[Freeipa-users] 2.20 dirsrv memory usage

[Stephen Ingram]

On Wed, Jul 18, 2012 at 12:28 PM, John Dennis <jdennis at redhat.com> wrote:
> On 07/18/2012 02:59 PM, Stephen Ingram wrote:
>>
>> On Wed, Jul 18, 2012 at 6:45 AM, Petr Vobornik <pvoborni at redhat.com>
>> wrote:
>>>
>>> On 07/17/2012 11:43 PM, Stephen Ingram wrote:
>>>
>>> 8><------
>>>
>>>
>>>>>>
>>>>>> I'm beginning to think this is just the Web UI itself instead of 389
>>>>>> although it is really difficult to tell. I've poured over the debug
>>>>>> logs and didn't see anything that caused me concern.
>>>>>>
>>>>>> It's certainly usable, but I just got really spoiled by the
>>>>>> unbelievable quickness of 2.1.3. When your release notes indicate it
>>>>>> should be faster, what are you comparing it to? Maybe this only
>>>>>> happens with upgraded instances and not fresh installs.
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> It is always possible something didn't get upgraded properly but I've
>>>>> done
>>>>> 2.1.3 -> 2.2.0 upgrades and haven't seen this. When we say something is
>>>>> faster we're always referring to the previous version (or versions).
>>>>
>>>>
>>>>
>>>> Maybe I was just lucky with 2.1.3. On a first load it might take some
>>>> time to load the "frame" as I call it. But the data would load almost
>>>> instantaneously from there (certainly no more than 1 s) as you moved
>>>> from page to page. Here, even if I return to the same page, the system
>>>> acts as if the data is begin fetched for the very first time as it is
>>>> no faster than the first load. Maybe that is significant to the
>>>> problem?
>>>
>>>
>>>
>>> I think the culprit is Web UI paging capabilities introduced in 2.2. With
>>> lot of users, responses might grow in size. You can check their size and
>>> duration in browser developers tools. I suggest chrome/chromium - press
>>> F12
>>> and choose 'network' tab.
>>>
>>> This new feature can't be disabled in configuration. To test if the
>>> slowdown
>>> is done by paging you can (at own risk) replace line
>>> /usr/share/ipa/ui/facet.js:538
>>>
>>> that.pagination = spec.pagination === undefined ? true : spec.pagination;
>>>
>>> with:
>>>
>>> that.pagination = false;
>>>
>>> Note: It will break some other parts of the UI - so for testing only.
>>
>>
>> I've made the substitution in the code (was line 507 for me-do I have
>> a different version?). Looking at the time chart in Chrome I see that
>> the bulk of the time is for /ipa/session waiting. Would "waiting" mean
>> waiting for the directory server or memcached?
>
>
> Actually neither, it means waiting for a response from the web server
> (technically it's making an RPC call via HTTP Ajax). The RPC call needs to
> go through the web server, memcached, and typically will invoke one or more
> directory server queries, and run a bunch of Python to massage everything
> before the RPC returns with the result.
>
> It doesn't look like you've got much difference in times between with
> pagination on and pagination off. I don't know the pagination code but I
> suspect it's run after the RPC call returns so the RPC timing is not telling
> us much with respect to that.
>
> Waiting for up to 3 seconds for an RPC call does seem on the high side. Do
> you have a lot of LDAP data?

No. 49 users, 17 hosts, 25 services, 6 DNS zones, only 1 of which has
any significant amount of hosts in it.

> But really, unless we get timing results for each component we're grasping
> at straws :-(

Understood.

Steve