[Engine-devel] the future of template cloning

Tue Jan 17 16:04:42 UTC 2012

On 01/17/2012 10:49 AM, Livnat Peer wrote:
> On 17/01/12 17:04, Omer Frenkel wrote:
>>
>> ----- Original Message -----
>>> From: "Jon Choate"<jchoate at redhat.com>
>>> To: engine-devel at ovirt.org
>>> Sent: Tuesday, January 17, 2012 3:52:34 PM
>>> Subject: Re: [Engine-devel] the future of template cloning
>>>
>>> On 01/17/2012 07:29 AM, Livnat Peer wrote:
>>>> On 17/01/12 11:45, Ayal Baron wrote:
>>>>> ----- Original Message -----
>>>>>> On 17/01/12 10:46, Itamar Heim wrote:
>>>>>>> On 01/17/2012 10:32 AM, Omer Frenkel wrote:
>>>>>>>> ----- Original Message -----
>>>>>>>>> From: "Itamar Heim"<iheim at redhat.com>
>>>>>>>>> To: "Jon Choate"<jchoate at redhat.com>
>>>>>>>>> Cc: engine-devel at ovirt.org
>>>>>>>>> Sent: Monday, January 16, 2012 7:26:24 PM
>>>>>>>>> Subject: Re: [Engine-devel] the future of template cloning
>>>>>>>>>
>>>>>>>>> On 01/16/2012 06:16 PM, Jon Choate wrote:
>>>>>>>>>> On 01/16/2012 10:58 AM, Itamar Heim wrote:
>>>>>>>>>>> On 01/16/2012 05:46 PM, Jon Choate wrote:
>>>>>>>>>>>> On 01/16/2012 09:46 AM, Livnat Peer wrote:
>>>>>>>>>>>>> On 12/01/12 22:45, Ayal Baron wrote:
>>>>>>>>>>>>>> ----- Original Message -----
>>>>>>>>>>>>>>> We are going to be able to store the disks for a
>>>>>>>>>>>>>>> template
>>>>>>>>>>>>>>> on
>>>>>>>>>>>>>>> different storage domains due to the multiple storage
>>>>>>>>>>>>>>> domain
>>>>>>>>>>>>>>> feature. Cloning a template will still be possible, but
>>>>>>>>>>>>>>> will
>>>>>>>>>>>>>>> it
>>>>>>>>>>>>>>> provide any value? Thoughts?
>>>>>>>>>>>>>> I see no relation between the two options.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Scenario 1: I can create a VM with a single disk and
>>>>>>>>>>>>>> create
>>>>>>>>>>>>>> a
>>>>>>>>>>>>>> template from it.
>>>>>>>>>>>>>> I would still want to be able to clone the template in
>>>>>>>>>>>>>> order
>>>>>>>>>>>>>> to
>>>>>>>>>>>>>> provision VMs from it on different domains.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Scenario 2: same thing with multiple disks on same
>>>>>>>>>>>>>> domain.
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> Scenario 3: I have a template with 2 disks on 2 different
>>>>>>>>>>>>>> domains
>>>>>>>>>>>>>> (domain A and domain B) and I want to have another copy
>>>>>>>>>>>>>> of
>>>>>>>>>>>>>> the
>>>>>>>>>>>>>> template on domain C and domain D
>>>>>>>>>>>>>>
>>>>>>>>>>>>> Hi Jon,
>>>>>>>>>>>>>
>>>>>>>>>>>>> After talking to Michael Pasternak it seems that we did
>>>>>>>>>>>>> not
>>>>>>>>>>>>> implemented
>>>>>>>>>>>>> copyTemplate in the REST API, it seems to be a gap that we
>>>>>>>>>>>>> have.
>>>>>>>>>>>>>
>>>>>>>>>>>>> This gap is playing in our favor, we can remove the
>>>>>>>>>>>>> copyTemplate
>>>>>>>>>>>>> verb
>>>>>>>>>>>>> and introduce copyDisk verb.
>>>>>>>>>>>>>
>>>>>>>>>>>>> The template disks can be copied to another SD.
