do options 2 & 3 too. So I think that it'd be worth trying to add some form of cloning capability to virt-install too. Thinking how it might work.. Currently to install from HTTP you might do virt-install --name myvm --ram 500 --mac 54:24:52:23:55:43 --vnc --vncunused --url http://download.fedora.redhat.com/pub/fedora/linux/core/6/x86_64/os/ --filesize 5 --file /var/lib/xen/images/myvm.img So, how might it work with cloning ? Basically install of the --url arg to point to an install source, we'd allow '--clone VMNAME' to point to an existing virtual machine. The main config would be determined by the settings of that VM, with the command line args being used to override specific bits. As a bare minimum example one would do virt-install --name othervm --clone myvm --file /var/lib/xen/images/othervm.img But for a more complex case one could use all the various args virt-install --name othervm --clone myvm --uuid 2452:ef23:ee3e:2232:442424252525ffe --vncport 5902 --mac 42:55:23:66:23:44 --file /var/lib/xen/images/myvm.img --file /var/lib/xen/images/myvmdata.img Now without prototyping it, I'm not sure how much code re-use we'd get within virt-install, so it may turn out that it is simpler to just have a separate tool called virt-clone for this purpose. In either case I reckon we could distribute this as part of the main virtinst codebase, so it is still easy for us to access the functionality from virt-manager. Regards, Dan. -- |=- Red Hat, Engineering, Emerging Technologies, Boston. +1 978 392 2496 -=| |=- Perl modules: http://search.cpan.org/~danberr/ -=| |=- Projects: http://freshmeat.net/~danielpb/ -=| |=- GnuPG: 7D3B9505 F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 -=|