Event Stability

[CEA-SEPIA]

Hello,


I have some problem of stability in my application. I believe that the
problem came with the pulling of event, it's seem that several event
packet are lost for random client.

We made a test with 6 client for 20mn. At use a lot of event has been
lost.

I use ucengine 0.3.0 (tested also on 0.4.1) and the api JS that you
provide.

Can you provide a test of stability for your server and your api JS
for a big amount of client (about 20 clients) with a strong activity
(about 5 event by second and by client) and several host from your
network.


Thanks a lot

[François de Metz]

Ok, It's a big issue. Can you help us to diagnostic you system? Can you
provide your tests?
If confirmed, we will fix this issue urgently.

François

[CEA-SEPIA]

We made two tests for now we aren't sure about the conclusion yet.
We prepare a test procedure to prove it without a doubt.

Test 1 (28/03/2011):
* Conditions:
** UCE: 0.3.0 with patch on roster and file listing
** Brick RED5
** 9 users
** 30minutes

* Results:
** 3504 events on /api/0.3/event/
** 2249 events on /api/0.3/event/demo

Test 2 (5/04/2011):
* Conditions:
** UCE: 0.3.0 with patch on roster and file listing
** Brick RED5
** 5 users
** 20minutes

* Results:
** 2186 events on /api/0.3/event/
** 1573 events on /api/0.3/event/demo

General conclusion:
* A loss of syncronous in several client make think of a loss in event
* A lot more of what is needed is transported by the /event/

>  signature.asc
> < 1 000AfficherTélécharger

[François de Metz]

On 05/04/2011 17:06, CEA-SEPIA wrote:
> We made two tests for now we aren't sure about the conclusion yet.
> We prepare a test procedure to prove it without a doubt.
>
> Test 1 (28/03/2011):
> * Conditions:
> ** UCE: 0.3.0 with patch on roster and file listing
> ** Brick RED5
> ** 9 users
> ** 30minutes
>
> * Results:
> ** 3504 events on /api/0.3/event/
> ** 2249 events on /api/0.3/event/demo
>
> Test 2 (5/04/2011):
> * Conditions:
> ** UCE: 0.3.0 with patch on roster and file listing
> ** Brick RED5
> ** 5 users
> ** 20minutes
>
> * Results:
> ** 2186 events on /api/0.3/event/
> ** 1573 events on /api/0.3/event/demo
>
> General conclusion:
> * A loss of syncronous in several client make think of a loss in event
> * A lot more of what is needed is transported by the /event/

This is normal behaviour. We have two rings of pubsub in U.C.Engine.

- meeting: only events from the meeting
- global: all events from all meetings and global events like user
creation, user connection, ... http://docs.ucengine.org/events.html

This is probably a point not really good documented. We will enhance
this subject quickly!

I hope that this explanation solve your problem.

Fran�ois

>
> On 5 avr, 10:27, Fran�ois de Metz<franc...@2metz.fr> wrote:
>> On 04/04/2011 18:13, CEA-SEPIA wrote:
>>

>>
>>
>>> Hello,
>>
>>> I have some problem of stability in my application. I believe that the
>>> problem came with the pulling of event, it's seem that several event
>>> packet are lost for random client.
>>
>>> We made a test with 6 client for 20mn. At use a lot of event has been
>>> lost.
>>
>>> I use ucengine 0.3.0 (tested also on 0.4.1) and the api JS that you
>>> provide.
>>
>>> Can you provide a test of stability for your server and your api JS
>>> for a big amount of client (about 20 clients) with a strong activity
>>> (about 5 event by second and by client) and several host from your
>>> network.
>>
>> Ok, It's a big issue. Can you help us to diagnostic you system? Can you
>> provide your tests?
>> If confirmed, we will fix this issue urgently.
>>

>> Fran�ois
>>
>> signature.asc
>> < 1 000AfficherT�l�charger

[CEA-SEPIA]

Hello,

Ok. I will dramatically reduce my number of request this way :) The
puller on /event/MEETING and /event/ was from the time were the
internal event was not sended on /event/MEETING :)

Isn't a security issue that if I create an other meeting and look at
my /event/ on a other meeting I obtain the event of the second
meeting ?

That explanation give me some time but I'm pretty sure of the loss of
event so potentially we will have a problem of stability when the
activity on the server will grown up so it's does solve temparary my
problem.

Thanks a lot

> creation, user connection, ...http://docs.ucengine.org/events.html

[Victor Goya]

It could be a security issue if you are not restrictive enough on the
access rights you set. For example, you can restrict the right to get
event from the global entry point (/event/) only for a small subset of
event type (like 'internal.presence.add').

--
Victor Goya

On 04/06/2011 10:59 AM, CEA-SEPIA wrote:
> Hello,
>
> Ok. I will dramatically reduce my number of request this way :) The
> puller on /event/MEETING and /event/ was from the time were the
> internal event was not sended on /event/MEETING :)
>
> Isn't a security issue that if I create an other meeting and look at
> my /event/ on a other meeting I obtain the event of the second
> meeting ?
>
> That explanation give me some time but I'm pretty sure of the loss of
> event so potentially we will have a problem of stability when the
> activity on the server will grown up so it's does solve temparary my
> problem.
>
> Thanks a lot
>

[François de Metz]

On 06/04/2011 10:59, CEA-SEPIA wrote:
> Hello,
>
> Ok. I will dramatically reduce my number of request this way :) The
> puller on /event/MEETING and /event/ was from the time were the
> internal event was not sended on /event/MEETING :)
>
> Isn't a security issue that if I create an other meeting and look at
> my /event/ on a other meeting I obtain the event of the second
> meeting ?

Obviously you can disallow users to poll /event/ like Victor said :).

> That explanation give me some time but I'm pretty sure of the loss of
> event so potentially we will have a problem of stability when the
> activity on the server will grown up so it's does solve temparary my
> problem.

Ok we need to explore the point. I'm waiting for your test case.

François