[Engine-devel] Fwd: Problem in REST API handling/displaying of logical networks

[Michael Pasternak]

On 07/05/2012 12:06 PM, Livnat Peer wrote:
> On 05/07/12 11:56, Michael Pasternak wrote:
>> On 07/05/2012 11:40 AM, Livnat Peer wrote:
>>> On 05/07/12 11:31, Michael Pasternak wrote:
>>>> On 07/05/2012 10:51 AM, Livnat Peer wrote:
>>>>>>>>> Actually the API has the same concept as you suggest for storage
>>>>>>>>>>>>> domains.
>>>>>>>>>>>>> At the top level you don't have a status field, but under data
>>>>>>>>>>>>> center level, where it's valid then you get the status property.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Same should go for networks.
>>>>>>>>>>>>> The status property should be added only where it's valid, in
>>>>>>>>>>>>> this
>>>>>>>>>>>>> case the cluster level sub-collection
>>>>>>>>>>>
>>>>>>>>>>> so sounds like we want to declare these properties deprecated to be
>>>>>>>>>>> able
>>>>>>>>>>> to remove them in a future version?
>>>>>>>>>
>>>>>>>>> I guess so,
>>>>>>>>> The question is, are there other location where the status property
>>>>>>>>> (or any other property) exists at an irrelevant level. Unless we
>>>>>>>>> want to go into the effort of mapping them all now we probably need
>>>>>>>>> to define a concept and anything not complying to that is a bug that
>>>>>>>>> is allowed to be fixed without considering it as breaking the API.
>>>>>>>>>
>>>>>>>>> Thoughts?
>>>>>>>>>
>>>>>>> +1 
>>>>>>> I agree that this is a bug and I DO suggest we  go into the effort of reviewing the other objects as well.
>>>>>>> This is too major to just fix this one, and wait until we bump into another one...
>>>>> Mike i see there a general consensus that this is a bug and the top
>>>>> level entity should be a DC network.
>>>>
>>>> i disagree that <status> should be completely removed, instead as bug fix it
>>>> should contain different members: ATTACHED|UNATTACHED (same concept we using in
>>>> /api/storagedomains/xxx)
>>>
>>> first you agree we should remove the status as it is today as it does
>>> not indicate anything to the user.
>>
>> http://lists.ovirt.org/pipermail/engine-devel/2012-July/002009.html
>>
>>>
>>> second you suggest that we'll add attached unattached status, I don't
>>> see value in it unless you specify the clusters it is attached to as a
>>> sub - collection, I don't see us getting to this anytime soon.
>>
>> exactly on opposite, if network would have /clusters links sub-collection,
>> <status>attached|unattached<status> will not be needed as it obvious by
>> absence or existence of clusters links in sub-collection,
>>
>> the use-case is: when you have N networks in DC and want to find unused one
>> to attach it to cluster.
>>
> 
> I don't see the logic in this use case, you don't attach a network to a
> cluster because it is unused, you attach it because you want to use it,
> having it attached to another cluster does not make much of a difference.
> 
> I don't see the need for the attached/detached property.
> I do think that adding a sub-collection of attached cluster can give
> some value to the user but again not an immediate action.

I'll give you one scenario and I'm sure there are lot more:
delete all unused networks ....

> 
> 
>> (without this <status> you'll have to traverse over all networks against all
>> clusters to find one unused)
>>
>>>
>>> we can always add it later and it does not change the fact that the API
>>
>> the solution is very simple and does not require any resources:
>>
>> 1. to enum NetworkStatus add new (default) member UNATTACHED
>> 2. clients will show UNATTACHED if NetworkStatus == UNATTACHED
>>    or ATTACHED otherwise
>>
>>> changes.
>>>
>>>
>>>>
>>>>> Can you please open a bug / update the existing bug to reflect that.
>>>>>
>>>>> Thanks, Livnat
>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>
>>>
>>
>>
> 
> 


-- 

Michael Pasternak
RedHat, ENG-Virtualization R&D