Also forgot to mention, this starts happening at around 1200cps. And using SIPp I accelerate from 500cps to 1200cps very quick. So it might be related to concurrency is what I'm assuming. Thanks.

- Jayesh

On Tue, Sep 22, 2015 at 12:40 PM Jayesh Nambiar <jayesh1017@gmail.com> wrote:
Hi,
Took time to get back with the test results as I was trying to analyze too much myself before getting back on the list. Also I was stuck with weird error of module version mismatch, but I figured and solved that issue.
So I do see the logs "frame size mismatch" when I miss the events. Since the logs are too huge, I am sending it in pastebin. Here it is:
http://pastebin.com/MzTpYsrk

In Kamailio Logs, I see:
DEBUG: evapi [evapi_dispatch.c:361]: evapi_recv_client(): frame size mismatch the ending char (h):[{"event":"REGISTER","tindex":"12603","tlabel":"626010915","PhoneNumber":"42000","D146:"event":"REGISTER","tindex":"3639","tlabel":"1889788252","P] (146)

Whereas the messages in ngrep looks good where the messages are split in two TCP packets. The first packet ends at D and the second packet is the proper continuation which send the rest. But it feels Kamailio skipped the beginning of the message and started parsing at colon again and it thinks the netstring started with D146 which is incorrect. Hence, it never emitted any event for the messages in the first chunk.

Hoping this to be helpful !! Thanks.

- Jayesh


On Mon, Sep 21, 2015 at 4:37 PM Daniel-Constantin Mierla <miconda@gmail.com> wrote:
Hello,

I pushed another commit to add more debug messages to see if the event route is supposed to be executed or not. Can you run the tests again and give again the log messages for missing event route executions?

Cheers,
Daniel


On 21/09/15 09:04, Jayesh Nambiar wrote:
Hi Daniel,
I have captured debug messages after testing with your latest commit on master. Once again, here's what I'm doing exactly for my test:
1) Use SIPp to send OPTIONS towards Kamailio.
2) On getting options send event as follows:
evapi_async_relay("{\"event\":\"REGISTER\",\"tindex\":\"$T(id_index)\",\"tlabel\":\"$T(id_label)\",\"PhoneNumber\":\"$avp(phone_number)\",\"DeviceId\":\"$avp(device_id)\",\"CallId\":\"$ci\"}");
3) There's a client connected which listens for messages on this socket, parses the netstring, and sends same data back as netstring to Kamailio.
4) On the event_route[evapi:message-received, I do the following:
xlog("L_INFO", "GOT [$evapi(msg)] from $evapi(srcaddr):$evapi(srcport)\n");
    if($evapi(msg)=~"REGISTER" && $evapi(msg)=~"tindex") {
      jansson_get_field("$evapi(msg)", "tlabel", "$var(tlabel)");
    jansson_get_field("$evapi(msg)", "tindex", "$var(tindex)");
       $var(t_index) = $(var(tindex){s.int});
    $var(t_label) = $(var(tlabel){s.int});
       t_continue('$var(t_index)', '$var(t_label)', 'REGISTER_RESPONSE');
    exit;
    }
5) On route[REGISTER_RESPONSE], I send 200 OK to SIPp

Now here are detailed logs for which neither SIPp didnt get a response nor the GOT [$evapi(msg)] from $evapi(srcaddr):$evapi(srcport)\n"); got logged in the syslog:

First Event:
DEBUG: evapi [evapi_dispatch.c:598]: evapi_relay(): relaying event data [{"event":"REGISTER","tindex":"26266","tlabel":"587925078","PhoneNumber":"21956","DeviceId":"abcd1234abcd1234","CallId":"23021-12910@198.24.63.39"}] (146)
DEBUG: evapi [evapi_dispatch.c:623]: evapi_relay(): sending [0x7f132588de68] [146:{"event":"REGISTER","tindex":"26266","tlabel":"587925078","PhoneNumber":"21956","DeviceId":"abcd1234abcd1234","CallId":"23021-12910@198.24.63.39"},] (151)

