On 6/28/06, Huang Zhen <[EMAIL PROTECTED]> wrote:
Lars Marowsky-Bree wrote:
> On 2006-06-28T09:18:36, Huang Zhen <[EMAIL PROTECTED]> wrote:
>
>
>>You made several mistakes here, I would like to point out them.
>>
>>MISTAKE 1 : Empty group is invalid configuration
>>Current crm-1.0.dtd shows :
>><!ELEMENT group (meta_attributes*, instance_attributes*, primitive*)>
>>So it's your code wrong instead of GUI.
>
>
> Uhm. Now you ain't be helpful either. The configurations the GUI created
> _were_ illegal up to the point where we changed the DTD and the
> code to match.
We decided to something, then we should follow the decision, right?
The decision was to allow empty groups in the CIB by changing the DTD.
That is NOT the same thing as you relying on yet more undocumented PE
behavior, ie. that non-meaninfgul objects are available in the results
from the PE.
The fact is, the DTD had been changed for several days.
But the code of PE hadn't changed according to the new DTD until I pushed it.
So who should be responsible for this broken?
See above. Relying on undocumented behavior is always a mistake.
>>MISTAKE 2 : Empty group is need by GUI.
>>No, it's easy to remove the function of create empty group in GUI.
>>What GUI need do is just remove the item in the type list in new item
>>window.
>>So why we insist that we should have empty group?
>>The answer is clear, our users need it.
>>Even expert like alanr thinks that an empty group should be there reasonbly.
>>And, lars also agrees with this.
>
>
> Please don't misrepresent me. I don't agree with either Alan _or_ Andrew
> ;-)
>
> I think that
>
> - empty groups are harmless, and if we must have them, I have given up
> complaining;
>
> - empty groups are _meaningless_, so the GUI shouldn't create them.
> (Meaningless configuration is redundant.)
>
> So, while I would prefer a better workflow in the GUI which didn't
> require the user to create empty complex objects (ie, constructing at
> least the object as one, or guiding the user through the workflow of
> populating it with at least what he already knows, which certainly will
> be at least _one_ resource, why else would he create a group? The GUI
> certainly sometimes confuses me greatly there), it isn't me writing the
> GUI, so there's no point in resisting a harmless change for too long.
>
> Do note though that this causes rather needless churn in the cluster -
> every single change not only causes the inevitable CIB updates to be
> broadcast, but triggers a rerun of the PE/TE cycle. That is indeed
> wasteful.
>
> One way around it would be to make the changes in one go; another would
> be to flag the CIB updates of the GUI as "don't start the PE yet" (until
> either the admin triggers the PE explicitly or some other event in the
> cluster causes a transition).
>
>
>>MISTAKE 3 : Empty containers are meanless.
>
>
> Well, they are meaningless in the sense of not having any effect on the
> cluster, that is certainly true. (Which is also why they are harmless.)
>
>>From the point of view as a placeholder during configuration, they are
> of course not entirely meaningless, that I'll concede.
>
> I also think that this topic has been beaten to death by everyone and
> needs to be given a rest for a couple of weeks.
>
>
> Sincerely,
> Lars Marowsky-Brée
>
--
Best Regards,
Huang Zhen
Linux-HA, Linux Technology Center, China Systems & Technology Lab
China Development Labs, Beijing Tel: 86-10-82782244 Ext. 2845
Email: [EMAIL PROTECTED]
_______________________________________________________
Linux-HA-Dev: [email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
_______________________________________________________
Linux-HA-Dev: [email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
