[libvirt] [PATCH 0/5]Add startupPolicy attribute support for hard disks

[Guannan Ren]

On 03/19/2013 08:21 PM, Daniel P. Berrange wrote:
> On Tue, Mar 19, 2013 at 03:56:40PM +0800, Guannan Ren wrote:
>> On 03/18/2013 10:42 PM, Daniel P. Berrange wrote:
>>> On Mon, Mar 18, 2013 at 05:10:17PM +0800, Guannan Ren wrote:
>>>> The set of patches is trying to add 'startupPolicy' attribute support
>>>> to the source element of hard disks. Policy levels are using the
>>>> mandatory, requisite, optional levels as originally documented.
>>>>
>>>> For the 'optional' policy, there is a little difference from CDROM and
>>>> Floppy which only drop its source path, for disks, if missing, the
>>>> checking function will drop their definitions, because qemu doesn't
>>>> allow missing source path for hard disk.
>>> What is the motivation for this feature ? I personally find even the
>>> existing CDROM code for this to be of rather dubious value, so would
>>> like to see a clear use case for why we need to expand this hack.
>>>
>>> Daniel
>>     There is a real case from a libvirt customer request, which needs
>>     to make disk of block type use "optional" policy.
>>
>>     "The DR concept for our virtualization infrastructure is based upon
>>      host-based mirroring of independent SAN LUNs.
>>      Because RH does not support CLVM spanning multiple data centers
>>      we have to pass the LUNs to the KVM and mirror inside of the KVM.
>>      If there is an outage taking down a storage box or an entire DC,
>>      we will be unable to migrate, relocate or start any KVM,
>>      despite the KVM being able to run just on one side of the mirror."
>>
>>      "Declare all devices as optional. If there aren't any mirror
>> parts left,
>>       the KVM will fail, but that is an acceptable behaviour that will be
>>       recognized by our monitoring software. There is no need to have
>>       libvirt guard against failure of storage devices in such a setup. "
> Ok, so they are doing multi-pathing inside the guest, and so don't care
> if some of the LUNs on the host are not available when started. This
> makes more sense now as a feature.
>
> Regards,
> Daniel

    Okay, thanks, I will rebase this patchset because [3/5] has been pushed
    by the same patch from Jiri.