DEBUG: evapi [evapi_dispatch.c:488]: evapi_recv_notify(): received [0x7f132588de68] [146:{"event":"REGISTER","tindex":"26266","tlabel":"587925078","PhoneNumber":"21956","DeviceId":"abcd1234abcd1234","CallId":"23021-12910@198.24.63.39"},] (151)
NOTICE: evapi [evapi_dispatch.c:290]: evapi_recv_client(): {0} [198.24.63.45:48881] - received [146:{"event":"REGISTER","tindex":"26266","tlabel":"587925078","PhoneNumber":"21956","DeviceId":"abcd1234abcd1234","CallId":"23021-12910@198.24.63.39"},] (151) (18)

Second Event:
DEBUG: evapi [evapi_dispatch.c:598]: evapi_relay(): relaying event data [{"event":"REGISTER","tindex":"32244","tlabel":"1637923412","PhoneNumber":"25597","DeviceId":"abcd1234abcd1234","CallId":"27182-12910@198.24.63.39"}] (147)
  DEBUG: evapi [evapi_dispatch.c:623]: evapi_relay(): sending [0x7f1325884410] [147:{"event":"REGISTER","tindex":"32244","tlabel":"1637923412","PhoneNumber":"25597","DeviceId":"abcd1234abcd1234","CallId":"27182-12910@198.24.63.39"},] (152)

  DEBUG: evapi [evapi_dispatch.c:488]: evapi_recv_notify(): received [0x7f1325884410] [147:{"event":"REGISTER","tindex":"32244","tlabel":"1637923412","PhoneNumber":"25597","DeviceId":"abcd1234abcd1234","CallId":"27182-12910@198.24.63.39"},] (152)
  NOTICE: evapi [evapi_dispatch.c:290]: evapi_recv_client(): {0} [198.24.63.45:48881] - received [147:{"event":"REGISTER","tindex":"32244","tlabel":"1637923412","PhoneNumber":"25597","DeviceId":"abcd1234abcd1234","CallId":"27182-12910@198.24.63.39"},] (152) (85)

I had three such events which got missed out of some 25000 odd messages sent from SIPp. Other major difference that I see in the debug logs for problematic events are that there is a positive number in the second parentheses of evapi_recv_client() function. For the events that were invoked successfully the value of second parentheses is 0 for evapi_recv_client() function. Hope the debugging helps.

Debug log for a successful event:
DEBUG: evapi [evapi_dispatch.c:598]: evapi_relay(): relaying event data [{"event":"REGISTER","tindex":"22127","tlabel":"1896682192","PhoneNumber":"73168","DeviceId":"abcd1234abcd1234","CallId":"49957-13056@198.24.63.39"}] (147)
  DEBUG: evapi [evapi_dispatch.c:623]: evapi_relay(): sending [0x7f132568e850] [147:{"event":"REGISTER","tindex":"22127","tlabel":"1896682192","PhoneNumber":"73168","DeviceId":"abcd1234abcd1234","CallId":"49957-13056@198.24.63.39"},] (152)

  DEBUG: evapi [evapi_dispatch.c:488]: evapi_recv_notify(): received [0x7f132568e850] [147:{"event":"REGISTER","tindex":"22127","tlabel":"1896682192","PhoneNumber":"73168","DeviceId":"abcd1234abcd1234","CallId":"49957-13056@198.24.63.39"},] (152)
  NOTICE: evapi [evapi_dispatch.c:290]: evapi_recv_client(): {0} [198.24.63.45:48881] - received [147:{"event":"REGISTER","tindex":"22127","tlabel":"1896682192","PhoneNumber":"73168","DeviceId":"abcd1234abcd1234","CallId":"49957-13056@198.24.63.39"},] (152) (0)

Thanks,

- Jayesh
On Sat, Sep 19, 2015 at 3:55 AM Daniel-Constantin Mierla <miconda@gmail.com> wrote:
I pushed a commit to add more debug message while processing received data. You can use debugger module to set a higher debug level for evapi module in order to see what happens.

I checked the netstring packet size and it is invalid (unless email stripped some white chars there) -- for example in:


145:{"event":"REGISTER","tindex":"55567","tlabel":"627458699","PhoneNumber":"20711","DeviceId":"abcd1234abcd1234","CallId":"21225-3848@5.6.7.8"},

the size is 141, not 145 -- it looks like the size includes the size itself plus the delimiters ':,'. The size is only the effective data, see:

https://en.wikipedia.org/wiki/Netstring

If evapi gets a packet with an invalid size, then it discards the buffer content.

