Hello,

I think you have discovered a bug I made using variadic functions.

Very soon I gonna send a patch to correct it.


Thanks,
Vicente.

On 11/28/2013 10:14 AM, Tuan Viet Nguyen wrote:

Hello Vicente,

Thank you for your reply, you'll find my answer below

On Thu, Nov 28, 2013 at 12:03 AM, Vicente Hernando <vhernando@systemonenoc.com> wrote:

Hello,

also full steps to crash kamailio and reproduce the error would be good.

Here is the architecture

A <--> Asterisk <--> Kamailio 1 <---> kamailio2 <--- ISP---> mobile

Kamailio 1 & 2 are connected to a local redis server

1/ I restarted the redis server

2/ From the mobile I made a call to A then cancelled it. In the script of kamailio1, if a call has missed or failed, it sends a message to the redis. And in this case, it crashes

On 11/27/2013 11:35 PM, Daniel-Constantin Mierla wrote:

Hello,

can you give the full output for 'bt full' with gdb on the core file? You gave only partial list of the frames, not being enough to see the execution trace.

Cheers,
Daniel

On 11/27/13 6:52 PM, Tuan Viet Nguyen wrote:

Hello,

I'll try to shut down the redis server to test the behavior of kamailio and it has crashed if a call is received and then cancelled.

1/The kamailio version is 4.0.4

2/ Kamailio log

/usr/local/sbin/kamailio[25333]: ERROR: ndb_redis [redis_client.c:364]: redisc_exec(): Redis error: Server closed the connection
/usr/local/sbin/kamailio[25361]: : <core> [pass_fd.c:293]: receive_fd(): ERROR: receive_fd: EOF on 13
/usr/local/sbin/kamailio[25328]: ALERT: <core> [main.c:788]: handle_sigs(): child process 25333 exited by a signal 11
/usr/local/sbin/kamailio[25328]: ALERT: <core> [main.c:791]: handle_sigs(): core was generated

I assume you disconnect redis server and don't reconnect it. It is that correct?

Then this line is an error but it should recover from that. I probably should set this as a warning instead an error.

/usr/local/sbin/kamailio[25333]: ERROR: ndb_redis [redis_client.c:364]: redisc_exec(): Redis error: Server closed the connection

Yes, it has been restarted

3/ Interesting information in the core
#3  0x00007fc79412893d in redisvCommand (c=0x64657461, format=0x9 <Address 0x9 out of bounds>, ap=0x30, ap@entry=0x7fff0ff56aa8) at hiredis.c:1304
No locals.
#4  0x00007fc794341713 in redisc_exec (srv=srv@entry=0x7fff0ff56be0, res=res@entry=0x7fff0ff56c00, cmd=cmd@entry=0x7fff0ff56bf0) at redis_client.c:368
        rsrv = 0x7fc794565150
        rpl = 0x7fc7946fab70
        c = 0 '\000'
        ap = {{gp_offset = 48, fp_offset = 48, overflow_arg_area = 0x7fff0ff56bb0, reg_save_area = 0x7fff0ff56ac0}}
        __FUNCTION__ = "redisc_exec"
#5  0x00007fc79433b781 in w_redis_cmd5 (msg=<optimized out>, ssrv=<optimized out>, scmd=<optimized out>, sargv1=<optimized out>, sargv2=0x7fc7946f7bf0 "p\243_\224\307\177", sres=0x7fc7946f7c50 " \253_\224\307\177") at ndb_redis_mod.c:250
        s = {{s = 0x7fc7945fb300 "kamailio_redis", len = 14}, {s = 0x7fc7945f5f50 "PUBLISH %s %s", len = 13}, {s = 0x7fc7945fab20 "r", len = 1}}
        arg1 = {s = 0x7fc7945f5f80 "notification", len = 12}
        arg2 = {
          s = 0x7fc794551c60 "info XXX"...,
          len = 212}
        c1 = 0 '\000'
        c2 = 0 '\000'
        __FUNCTION__ = "w_redis_cmd5"

In the source code:

    rpl->rplRedis = redisvCommand(rsrv->ctxRedis, cmd->s, ap );
    if(rpl->rplRedis == NULL)
    {
        /* null reply, reconnect and try again */
        if(rsrv->ctxRedis->err)
        {
            LM_ERR("Redis error: %s\n", rsrv->ctxRedis->errstr);
        }
        if(redisc_reconnect_server(rsrv)==0)
        {
            rpl->rplRedis = redisvCommand(rsrv->ctxRedis, cmd->s, ap);
        }
    }

First redisvCommand executes but returns nothing. Then it shows a redis error.

It tries to reconnect and it manages to connect ?? because it shows no more errors.

And then executes redisvCommand again and crashes.

If server is down it should not be able to connect and so not to execute redisvCommand again.

According to the core, we MUST be in this case if(redisc_reconnect_server(rsrv)==0)

But I am wondering how the first redisvCommand can succeed before the reconnection ? (the connection kamailio1 <-> redis has already been taken down). Does all the redis context always there when we first call redisvCommand? 

 

May be I would get more clues with more information.

Regards,
Vicente.

Thank you

Regards,

 

I've found one of post that this issue has been fixed but it seems that it's always the case ..
http://www.mail-archive.com/search?l=sr-users@lists.sip-router.org&q=subject:%22Re%3A+%5BSR-Users%5D+ndb_redis+module+fails+after+a+while%22

Do you have any idea?

Thank you

_______________________________________________
sr-dev mailing list
sr-dev@lists.sip-router.org
http://lists.sip-router.org/cgi-bin/mailman/listinfo/sr-dev

-- 
Daniel-Constantin Mierla - http://www.asipto.com
http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda

_______________________________________________
sr-dev mailing list
sr-dev@lists.sip-router.org
http://lists.sip-router.org/cgi-bin/mailman/listinfo/sr-dev