Ah so, this makes sense. I guess you just need to adjust and fine tune some TCP parameters, take a look at this,
www.opensips.org/Documentation/Script-CoreParameters-1-8#toc75 [2] Important flags to optimize are, tcp_connect_timeout tcp_send_timeout tcp_no_new_conn_flag tcp_keepxxxx Also as you mentioned, the IP in error message is different and unreachable, which leads me to question, which node is handing NAT? i guess it would be opensips (since it seems to be edge proxy in your setup, comparable to P-CSCF in IMS), so you need to make sure you handle the NAT correctly on it (with NAT keepalive etc. setup). Thank you. On 2013-11-15 14:59, Gavin Murphy wrote: > Yes, we are bridging between TCP and UDP. There are no custom scripts being used, just the core OpenSIPS scripting, with OpenSIPS acting just as a proxy. We had another event occur last evening and managed to capture some interesting information that suggests the problem is related to blocking tcp connections: > > aaa@ip-10-72-7-129:/dir$ grep "ERROR:core:tcp_blocking_connect: timeout 10 s elapsed from 10 s" opensips.log | sed 's/:.. ip.*//'| uniq -c > 2 Nov 14 06:50 > 2 Nov 14 06:53 > 2 Nov 14 06:55 > 1 Nov 14 06:56 > 2 Nov 14 06:57 > 2 Nov 14 07:00 > .... > 2 Nov 14 13:53 > 3 Nov 14 13:54 > 2 Nov 14 13:55 > 23 Nov 14 13:56 > 29 Nov 14 13:57 > 30 Nov 14 13:58 > 30 Nov 14 13:59 > > While I'm not familiar with the inner workings of OpenSIPS, it seems like there is something happening that may be causing all of the child processes to block while trying to establish outbound TCP connections. Unfortunately the "tcp_blocking_connect: timeout" error (and the subsequent "tcp_blocking_connect failed" error) doesn't indicate the address that caused the failure, and there's no way that I can see to correlate the error back to a request being processed. But I suspect it's a case where a client connected via TCP, registered with our registrar (where the path to the client, including the address of the TCP connection endpoint on OpenSIPS that has an ephemeral port, is recorded), but when we go to send a bunch of requests to that client it is no longer there and it causes all of the opensips child processes to effectively block trying to establish a connection to an address it will never be able to connect to. In the end 477 Send Failed is returned in those cases, but I think the child trying to establish the connection is blocked until the connect times out, and things start piling up behind it, leading to a vicious circle. > > Assuming all of the analysis is correct, can we prevent OpenSIPS from spending time trying to establish a connection that can never be established? > > Thanks, > > Gavin > > On 15/11/2013 7:00 AM, [email protected] [1] wrote: > >> Seems opensips is acting as transport bridge between udp and tcp, so, >> do the SIP replies actually arrive at same interface (transport + ip + >> port) from where the request was sent, and opensips is listening to that >> interface? Can you share SIP trace with us? >> >> Also, are you using any custom application, e.g. perl / lua script etc. >> in opensips dial plan, that might be blocking the opensips thread that >> is managing this transaction. As far as i remember opensips 1.8.x does >> not have async processing, therefore, if any script or module command >> that consumes time e.g. custom sql query, external script or shell >> command etc. would block opensips and any responses received during that >> time are likely to be ignored. >> >> Thank you. >> >> On 2013-11-14 16:05, Gavin Murphy wrote: >> >>> Hi all, >>> >>> We're seeing a possible issue in OpenSIPS related to the timely >>> processing of replies. For example, OpenSIPS receives a REGISTER and >>> passes it on to our application server (which is the registrar). The >>> registrar receives it within the same second and generates the >>> response (a 401 initially) within 0.003 seconds. However OpenSIPS >>> doesn't appear to receive that message, and half a second after the >>> first REGISTER is sent, there is a retransmission after 0.5 second, >>> which is again received by our registrar and a response is generated >>> very quickly. But still the 401 doesn't get received and/or processed >>> by OpenSIPS. The retransmission happens again a few times as per RFC >>> 3261. Eventually, almost 30 seconds later OpenSIPS logs that the >>> reply >>> to the REGISTER has been received. >>> >>> Based on the evidence it seems that there is no problem with OpenSIPS >>> sending the request and it being received by our registrar. There >>> also >>> doesn't appear to be any issues with the retransmission by OpenSIPS >>> at >>> the right intervals, nor does the registrar appear to be introducing >>> any delays. There is not much other traffic going on at the same >>> time, >>> but there are other REGISTERs that are getting the same