[one-users] Using Xen phy devices in VM templates

Stefan Freitag stefan.freitag at udo.edu
Mon Feb 22 00:14:59 PST 2010 

Hello Ruben

taking the lvm driver as example helped to solve this issue :-)


Cheers
Stefan


Ruben S. Montero schrieb:
> Hi
>
> Yes you are right, I only checked the RAW option with the manual, (and
> it says that you can use several disk attributes). However I've tested
> in our machines and you are right Xen only use the last one...
>
> Then, the best way is to ssh the cluster node and link there the
> device, that should be pretty straightforward. Maybe you can take a
> look to tm_clone.sh in the lvm driver, there you can see an example of
> ssh the host an link a physical volume.
>
>
> Cheers
>
> Ruben
>
>
>
>
> On Fri, Feb 19, 2010 at 2:19 PM, Stefan Freitag <stefan.freitag at udo.edu>
> wrote:
>> Hello Ruben,
>>
>> I already thought about this idea, but then it would not be sufficient
>> to
>> pass only
>> one of the disks via the RAW section. All three of them would be needed.
>>
>> This is because, when I use the OpenNebula DISK section for two of my
>> disks and the
>> RAW section for the third disk, the resulting xen configuration file
>> will contain two lines  with the disk keyword, right?
>> E.g.:
>>
>> disk = [ 'file:/vm/udo-wn099_root.img,xvda,w',
>> 'file:/vm/udo-wn099_swap.img,xvdb,w']
>> disk = [ 'phy:/dev/cciss/c0d0p4,xvdc,w']
>>
>> From what I know about Xen this is somehow an invalid configuration. Xen
>> will only
>> use one of both lines and ignore the other :-(
>>
>>
>>
>> Cheers
>> Stefan
>>
>>
>> Ruben S. Montero schrieb:
>>> Sorry, I meant:
>>>
>>>
>>> RAW     = [
>>>       type  = "xen",
>>>       data  = "disk=['phy:/dev/cciss/c0d0p4,xvdc,w']"
>>> ]
>>>
>>> On Fri, Feb 19, 2010 at 12:23 PM, Ruben S. Montero
>>> <rubensm at dacya.ucm.es>
>>> wrote:
>>>> Hi,
>>>>
>>>> I think you can try two approaches:
>>>>
>>>>  * Modify the TM script so it ssh the node to make the link. You can
>>>> use the tm_clone.sh as starting point from the ssh suite to do so.
>>>>
>>>>  * As you do not really require anything from the TM. You can use the
>>>> RAW attribute, add:
>>>> RAW     = [
>>>>      type  = "xen",
>>>>      data  = " 'phy:/dev/cciss/c0d0p4,xvdc,w' ]"
>>>> ]
>>>>
>>>> Cheers
>>>>
>>>> Ruben
>>>>
>>>> On Fri, Feb 19, 2010 at 11:14 AM, Stefan Freitag
>>>> <stefan.freitag at udo.edu> wrote:
>>>>> Hi Ruben,
>>>>>
>>>>> thanks for the reply.
>>>>> I added the block and clone information to the template and
>>>>> tried to deploy an appliance.
>>>>>
>>>>> From what I see by browsing the log files, OpenNebula is now trying
>>>>> to
>>>>> create a link from <OpenNebula Server>:/dev/cciss/c0d0p4 to
>>>>> <TargetHost>:/vm/$VMID/images/disk.2.
>>>>>
>>>>> == transfer prolog log snippet==
>>>>> LN one:/dev/cciss/c0d0p4 udo-bl6107:/vm/72/images/disk.2
>>>>> == transfer prolog log snippet==
>>>>>
>>>>> == vm log snippet ==
>>>>> Command execution fail: /opt/one/lib/tm_commands/ssh/tm_ln.sh
>>>>> one:/dev/cciss/c0d0p4 udo-bl6107:/vm/72/images/disk.2
>>>>> == vm log snippet ==
>>>>>
>>>>> That's not ok for my use case :-(
>>>>>
>>>>> The device /dev/cciss/c0d0p4 is not a shared one and therefore does
>>>>> not
>>>>> exist at the server running OpenNebula. Each clusternode has its down
>>>>> device /dev/cciss/c0d0p4 as this created as partition the the
>>>>> internal
>>>>> blade server hard disk.
>>>>>
>>>>> Is it possible to pass a type of disk to OpenNebula, that is just
>>>>> piped
>>>>> through and added to the Xen configuration file? Like
>>>>>
>>>>> DISK = [
>>>>>   type ="local"
>>>>>   source="/dev/cciss/c0d0p4",
>>>>>   target   = "xvdc",
>>>>>   readonly = "no" ]
>>>>>
>>>>> resulting in the needed phy entry in the configuration file?
>>>>>
>>>>>
>>>>> Kind regards
>>>>> Stefan
>>>>>
>>>>>
>>>>> Ruben S. Montero schrieb:
>>>>>> Hi
>>>>>>
>>>>>> That should be :
>>>>>> DISK = [