>>>>>>>>>>>>> When creating a VM from template the user can choose per
>>>>>>>>>>>>> disk
>>>>>>>>>>>>> the
>>>>>>>>>>>>> destination SD (only SD with the disks are eligible
>>>>>>>>>>>>> candidates).
>>>>>>>>>>>> wait, when creating a VM from a template, the user won't
>>>>>>>>>>>> get a
>>>>>>>>>>>> choice
>>>>>>>>>>>> will they? Won't the VM disks have to go on the same
>>>>>>>>>>>> storage
>>>>>>>>>>>> domain as
>>>>>>>>>>>> the template disks they were created from?
>>>>>>>>>>> yes, but the template disks can be copied to multiple
>>>>>>>>>>> storage
>>>>>>>>>>> domains,
>>>>>>>>>>> so the user can choose for the VM/disk which storage domain
>>>>>>>>>>> to
>>>>>>>>>>> create
>>>>>>>>>>> them from (per storage domains that have copies of that
>>>>>>>>>>> disk)
>>>>>>>>>> OH! I totally misunderstood. So what you are saying is that a
>>>>>>>>>> template
>>>>>>>>>> can have N number of copies of the same disk each on a
>>>>>>>>>> different
>>>>>>>>>> storage
>>>>>>>>>> domain. I had thought that if you wanted that type of
>>>>>>>>>> situation
>>>>>>>>>> you
>>>>>>>>>> would have multiple copies of the template itself too.
>>>>>>>> yes, one copy of disk per domain though.
>>>>>>>>
>>>>>>>>>> Just to be clear, does this mean that the plan is to phase
>>>>>>>>>> out
>>>>>>>>>> the
>>>>>>>>>> current clone template command and instead implementing a
>>>>>>>>>> clone
>>>>>>>>>> disk
>>>>>>>>>> command so that a template can duplicate its disks
>>>>>>>>>> individually?
>>>>>>>>> pretty much, yes.
>>>>>>>>> though i'd imagine 'clone template' would still be useful to
>>>>>>>>> have
>>>>>>>>> for
>>>>>>>>> the user. not sure if it implies core should expose it as well
>>>>>>>>> to
>>>>>>>>> allow
>>>>>>>>> easier usage at UI level for such a task.
>>>>>>>> we can leave it untouched - means copyTemplate get 1
>>>>>>>> destination
>>>>>>>> domain, and copies all disks to it,
>>>>>>>> but i think it will be unusable (and problematic - what if one
>>>>>>>> of
>>>>>>>> the
>>>>>>>> disks already exists on the destination?),
>>>>>>> then don't copy it, it is already there
>>>>>>>
>>>>>> I agree with Omer, there is no reason to support copy template,
>>>>>> if
>>>>>> the
>>>>>> user wants to clone all the disks he can use multiple actions, we
>>>>>> don't
>>>>>> need a specific verb for this.
>>>>> Reason or lack of depends on the common usage.
>>>>> If we assume that normally all disks of a template would reside on
>>>>> the same domain then it makes sense to have a verb to copy the
>>>>> template in its entirety and not burden the user.
>>>>> The general recommendation should be to use a single storage
>>>>> domain, so I think there is room for such a verb.
>>>>>
>>>>>> If the UI chooses to expose such operation it will use the
>>>>>> multipleRunAction API which makes it easier to expose to the user
>>>>>> partial success, we could clone disk A and Disk B but Disk C
>>>>>> failed
>>>>>> etc.
>>>>> The multipleRunAction is for user marking multiple objects in GUI
>>>>> and running an action on all of these objects.
>>>> Exactly, choosing the disks to copy to the storage domain.
>>>>
>>>>> Here however, the action the user wants is to copy 1 object (the
>>>>> template) which has sub objects and it should run as a single
>>>>> action.
>>>> We are not cloning the template itself we are only cloning the
>>>> template
>>>> disks.
>>>>
>>>>
>>>>> For example, if there is enough space on the target domain for 2/4
>>>>> disks then using multipleRunAction would result in 2 disks being
>>>>> copied and 2 failing.
>>>>> If however it is a single action then the free space test would
>>>>> fail the entire action and user would be able to choose if he
>>>>> wants to copy just 2.
