[libvirt] [PATCH] Allow off-line removal of devices via xend

[Cole Robinson]

Guido Günther wrote:
> On Tue, Feb 10, 2009 at 09:32:49PM +0100, Stefan de Konink wrote:
>> Guido Günther wrote:
>>> On Mon, Feb 09, 2009 at 11:30:30AM -0500, Cole Robinson wrote:
>>> [..snip..] 
>>>> No fault of this patch, but this makes me ask the question: what exactly
>>>> are attachDevice and detachDevice supposed to do?
>>>>
>>>> In the qemu driver, it is entirely hotplug/hotunplug, and deliberately
>>>> fails on an inactive domain. This patch seems to add offline support
>>> We don't even persist the devices being hot{added,removed} into the
>>> domain xml so acting on inactive domains has no meaning or am I missing
>>> something?
>> When the domain is enabled, and you store it... you basically get that  
>> behavior.
> Exactly my point: when the domain is *enabled*.

I'm a little lost. My point was basically that with this patch to xend,
attach/detach has different meanings between the qemu + xend driver.

qemu :   vm active   : attempt to hotplug the device and persist xml

                       changes on success, raise error if device is not
                       not hotpluggable

         vm inactive : deliberately fail: 'vm must be active'

xen  :   vm active   : attempt to hotplug device, unknown if changes are
                       persisted, unknown result if device is not
                       hotpluggable (unknown to me, anyways)
	
	 vm inactive : _dont_ fail, add device to VM config

The libvirt docs for attach/detach device are:

attachDevice: Create a virtual device attachment to backend.
detachDevice: Destroy a virtual device attachment to backend.

Not very helpful. All I want is some understanding of what attach/detach
is supposed to mean :)

Adding the device to the xml for an inactive domain is fine by me
(though it would essentially be a convenience function), but I would
like to ensure that if the domain is active and the device is not
hotpluggable, we throw a failure. Without that, there isn't any way to
determine that attachDevice on the active domain did anything
immediately useful.

Thanks,
Cole