athena eats my cpu

Hi all

I implemented a cardgame, which makes approximately one call up to the 
client per second in autoplay mode. Everything is fine with one client, 
the server runs for days, no apparent problem. Now, if I move my app 
from Windows to a hefty OpenBSD V4.1) server class machine, multihomed 
Internet connection, RAID, ... and connect 9 more clients I see an 
initial CPU load of say 10%, everything is working seemingly fine, but 
now things get out of hand: CPU usage is increasing steadily up until 
where the server disconnects clients. Different type of reactors show 
the same basic behavior, only the gradient of the increase varies. Linux 
(Ubuntu server, latest version) shows the same symptoms. Windows comes 
in first, with barely an increase in CPU usage detectable.

gc, cProfile to the rescue, I studied printouts for hours and tried to 
detect a pattern, but so far no luck, I am out of my wits.

Has someone out there experienced such problems and is willing to give 
me a hint, I am grateful for the smallest straw.

Thanks, Werner

Jean-Paul Calderone wrote:
> On Thu, 26 Jul 2007 08:22:27 +0200, Werner Thie 
> <[email protected]> wrote:
>> I am suffering from the same problem in athena like Michael Graz 
>> described in his post from the 12.Jan 2007.
>> Jean-Paul asked for an example, Michael provided one and silence since 
>> then.
>>
>> After having received no answer to my own post regarding this issue I 
>> bring up this subject again, since it is observable since at least 
>> nevow 0.7 in the widgets example. (I have no earlier version anymore). 
>> If somebody could provide some pointers where to hook into I might 
>> invest time to fix this.
>>
>> It's very annoying to have to clean the logs for customers from a 
>> gazillion 'Athena transport duplicate message, discarding...' even 
>> when their harmless.
> 
> I can think of a few possible courses of action.
> 
> I could remove the log message.  The duplicates would still occur, of
> course, but no one would know about them.
> 
> Someone else could tackle Twisted ticket #638, allowing you to avoid
> actually recording this event in the log file.
> 
> Someone else could work on improving Athena such that duplicates occur
> only in the case where they really cannot be avoided.  This might result
> in a small enough number of messages so that they wouldn't be a bother
> to your customers, or it might leave the code in a state where the log
> message really isn't conveying _any_ useful information and it could be
> removed.
> 
> I probably only have time to address this in the first way myself.  Any
> contributions towards the other two solutions would be welcome, though.
> 
> Jean-Paul
> 
> _______________________________________________
> Divmod-dev mailing list
> [email protected]
> http://divmod.org/users/mailman.twistd/listinfo/divmod-dev