Raphael,
I can reliably reproduce the problem by running a looping xmlrpc client calling
calls() once a second, while putting more than 1024 participants into a
conference. With each new call over xmlrpc, a new connection is created.
Eventually the socket number that gets accepted is greater than 1024, beyond
the
limit of fd_set.
I looked at XmlRpcDispatch.cpp at the tip of 1.4 and it still looks to be
implemented using select(). But maybe I am not seeing the right version.
Yesterday I rewrote XmlRpcDispatch::waitForAndProcessEvents() using poll(). I
reran the test and it does not seg fault anymore. I am glad to provide these
changes if you like.
Thanks,
Brent
________________________________
On 31.03.11 12:21, Raphael Coeffic wrote:
>>> Am Wed, 30 Mar 2011 11:50:05 -0700 (PDT)
>>> schrieb B Kaser<[email protected]>:
>>>
>>>> Hello,
>>>>
>>>> We are developing a conferencing application on SEMS 1.4.0. It is
>>>> generally very stable, but under load I am getting a core dump in
>>>> xmlrpc2di when periodically entering commands, in this case calls().
>>>> I have seen the same thing on 1.3. Is it possible this WorkerThread
>>>> instance is getting deleted? Thanks.
>>>>
>>
>>
>> Could you please describe us the steps to reproduce the issue a little
>> bit more?
>>
>And maybe retry with the head of 1.4 branch from git?
>-Raphael.
_______________________________________________
Sems mailing list
[email protected]
http://lists.iptel.org/mailman/listinfo/sems