>>>>> Note that in this case, actually performing the copy of the 2
>>>>> disks is detrimental as it can negatively affect VMs on this
>>>>> domain.
>>>>>
>>>>>>
>>>>>>>> what the user really wants is to specify which disks to copy
>>>>>>>> and destination per disk, and i don't see a reason to create a
>>>>>>>> backend
>>>>>>>> command to do it
>>>>>>>>
>>>>>>>>> _______________________________________________
>>>>>>>>> Engine-devel mailing list
>>>>>>>>> Engine-devel at ovirt.org
>>>>>>>>> http://lists.ovirt.org/mailman/listinfo/engine-devel
>>>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> Engine-devel mailing list
>>>>>>> Engine-devel at ovirt.org
>>>>>>> http://lists.ovirt.org/mailman/listinfo/engine-devel
>>>>>> _______________________________________________
>>>>>> Engine-devel mailing list
>>>>>> Engine-devel at ovirt.org
>>>>>> http://lists.ovirt.org/mailman/listinfo/engine-devel
>>>>>>
>>>> _______________________________________________
>>>> Engine-devel mailing list
>>>> Engine-devel at ovirt.org
>>>> http://lists.ovirt.org/mailman/listinfo/engine-devel
>>> I would like this discussion to continue but I need some short-term
>>> closure since I only have two days to get something into code. He is
>>> what I am proposing to do:
>>>
>>> 1) leave clone template as it is. It will try to copy all of the
>>> disks
>>> to the same storage domain.
>>> 2) Expose a copy disk command so that the user can select a single
>>> disk
>>> and create a copy of it on another storage domain.
>>>
>>> Is everyone ok with starting there and then refining once we can
>>> reach
>>> more of a consensus on what the end product should be?
>> fine by me, although i think we should remove the clone template (actually called copy template),
>> as i think the copy disk gives a way to do it, with more granularity,
>> and the user will know exactly what to expect.
>>
>
> We are removing the cloneTemplate and introducing cloneTemplatedisk,
> following the above discussion does anyone have strong objection for
> doing this?
>

We can't remove cloneTemplate until it is removed from the UI or else we 
will break functionality.  For now we are just going to ensure that it 
works as it always has until the UI is ready to remove it.

>> unfortunately, there is another issue,
>> currently, since template can be fully on a domain or not,
>> remove template can get list of storage domains to remove the template from,
>> and if the list contains all the domain the template is on, the template is fully deleted from the db,
>> if its partial, then only the copies of the disks are removed from the domain.
>>
> Worth Noting that we have another REST API gap that we can use in our
> favor, the delete template from storage domain is not modeled yet and we
> don't need to support it.
>
>> what i suggest is, making it simple:
>> remove template will remove the template, and all of the disks copies from all the domains.
> +1 for that approach.
>
>> we will need new removeTemplateDisk command that will remove a disk from a domain,
>> as long there is another copy of the disk.
>> if its the last copy it should fail, as removing the disk from all the domains will change the template,
>> which is not a supported flow.
>>
>> thoughts?
> I would implement a single verb - removeDisk, it has optional parameter
> storage domain.
> If the storage domain is not specified it is simply removing a disk (for
> template, VM or floating) if a SD is specified in case there is only one
> image that represent this disk (the only use case for VMs) we remove the
> disk (same as no passing no SD except for extra validation) if more than
> one image is associated with this disk (the image-SD map has more than
> one entry) then remove the relevant mapping.
>
> Note: in case of deleting the disk and there is only one image
> associated with the disk we should remove the device from the vm_devices
> table as well.
>
> Implementing a single generic verb makes it easier to keep backwards
> compatibility when the model changes.
>
> Livnat
>
>
>
>>> _______________________________________________
>>> Engine-devel mailing list
>>> Engine-devel at ovirt.org
>>> http://lists.ovirt.org/mailman/listinfo/engine-devel
>>>
>> _______________________________________________
>> Engine-devel mailing list
>> Engine-devel at ovirt.org
>> http://lists.ovirt.org/mailman/listinfo/engine-devel