Summary:ASTERISK-03006: issuing reload to add more iax2 peers causes all iax2 activity to die
Reporter:jolan (jolan)Labels:
Date Opened:2004-12-15 14:32:14.000-0600Date Closed:2004-12-23 15:27:40.000-0600
Versions:Frequency of
Environment:Attachments:( 0) asterisk_trace.txt
Description:i added 2 iax2 peers to iax.conf on our primary server and issued a reload to asterisk.  suddenly all of the servers connected to the primary server showed the primary as being unreachable. the console was a little bit unresponsive, but sip calls appeared to still be making progress.

the primary server had about 10 IAX2 channels in use when i issued the reload so i may think this is a locking issue.

the very last bit of the gdb trace (attached) seems to confirm this:

Thread 20 (Thread -165676112 (LWP 20167)):
#0  0x0010b7a2 in _dl_sysinfo_int80 () from /lib/ld-linux.so.2
#1  0x002e9eee in __lll_mutex_lock_wait () from /lib/tls/libpthread.so.0
#2  0x002e6e09 in _L_mutex_lock_156 () from /lib/tls/libpthread.so.0
#3  0xf6600010 in ?? ()
#4  0x00000674 in ?? ()
ASTERISK-1  0x00fe3100 in ?? () from /usr/local/lib/asterisk/modules/chan_iax2.so
ASTERISK-2  0x00000001 in ?? ()
ASTERISK-3  0xf66a40f0 in ?? ()
ASTERISK-4  0xf61fc818 in ?? ()
ASTERISK-5  0x00fc3931 in new_iax (sin=0xfe47e4, lockpeer=3055342) at lock.h:210
Previous frame identical to this frame (corrupt stack?)
#0  0x0010b7a2 in _dl_sysinfo_int80 () from /lib/ld-linux.so.2

the server platform is Fedora Core 2.

please let me know if you need any more details as i will happily provide them.

this happened once before in the same situation but i didn't obtain a trace at that time.
Comments:By: Mark Spencer (markster) 2004-12-15 14:44:51.000-0600

Can you please duplicate the problem on CVS head and also build with "make clean ; make valgrind" so that the backtrace will be meaningful.  Thanks.

By: Mark Spencer (markster) 2004-12-19 18:55:17.000-0600

Okay one more time before we close this out for lack of interest.  Please try to duplicate the problem on CVS and built wiht "make clean ; make valgrind" so the backtrace is meaningful.  Thanks!

By: jolan (jolan) 2004-12-19 18:57:34.000-0600

i will get to this either monday or tuesday night.

By: jolan (jolan) 2004-12-23 01:12:41.000-0600

this doesn't occur w/cvs.

By: Mark Spencer (markster) 2004-12-23 06:00:13.000-0600

Okay looks like it's already fixed in CVS.

By: Russell Bryant (russell) 2004-12-23 15:27:40.000-0600

Ok, I just found a locking fix that went in in November into CVS head that I missed.  Hopefully that will fix your problem.  If you still have the problem after using the latest code from the stable branch, feel free to open a new bug.

If you provide another backtrace, be sure to do it with "make valgrind".