[Engine-devel] [RFC] ovirt-engine - vdc_config default options

[Itamar Heim]

On 09/13/2012 11:36 AM, Alon Bar-Lev wrote:
>
>
> ----- Original Message -----
>> From: "Eli Mesika" <emesika at redhat.com>
>> To: "Alon Bar-Lev" <alonbl at redhat.com>
>> Cc: engine-devel at ovirt.org, "Livnat Peer" <lpeer at redhat.com>
>> Sent: Thursday, September 13, 2012 10:46:41 AM
>> Subject: Re: [Engine-devel] [RFC] ovirt-engine - vdc_config default options
>>
>>
>>
>> ----- Original Message -----
>>> From: "Alon Bar-Lev" <alonbl at redhat.com>
>>> To: "Eli Mesika" <emesika at redhat.com>
>>> Cc: engine-devel at ovirt.org, "Livnat Peer" <lpeer at redhat.com>
>>> Sent: Wednesday, September 12, 2012 3:57:13 PM
>>> Subject: Re: [Engine-devel] [RFC] ovirt-engine - vdc_config default
>>> options
>>>
>>>
>>> Hello Eli,
>>>
>>> ----- Original Message -----
>>>> From: "Eli Mesika" <emesika at redhat.com>
>>>> To: "Alon Bar-Lev" <alonbl at redhat.com>
>>>> Cc: engine-devel at ovirt.org, "Livnat Peer" <lpeer at redhat.com>
>>>> Sent: Wednesday, September 12, 2012 3:48:00 PM
>>>> Subject: Re: [Engine-devel] [RFC] ovirt-engine - vdc_config
>>>> default
>>>> options
>>>>
>>>>>>>>> QUESTIONS
>>>>>>>>>
>>>>>>>>> 1. Why do we store default values in database?
>>>>
>>>> I wanted to add few important points
>>>> a) We have default values in DB in order to enable overriding
>>>> values
>>>> in the 0000_config.sql file only if a customer did not change the
>>>> default value, if a customer changed the default in some entries,
>>>> we
>>>> want to honour the customer settings
>>>
>>> Default != value.
>>> Default is what should be used if not specified explicitly
>>>
>>> Current implementation puts the default value as a value, so you
>>> cannot distinguish between what user enforced.
>>>
>>> When a default is changed, because of field feedback, we should
>>> push
>>> this into the database instead of keeping in database only options
>>> that were modified by the user and fetch the default from the
>>> option
>>> metadata.
>>>
>>> In another words.... there should be no 000_config.sql, the option
>>> table should be empty as long as the user does not modify any of
>>> the
>>> options.
>>>
>>
>> Lets assume that we are going for it, how would you upgrade from the
>> current state to your suggested solution keeping all user current
>> settings ?
>
> Either by:
>
> 1. leaving all existing data within database... so current users will not enjoy updating defaults in future. I really don't like this solution, but it will work.
>
> 2. during upgrade, go over the values, and remove all that matches the metadata default. This way, future change in metadata will apply.
>
>>
>>>> b)There may be other requirements on configuration that are
>>>> easier
>>>> to
>>>> manage in the database, for example, I have heard that one of the
>>>> suggested features regarding configuration is to keep a kind of
>>>> configuration history and know exactly which configuration values
>>>> was changed, when and by whom.
>>>
>>> There is no conflict. You can have audit table when modifying
>>> options.
>>> Keep in mind that I discuss the option metadata.
>>> I believe you are discussing the option data.
>> You are right
>>
>>>
>>>> c) The version mechanism in the config is working like this :
>>>> there
>>>> is a 'general' version , means that this value is not version
>>>> dependant, or the version can be a real version like 3.1 3.2 etc
>>>> ,
>>>> in such case the value is version dependant and an entry is
>>>> required
>>>> for each version.
>>>
>>> Again, there is no conflict, as the default value within the
>>> metadata
>>> can be version specific too.
>> I agree , however we should consider all changes in current code +
>> tools + upgrade since this seems a major change
>
> Right. I outlined all changes I collected in the initial message.
>
>>>
>>> Having said that, I don't understand how two different versions can
>>> work with the different data models and share one database and
>>> options.
>>>
>>>> I agree however, that the default value in the code is redundant
>>>> and
>>>> error prone and should be removed.
>>>
>>>  From what you wrote above, I don't understand how you reached to
>>> this
>>> conclusion. Can you please explain?
>> I mean that we have now defaults in code (ConfigValues.java) and in
>> the database they may not match.
>> I think as you that only one place defined the defaults, so , it
>> should be removed from code which is less flexible if we want to
>> change something without recompiling ...
>
> If we decide to use java annotations for metadata, the default will be in annotation.
> If we decide to use XML for metadata, the default will be in XML.
> At any case, I would like to reach to a state where the metadata is at one place, and not spread between code, property/xml, database.

I may be missing something.
are you suggesting moving all the vdc_options table to an xml file?
I assume it will be placed under /etc?

then if we update the rpm, the edited config will remain unmodified, and 
we'll get the new one as rpmsave (iirc).
then during upgrade process we'll need to merge these files?