openstack team mailing list archive

[Jay Pipes <jaypipes@xxxxxxxxx>]

On 07/28/2012 11:10 AM, Gaurab Basu wrote:
> Another thing I would like to know is whether it uses snapshot mechanism
> over time.

What is "it" you are referring to above? Are you asking whether Nova
automatically takes snapshots of images over time? If so, no, it does
not. If a user requests a snapshot of a launched instance, then Nova
will issue snapshot commands -- in the case of the libvirt driver, these
commands would be qemu-img snapshot -c <SNAPSHOT_NAME> <IMAGE_PATH>.

> I mean how does the copy on write functionality works. Does it keep the
> diff snapshots over time ( or something else ).

Not sure here whether you are asking how QEMU's copy on write operations
work or whether Nova keeps the base images separate from any VM images.
If you are asking about the latter, the answer is that Nova will create
the virtual machine images by creating a COW image based on the base
image it pulls from Glance -- after making a resized copy of the base
image if it needs to do so to meet the needs of the requested image size
of the VM.

Snapshots that are taken of virtual machine images on a host are stored
by Nova in Glance.

> And does the diff work at the level of file or block level?

AFAIK, CoW and snapshot actions with QEMU are block-level.

> What is the format that the image is converted to after it is fetched
> from glance.

There may be no conversion needed at all... it depends on what the
format of the original base image that was stored in Glance. Conversion
between raw/iso and QCOW2 and vice versa is what you see in the code,
and is what is done during migration as Mikal mentioned below.

> I am fairly new to openstack.
> Can you point me to the specific files in the code where all these
> things are coded. I want to know the details of the
> present state.

grep for qemu-img in the nova/ directory. You'll see all the files that
call qemu-img commands and then you can go look in those files.

Best,
-jay

> Thanks again for your help.
> 
> Regards,
> Gaurab
> 
> On Sat, Jul 28, 2012 at 11:52 AM, Michael Still
> <michael.still@xxxxxxxxxxxxx <mailto:michael.still@xxxxxxxxxxxxx>> wrote:
> 
>     On 28/07/12 05:42, Gaurab Basu wrote:
>     > Hi,
>     >
>     > I am trying to figure out the technology that openstack uses when
>     > multiple VM's having the*same *base image (OS) are provisioned on a
>     > physical server.
>     > Does it use as many copy as the number of VM's or does it use the same
>     > base image and then copy on write.
>     >
>     > I need to understand the complete details. Can anybody share some
>     > details or point me to some place where I can find the details.
> 
>     Its pretty hard to provide a complete description of what happens,
>     because the code keeps changing. However, assuming you have copy on
>     write turned on (which is the default IIRC), and assuming that all of
>     the instances have the same disk size, then you end up with:
> 
>      - the image as fetched from glance, with possible format conversion
>      - that image resized to the size the instance requested
>      - a cow on write layer for each instance that is using that sized image
> 
>     The first should be smallish, the second can be quite large, and the
>     third will really depend on how much writing the instances are doing.
> 
>     Note that this all falls apart if instances are migrated, because as
>     part of the migration the copy on write layer is transformed into a full
>     disk image, which is what is shipped over to the new machine.
> 
>     Hope this helps,
>     Mikal
> 
> 
> 
> 
> 
> _______________________________________________
> Mailing list: https://launchpad.net/~openstack
> Post to     : openstack@xxxxxxxxxxxxxxxxxxx
> Unsubscribe : https://launchpad.net/~openstack
> More help   : https://help.launchpad.net/ListHelp
>