Summary:ASTERISK-03632: DeadLock related to Zap (seems)
Reporter:paradise (paradise)Labels:
Date Opened:2005-03-05 11:32:41.000-0600Date Closed:2011-06-07 14:10:38
Versions:Frequency of
Environment:Attachments:( 0) backtrace.txt
( 1) backtrace2.zip
( 2) backtrace3.txt
( 3) debug.txt
( 4) debug2.txt
( 5) debug3.txt
( 6) verbose.txt
( 7) verbose2.txt
( 8) verbose3.txt
Description:I get these deadlocks 2-3 times per day.
i think that the deadlock is related to Zap Channels because when i got the following message that's the time my * is deadlock:
Mar  5 09:28:53 WARNING[2457]: Avoided deadlock for 'Zap/6-1', 10 retries!
Mar  5 09:48:47 WARNING[2457]: Avoided deadlock for 'Zap/29-1', 10 retries!


i'm sure that:
- * compiled with make valgrind
- verbose 6
- debug 6
in gdb at deadlock time:
- info thread
- thread apply all bt
Comments:By: nick (nick) 2005-03-05 12:34:01.000-0600

So Asterisk actually deadlocks... as in, it's running but not doing anything. In other words, you're getting more behavior than just that message?

What kind of zap cards do you have in your system?

Also, please update to the latest CVS. Yours isn't that old, but it is a requirement when filing a bug report.


By: Mark Spencer (markster) 2005-03-05 15:19:56.000-0600

Nothing in your backtrace suggests a deadlock (they're all in poll), but portions of the backtrace do not have symbols.  If you compiled with -DDEBUG_THREADS you should be able to traverse the global channel list and see what is holding the lock in each case.

By: paradise (paradise) 2005-03-05 15:32:04.000-0600

i will try to get more info on deadlock...

By: paradise (paradise) 2005-03-05 15:35:57.000-0600

just to make sure, all i've done is:
- cd asterisk
- make clean
- uncomment DEBUG_THREADS = -DDEBUG_THREADS in Makefile
- make valgrind
is there anything missing?

By: paradise (paradise) 2005-03-06 04:30:16.000-0600

as my box is running on kernel 2.6.9:
found a potential bug about THREAD_DEBUG or whatever related to it and kernel 2.6.x (i hope that's not my fault) ;-)
so i submitted it as bug ASTERISK-3635

edited on: 03-07-05 20:45

By: Donny Kavanagh (donnyk) 2005-03-07 10:28:03.000-0600

I've also expirenced the 'Avoided Deadlock' message, except mine seem to only occur while using agents & queue's.  As well they occur on the Local device.  Asterisk seems to continue functioning normally however, at least in my case.

By: paradise (paradise) 2005-03-08 02:44:29.000-0600

another deadlock with more information sent.
backtrace file is about 1MB so it's zipped.
i'm sure that * is compiled with make valgrind
if there is still a way to get more information on deadlocks let me know.
now i have at least 2 of these deadlocks a day.

By: paradise (paradise) 2005-03-09 10:48:52.000-0600

Another DeadLock :-(
logs sent.

By: Donny Kavanagh (donnyk) 2005-03-09 22:17:15.000-0600

Is * actually locking up on you, or are you just seeing the above message?

By: paradise (paradise) 2005-03-10 00:29:03.000-0600

for sure it's dead. no incoming or outgoing call is accepted.
as i said when i saw "Avoided deadlock" on Zap channels my box is dead.
i also see the same message for SIP channels but my box don't lock on it.

By: Mark Spencer (markster) 2005-03-10 01:46:27.000-0600

The "backtrace 3" suggests that there is an issue related to manager.  However the line number does not correspond to unpatched CVS head.  Either redo it with unpatched CVS head, or find me on IRC so I can login and look at your core in real time.

By: Brian West (bkw918) 2005-03-14 23:44:19.000-0600

Any update?

By: Donny Kavanagh (donnyk) 2005-03-15 00:16:47.000-0600

bkw, if it means anything i can reproduce the warning message on a consistant basis and it occurs on the local channel.  My system does not deadlock however, just the Avoided Deadlock message.

By: paradise (paradise) 2005-03-15 08:57:15.000-0600

my deadlocks was on stable version.
now i'm switching to CVS HEAD as mark said to see if deadlocks occur again.

By: Mark Spencer (markster) 2005-03-15 14:23:35.000-0600

If it deadlocks again, please get a new backtrace.  Please do both a "thread apply all bt" and a "thread apply all bt full".  The shorter version makes it easier to spot likely candidate threads.  Alternatively you can generate a core by attaching gdb and doing "gcore <filename>" and then find me on IRC so I can actually try to debug your core file.

By: paradise (paradise) 2005-03-20 04:57:36.000-0600

yesterday i switched to CVS HEAD (03/18/2005) but after a while more than 50% of my sip clients lost their registrations!! :-( so i restart the sip clients which are (Snom, Xpro, Eyebeam, Grandstream, and AudioCodes) but again after a while (about 5-10 minutes) most of them lost their registrations.

so i had to switch back to latest CVS Stable, and the problem solved!
now, i'm going to report this issue as a new bug after testing it again with latest CVS HEAD.

any comments?

edited on: 03-20-05 05:23

btw, no issue in very latest CVS HEAD

edited on: 03-22-05 01:47

By: Mark Spencer (markster) 2005-03-26 11:48:39.000-0600

Okay marking as suspended in the absense of any further information.  If it happens again, please create a core with:

gdb ./asterisk <pid of asterisk>
(gdb) gcore mycorefile
(gdb) quit

and then i can login and diagnose.