>>>>>>     type="block"
>>>>>>     clone="no"
>>>>>>     source="/dev/cciss/c0d0p4",
>>>>>>     target   = "xvdc",
>>>>>>     readonly = "no" ]
>>>>>>
>>>>>> This should generate something like
>>>>>>
>>>>>> 'phy:$VM_DIR/var/$VMID/disk.3,xvdc,w'
>>>>>>
>>>>>> Note also the clone="no" part. The Transfer Manager will try to
>>>>>> setup
>>>>>> the disk in the VM home directory; if each cluster node has the same
>>>>>> scratch partition we can just link the device there. The script that
>>>>>> actually makes the link is tm_ln.sh in
>>>>>> $ONE_LOCATION/lib/tm_commands/nfs (or ssh if you are not using
>>>>>> NFS...), just in case you need to tune something...
>>>>>>
>>>>>> Cheers
>>>>>>
>>>>>> Ruben
>>>>>>
>>>>>>
>>>>>> On Thu, Feb 18, 2010 at 8:54 PM, Stefan Freitag
>>>>>> <stefan.freitag at udo.edu>
>>>>>> wrote:
>>>>>>> Dear all,
>>>>>>>
>>>>>>> at present I am using OpenNebula 1.4 to deploy virtual appliances
>>>>>>> to a Xen-based compute cluster. On each of the cluster nodes there
>>>>>>> exists a hard disk partition that should be used as scratch
>>>>>>> directory
>>>>>>> inside the virtual appliance (only 1 appliances is assigned to 1
>>>>>>> server
>>>>>>> at a time, so there is no conflict).
>>>>>>>
>>>>>>>
>>>>>>> I created a template to describe the appliances and got stuck. Here
>>>>>>> is
>>>>>>> what I did so far concerning the hard disks used in the virtual
>>>>>>> appliance:
>>>>>>>
>>>>>>> 1) that is the image containing the OS, boot and root directory
>>>>>>> DISK = [
>>>>>>>    source   =
>>>>>>> "/mnt/gridconfig/images/workernode/wn_sl54_x86_64.img",
>>>>>>>    target   = "xvda",
>>>>>>>    readonly = "no" ]
>>>>>>>
>>>>>>> 2) a swap partition that is created on the fly by OpenNebula
>>>>>>> DISK = [
>>>>>>>    type = swap,
>>>>>>>    size = 1024,
>>>>>>>    target   = "xvdb",
>>>>>>>    readonly = "no" ]
>>>>>>>
>>>>>>>
>>>>>>> 3) I thought that this could work
>>>>>>> DISK = [
>>>>>>>    source="/dev/cciss/c0d0p4",
>>>>>>>    target   = "xvdc",
>>>>>>>    readonly = "no" ]
>>>>>>>
>>>>>>> but in the OpenNebula documentation one can read that without a
>>>>>>> specifying a type, "disk" is assumed and I need a Xen phy: device.
>>>>>>>
>>>>>>> The thing I need to express with OpenNebula needs to be translated
>>>>>>>  to
>>>>>>> something like
>>>>>>>
>>>>>>> disk = [ 'file:/vm/udo-wn099_root.img,xvda,w',
>>>>>>> 'file:/vm/udo-wn099_swap.img,xvdb,w',
>>>>>>> 'phy:/dev/cciss/c0d0p4,xvdc,w'
>>>>>>> ]
>>>>>>>
>>>>>>> in Xen-speak.
>>>>>>>
>>>>>>>
>>>>>>> What do I need to specify to make use of the phy: partition located
>>>>>>> at
>>>>>>> each of the cluster nodes?
>>>>>>>
>>>>>>>
>>>>>>> Kind regards,
>>>>>>> Stefan
>>>>>>>
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> Users mailing list
>>>>>>> Users at lists.opennebula.org
>>>>>>> http://lists.opennebula.org/listinfo.cgi/users-opennebula.org
>>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Dr. Ruben Santiago Montero
>>>>>> Associate Professor, Complutense University of Madrid
>>>>>>
>>>>>> URL:    http://dsa-research.org/doku.php?id=people:ruben
>>>>>> Weblog: http://blog.dsa-research.org/?author=7
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Dr. Ruben Santiago Montero
>>>> Associate Professor, Complutense University of Madrid
>>>>
>>>> URL:    http://dsa-research.org/doku.php?id=people:ruben
>>>> Weblog: http://blog.dsa-research.org/?author=7
>>>>
>>>
>>>
>>>
>>> --
>>> Dr. Ruben Santiago Montero
>>> Associate Professor, Complutense University of Madrid
>>>
>>> URL:    http://dsa-research.org/doku.php?id=people:ruben
>>> Weblog: http://blog.dsa-research.org/?author=7
>>>
>>
>>
>>
>>
>>
>
>
>
> --
> Dr. Ruben Santiago Montero
> Associate Professor, Complutense University of Madrid
>
> URL:    http://dsa-research.org/doku.php?id=people:ruben
> Weblog: http://blog.dsa-research.org/?author=7
>

More information about the Users mailing list