See if the app on the other side of evapi connection builds netstrings with wrong size.

Cheers,
Daniel


On 18/09/15 23:19, Jayesh Nambiar wrote:
Here are the tests that I did:

With the patch applied, I see that Kamailio is invoking event individually for each netstring even when they come in different chunks. But I did see instances where when there were complete netstrings in a single chunk; kamailio did not raise an event for them. Here's the illustration:

146:{"event":"REGISTER","tindex":"16916","tlabel":"1753048277","PhoneNumber":"20708","DeviceId":"abcd1234abcd1234","CallId":"21220-3848@5.6.7.8"},144:{"event":"REGISTER","tindex":"7954","tlabel":"254315075","PhoneNumber":"20709","DeviceId":"abcd1234abcd1234","CallId":"21223-3848@5.6.7.8"},145:{"event":"REGISTER","tindex":"64529","tlabel":"599481568","PhoneNumber":"20709","DeviceId":"abcd1234abcd1234","CallId":"21222-3848@5.6.7.8"},145:{"event":"REGISTER","tindex":"46605","tlabel":"112015324","PhoneNumber":"20710","DeviceId"                                                                                                                                                                                                        

  :"abcd1234abcd1234","CallId":"21224-3848@5.6.7.8"},

The above two chunks contain 4 proper netstrings where the second chunk contains part of the 4th netstring. In this case Evapi properly raised 4 individual events.

  145:{"event":"REGISTER","tindex":"55567","tlabel":"627458699","PhoneNumber":"20711","DeviceId":"abcd1234abcd1234","CallId":"21225-3848@5.6.7.8"},143:{"event":"REGISTER","tindex":"28682","tlabel":"9676691","PhoneNumber":"20712","DeviceId":"abcd1234abcd1234","CallId":"21226-3848@5.6.7.8"},

This above chunk contains two complete netstrings but Kamailio never raised events for these two netstrings.

For the events that were not raised I see proper Kamailio Logs which is:
evapi_recv_client(): {0} [1.2.3.4:48873] - received [145:{"event":"REGISTER","tindex":"55567","tlabel":"627458699","PhoneNumber":"20711","DeviceId":"abcd1234abcd1234","CallId":"21225-3848@5.6.7.8"},143:{"event":"REGISTER","tindex":"28682","tlabel":"9676691","PhoneNumber":"20712","DeviceId":"abcd1234abcd1234","CallId":"21226-3848@5.6.7.8"},]

But I don't see the logs that I've written in the script after the event was raised, which means Kamailio did not invoke events for these two netstrings.

So in my tests, out of 27232 messages sent, there were approximately 27 messages for which the events were not raised by Evapi. The rate of messages started at 500cps and I stopped after I saw missed events at around 1200cps.

Thanks, and do let me know for any further tests or information required about the same.

- Jayesh



On Fri, Sep 18, 2015 at 2:55 PM, Jayesh Nambiar <jayesh1017@gmail.com> wrote:
Will have this tested by tomorrow and will get back to you. Thanks.

- Jayesh

On Fri, Sep 18, 2015 at 1:41 PM, Daniel-Constantin Mierla <miconda@gmail.com> wrote:
Hello,

I just pushed a patch to master branch that should cope with partial data received on tcp connection. No time to test at all, therefore any feedback will be appreciated.

Cheers,
Daniel


On 15/09/15 14:52, Daniel-Constantin Mierla wrote:
Hello,

I will look if there are options in libev to buffer data or try to implement a buffering mechanism locally for such cases.

Cheers,
Daniel

On 14/09/15 23:00, Jayesh Nambiar wrote:
Hello Daniel,
After further testing with evapi module, I figured that when Netstrings are used, an event route is invoked individually for each message even if if multiple netstring messages are received in a single TCP packet. But this doesn't work effectively when a single proper message is split-up in two packets. For Example, if a message arrives as:
12:Hello World!, 12:Hello World!, 12:Hello World! in a single packet, kamailio properly invokes the event route "evapi:message-received" thrice for every individual proper netstring message. 
But if the first packet contains:
12:Hello World!, 12:Hello
And Second Packet contains:
 World!, 12:Hello World!
the event route is invoked only once !!

