[Freeipa-devel] [PATCHES] 143-147 Improve performance with large groups

[Martin Kosek]

On 07/18/2013 04:53 PM, Jan Cholasta wrote:
> On 15.7.2013 17:21, Martin Kosek wrote:
>> On 07/15/2013 03:43 PM, Jan Cholasta wrote:
>>> On 15.7.2013 15:16, Martin Kosek wrote:
>>>> On 07/11/2013 12:15 PM, Alexander Bokovoy wrote:
>>>>> On Thu, 11 Jul 2013, Jan Cholasta wrote:
>>>>>> We can add WebUI improvements later. I have some WIP, but I need to discuss
>>>>>> it with Petr first (he's away this week).
>>>>> Ok.
>>>>>
>>>>> The patchset is in ipa-3-2 as well now.
>>>>
>>>> Just checking:
>>>>
>>>> 1) Jan, did you check size of these new indexes on IPA master with such a high
>>>> number of users? How big are they? I want to make sure that this won't create
>>>> an issue on upgrades to new 3.2.x.
>>>
>>> With 10k users, the indices ate roughly 250 MB of disk space.
> 
> Scratch that, my measurement method was stupid. It is just 2 MB of extra space.
> 
>>>
>>>>
>>>> 2) Does the patch set also fix the problem for Web UI? Currently, I think it
>>>> will still grab and process all member attributes even though it does not need
>>>> it. If the Web UI performance is still not sharp, I would rather leave this
>>>> ticket opened and let Jan&Petr cooperate on the Web UI part.
>>>
>>> I agree on keeping the ticket open.
>>>
>>> Honza
>>>
>>
>> Ok, I reopened the ticket. Petr and Jan, please cooperate on this one.
>>
>> Martin
>>
> 
> Added patch which adds new hidden option no_members to suppress membership
> processing for commands of all objects that have member attributes. This can be
> used by the WebUI to prevent member lookups where they are not necessary (as we
> discussed off-line with Martin and Petr).
> 
> Honza
> 

1) Should the new option really have "exclude='webui'? I thought that Web UI is
the main and only consumer of this option.

2) I would clearly state this is an internal option only, e.g.

+ doc=_('INTERNAL: suppress processing of membership attributes.'),

3) It would be nice to state that this option is mutually exclusive with --all,
but given it is internal anyway and there is no central place to define it, we
do not need to do that.

Martin