result. >>> >>> Here are some logs from OpenSIPS: >>> Initial relay to our registrar: >>> Nov 1 19:14:40 ip-10-72-7-129 rcs-opensips[13839]: RCS INFO: >>> ROUTE[2] (Inbound and Outbound) - REGISTER REQUEST relayed. Exiting. >>> Nov 1 19:14:40 ip-10-72-7-129 rcs-opensips[13839]: DBG:tm:t_unref: >>> UNREF_UNSAFE: [0x7f8f8b8d6518] after is 0 >>> Nov 1 19:14:40 ip-10-72-7-129 rcs-opensips[13839]: >>> DBG:core:destroy_avp_list: destroying list (nil) >>> >>> Retransmissions: >>> Nov 1 19:14:40 ip-10-72-7-129 rcs-opensips[13836]: >>> DBG:tm:retransmission_handler: retransmission_handler : request >>> resending (t=0x7f8f8b8d6518, REGISTER ... ) >>> Nov 1 19:14:41 ip-10-72-7-129 rcs-opensips[13836]: >>> DBG:tm:retransmission_handler: retransmission_handler : request >>> resending (t=0x7f8f8b8d6518, REGISTER ... ) >>> Nov 1 19:14:43 ip-10-72-7-129 rcs-opensips[13836]: >>> DBG:tm:retransmission_handler: retransmission_handler : request >>> resending (t=0x7f8f8b8d6518, REGISTER ... ) >>> Nov 1 19:14:53 ip-10-72-7-129 rcs-opensips[13836]: >>> DBG:tm:retransmission_handler: retransmission_handler : request >>> resending (t=0x7f8f8b8d6518, REGISTER ... ) >>> Nov 1 19:15:04 ip-10-72-7-129 rcs-opensips[13836]: >>> DBG:tm:retransmission_handler: retransmission_handler : request >>> resending (t=0x7f8f8b8d6518, REGISTER ... ) >>> >>> Reply "received" (IP,s TNs, and domains anonymized): >>> Nov 1 19:15:08 ip-10-72-7-129 rcs-opensips[13831]: >>> DBG:tm:t_reply_matching: REF_UNSAFE:[0x7f8f8b8d6518] after is 1 >>> Nov 1 19:15:08 ip-10-72-7-129 rcs-opensips[13831]: >>> DBG:tm:t_reply_matching: reply matched (T=0x7f8f8b8d6518)! >>> Nov 1 19:15:08 ip-10-72-7-129 rcs-opensips[13831]: DBG:tm:t_check: >>> end=0x7f8f8b8d6518 >>> Nov 1 19:15:08 ip-10-72-7-129 rcs-opensips[13831]: >>> DBG:tm:reply_received: org. status uas=0, uac[0]=0 local=0 >>> is_invite=0) >>> Nov 1 19:15:08 ip-10-72-7-129 rcs-opensips[13831]: RCS INFO: >>> PROCESSING ONREPLY ROUTE[2] (Reply for Inbound Request) >>> Nov 1 19:15:08 ip-10-72-7-129 rcs-opensips[13831]: RCS DEBUG: >>> ONREPLY_ROUTE[1] (Reply for Inbound Request) - Dump Request Info. >>> Nov 1 19:15:08 ip-10-72-7-129 rcs-opensips[13831]: RCS DEBUG: >>> ============= DUMP REPLY ============= >>> Nov 1 19:15:08 ip-10-72-7-129 rcs-opensips[13831]: RCS DEBUG: SIP >>> message buffer:#012SIP/2.0 401 Unauthorized#015#012Via: SIP/2.0/UDP >>> x.x.x.x:6000;branch=z9hG4bKf439.62091482.0;i=c04c3,SIP/2.0/TCP >>> >>> x.x.x.x:40042;branch=z9hG4bK1490773660;received=x.x.x.x;rport=40042#015#012Record-Route: >>> #015#012Call-ID: >>> 6d76ce9e-ed19-9701-1e75-f070eacd3400#015#012From: >>> ;tag=330928025#015#012To: >>> ;tag=Jmuh3YOvEmCA#015#012CSeq: 221716357 >>> REGISTER#015#012Contact: #015#012Require: >>> gruu#015#012WWW-Authenticate: Digest >>> >>> realm="domain.net",nonce="5273fda02d788dccba9301c8e1f68f078f8e95e5",qop="auth",opaque="004533235332435434ffac663e",algorithm=MD5#015#012Content-Length: >>> 0#015#012P-Associated-URI: >>> #015#012P-Preferred-Identity: >>> #015#012P-Access-Network-Info: >>> ADSL;utran-cell-id-3gpp=00000000#015#012Privacy: none#015#012#015#012 >>> >>> On the registrar the logs appear as follows: >>> [2013-Nov-01 19:14:40.185708] [14002] [6-inf] ===RECV==> REGISTER >>> (6d76ce9e-ed19-9701-1e75-f070eacd3400:+xxxxxxxxxx => +xxxxxxxxxx) >>> [2013-Nov-01 19:14:40.188331] [14002] [6-inf] >>> +xxxxxxxxxx) >>> [2013-Nov-01 19:14:40.721011] [14002] [6-inf] ===RECV==> REGISTER >>> (6d76ce9e-ed19-9701-1e75-f070eacd3400:+xxxxxxxxxx => +xxxxxxxxxx) >>> [2013-Nov-01 19:14:40.721162] [14002] [6-inf] >>> +xxxxxxxxxx) >>> [2013-Nov-01 19:14:41.723160] [14002] [6-inf] ===RECV==> REGISTER >>> (6d76ce9e-ed19-9701-1e75-f070eacd3400:+xxxxxxxxxx => +xxxxxxxxxx) >>> [2013-Nov-01 19:14:41.723262] [14002] [6-inf] >>> +xxxxxxxxxx) >>> >>> etc.... >>> >>> This "bogging down" seems to happen on a regular basis, and the only >>> way we are able to resolve it at the moment is to restart OpenSIPS. >>> Until it is restarted all of the registrations are essentially >>> failing. I believe it is v1.8.2 that we are running. >>> >>> Anyone have any thoughts as to what might be happening? >>> >>> Thanks, >>> >>> Gavin > > -- > Gavin Murphy > Vice President & CTO | www.newpace.ca - Real Technology Solutions > (e) [email protected] > (w) +1.902.406.8375 x 1002 > (m) +1.902.401.9445 > (f) +1.902.406.8377 -- Mit freundlichen Grüßen Muhammad Shahzad ----------------------------------- CISCO Rich Media Communication Specialist (CRMCS) CISCO Certified Network Associate (CCNA) Cell: +49 176 99 83 10 85 MSN: [email protected] Email: [email protected] Links: ------ [1] mailto:[email protected] [2] http://www.opensips.org/Documentation/Script-CoreParameters-1-8#toc75
_______________________________________________ Users mailing list [email protected] http://lists.opensips.org/cgi-bin/mailman/listinfo/users
