Hi Joegen
In more genereical way I've found that we have a problem with
uncorrectly closed socket from UA, this can be seen with an unfinished
sip stack that ends prematurely and with some softphone that crash or
(like linphone) allow to change the transport protocol on fly.
Using many different softphone make our server behave as I described,
with this patch seems that things go better.
I'm still testing so this aren't final results, what I really like to
know is your opinion about the validity of the approach, basically I
think that check if socket is broken before read or write on it seems
to be more safe way of manage.
Do you agree ?
On Tue, Nov 13, 2012 at 4:09 PM, Joegen Baclor <jbac...@ezuce.com> wrote:
> Domenico,
>
> Thanks for the patch. Just clarifying, this patch is for the behavior you
> specified in the August 3 post? If I'm correct, All I need to do to
> reproduce is send an INVITE using TCP, on receipt of 183, close the socket.
>
> -j
>
>
> On 11/13/2012 10:53 PM, Domenico Chierico wrote:
>
> Just to simplify tests here is the patch
>
> On Tue, Nov 13, 2012 at 3:14 PM, Domenico Chierico
> <domenico.chier...@sip2ser.it> wrote:
>
> Hi
> We have 1 sipxecs 4.4 with 50 users installed on kvm based virtual machine.
> We had the proxy that ran over 290% of cpu with an average cpu load
> close to 95%. Applying the review #22, the stuff start goes better and
> we are now close to 40% of cpu load.
>
> Some of this load come from the known SUBSCRIBE issue, but some others
> come from a strange behaviour of the tcp part of the sip stack that we
> found:
>
> - linphone client increases the load on sipXproxy, with his own
> strange keepalive method ("Jak" msg to the proxy) and switching the
> transport from tcp to udp.
>
> - Some other evidences come from my personal tests as I notify on 3 of
> August on dev-ml.
>
> Now I'm testing a solution that seems to work, but I wish to know your
> opinion. I've change the order of "if" statements into SipClient::run
> and I moved the branch about POLLERR and POLLHUP as first.
>
> On Fri, Aug 3, 2012 at 11:43 AM, Domenico Chierico
> <domenico.chier...@sip2ser.it> wrote:
>
> I'm just playing around with go(lang), and this days I was starting
> with sip stack implementation, just when messages starts float around
> I'd realize that I've written a DOS for proxy ..
> I just send INVITE to the proxy than reads for 100 and 180 and so I
> close the socket, at this point I got this into the logs forever:
>
> "2012-08-03T09:31:03.817653Z":43810:SIP:DEBUG:testpbx.labsip2ser.net:SipClientTcp-30:22CEF700:SipXProxy:"SipClient[SipClientTcp-30]::run
> resPoll= 1 revents: fd[0]= 0 fd[1]= 1d"
> "2012-08-03T09:31:03.817668Z":43811:KERNEL:DEBUG:testpbx.labsip2ser.net:SipClientTcp-30:22CEF700:SipXProxy:"OsSocket::isReadyToWrite
> poll returned 1 in socket: 21 0x7f5eec002070"
> "2012-08-03T09:31:03.817683Z":43812:SIP:DEBUG:testpbx.labsip2ser.net:SipClientTcp-30:22CEF700:SipXProxy:"SipClient[SipClientTcp-30]::run
> resPoll= 1 revents: fd[0]= 0 fd[1]= 1d"
> "2012-08-03T09:31:03.817698Z":43813:KERNEL:DEBUG:testpbx.labsip2ser.net:SipClientTcp-30:22CEF700:SipXProxy:"OsSocket::isReadyToWrite
> poll returned 1 in socket: 21 0x7f5eec002070"
> "2012-08-03T09:31:03.817714Z":43814:SIP:DEBUG:testpbx.labsip2ser.net:SipClientTcp-30:22CEF700:SipXProxy:"SipClient[SipClientTcp-30]::run
> resPoll= 1 revents: fd[0]= 0 fd[1]= 1d"
> "2012-08-03T09:31:03.817728Z":43815:KERNEL:DEBUG:testpbx.labsip2ser.net:SipClientTcp-30:22CEF700:SipXProxy:"OsSocket::isReadyToWrite
> poll returned 1 in socket: 21 0x7f5eec002070"
>
> I hope this helps..
>
> bye
> Domenico Chierico
>
>
>
> _______________________________________________
> sipx-users mailing list
> sipx-users@list.sipfoundry.org
> List Archive: http://list.sipfoundry.org/archive/sipx-users/
>
>
_______________________________________________
sipx-users mailing list
sipx-users@list.sipfoundry.org
List Archive: http://list.sipfoundry.org/archive/sipx-users/
