[Kimchi-devel] [RFC]Create VM from image

[Royce Lv]

On 2014年07月02日 11:23, Aline Manera wrote:
>
> On 07/01/2014 04:53 AM, Royce Lv wrote:
>> Currently kimchi just support create a vm from an ISO defined 
>> template. This requires duplicate installation.
>> This release we want to support create a vm from an installed image.
>>
>> This involves the following change:
>>
>> 1. Create a template from an image:
>>
>> POST /templates/template-1
>> {'base': 'abs_base_img_path'}
>>
>> Implementation:
>> (1)probe image os distro and version to get suggested config.
>> (2)Avoid image to be deleted.
>> (3)Avoid using ISO and base image at the same time.
>>
>> 2. Create a vm from image based template:
>> This part has already been covered by Aline's prototype patch.
>>
>> POST /vms/vm-1
>> {'storagepool': 'default', 'network': 'default', 'disks':[{'index':0, 
>> 'base': 'abs_path'}]}
>> or
>> {'storagepool': 'default', 'network': 'default', 'disks':[{'index':0, 
>> 'base': 'abs_path', 'volume': 'scsi_volume'}]}
>>
>> The create vm implementation need to be re-considered for the 
>> following points:
>> (1) when creating vm on scsi/iscsi pool, if volume size smaller than 
>> original image, need to be rejected.
>
> From that, I assume you will copy the img content to the scsi/iscsi 
> volume. Is that correct?
> If so, can't we use backing store 
> (http://libvirt.org/formatstorage.html#StorageVolBacking) for the 
> existing img and make the vm volume the RW img?
>
> During my prototype development, I used backing store and it worked 
> well with directory storage volumes
> But I haven't tested it with scsi/iscsi volumes
Yes, we will use the backing file, but since the original image may be a 
small image with a large size, the rest most part will be write to the 
SCSI/iSCSI volume,
to *guarantee* volume will not explode in write, the cow volume needs to 
be >= declared.
It is a secure choice, but yes, there will be some waste in space.
>
>> (2) when creating vm on directory based pool, volume generating logic 
>> is different with allocation new volume.
>> (2) xml generating need to be refactored.
>>
>> Anything else need to be considered, welcome your comments.
>