The above pattern is very much possible while sending and receiving packets over TCP Socket. Our tests for receiving an approximately 150 byte message over evapi socket at the rate of roughly 1000cps causes a lot of real events to be missed because of the above problem. You can never be sure when TCP will split messages in different chunks.
This definitely looks like a bug which makes it not very reliable at large scale deployments. Would really appreciate your inputs on this. Thanks;

- Jayesh



On Thu, Sep 10, 2015 at 4:01 PM, Jayesh Nambiar <jayesh1017@gmail.com> wrote:
Hi Daniel,
Thanks for the quick response. So if I do not use Netstrings, does Kamailio allow me to create a custom logic in the script. For eg. if I decide to use newline as a delimiter, can I keep buffering the message until I encounter the delimiter from the event route and then execute whatever I have to within the script??

Thanks,

- Jayesh

On Thu, Sep 10, 2015 at 1:29 PM, Daniel-Constantin Mierla <miconda@gmail.com> wrote:
Hello,

tcb is stream protocol and several messages can be queued on the pipe at the same time. That is the reason for netstring format, to be able to easily detect the boundaries of each message. If netstring format is enabled and kamailio receives several messages at once, it splits them and for each is executing the event route.

If netstring format is not used, the kamailio is executing the event route with the entire content that was read at once from the tcp connection.

Cheers,
Daniel


On 09/09/15 22:01, Jayesh Nambiar wrote:
Hello,
I'm exploring the evapi module for my kamailio to interface with an external node.js app for third party stuff like AAA, billing engine tasks, notifications and so on. I followed and took some ideas from the rtjson and evapi tutorial found here(http://kb.asipto.com/kamailio:k43-async-sip-routing-nodejs) to build the node.js app consuming events.
When I stress tested the scenario using SIPp and tried sending a lot of events at 300-350cps from Kamailio, I noticed that at times the client is receiving 2-3 events in a single message together although I do event_sync_relay once per SIP message received and have netstrings enabled. I believe this is a typical behavior of TCP and needs to be handled by the client using some kind of Netstring handler. Please correct me if I'm wrong.
And hence I'd like to know what particularly needs to be taken care of while writing a client that is listening for events on raw tcp socket and how does kamailio handle this situation while receiving messages over TCP socket?? Does kamailio recognize the end of netstring properly on evapi:message-received and give exactly one message to take care of on every "message-received" event or should that be handled in the script somewhere !!
I also referred cgrates client over evapi example which is written in GO, but I couldnt find them handling TCP streams clearly either.
I'd really appreciate some expert suggestion here to make an informed decision on using the evapi module for a large scale solution.

Thanks,

- Jayesh




_______________________________________________
SIP Express Router (SER) and Kamailio (OpenSER) - sr-users mailing list
sr-users@lists.sip-router.org
http://lists.sip-router.org/cgi-bin/mailman/listinfo/sr-users

-- 
Daniel-Constantin Mierla
http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
Book: SIP Routing With Kamailio - http://www.asipto.com
Kamailio Advanced Training, Sep 28-30, 2015, in Berlin - http://asipto.com/u/kat

_______________________________________________
SIP Express Router (SER) and Kamailio (OpenSER) - sr-users mailing list
sr-users@lists.sip-router.org
http://lists.sip-router.org/cgi-bin/mailman/listinfo/sr-users




-- 
Daniel-Constantin Mierla
http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
Book: SIP Routing With Kamailio - http://www.asipto.com
Kamailio Advanced Training, Sep 28-30, 2015, in Berlin - http://asipto.com/u/kat

-- 
Daniel-Constantin Mierla
http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
Book: SIP Routing With Kamailio - http://www.asipto.com
Kamailio Advanced Training, Sep 28-30, 2015, in Berlin - http://asipto.com/u/kat



-- 
Daniel-Constantin Mierla
http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
Book: SIP Routing With Kamailio - http://www.asipto.com
Kamailio Advanced Training, Sep 28-30, 2015, in Berlin - http://asipto.com/u/kat

-- 
Daniel-Constantin Mierla
http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda
Book: SIP Routing With Kamailio - http://www.asipto.com
Kamailio Advanced Training, Sep 28-30, 2015, in Berlin - http://asipto.com/u/kat