Help Debugging A Possible SIP Channel Leak In 11.17.0, Possible Race Condition

Home » Asterisk Users » Help Debugging A Possible SIP Channel Leak In 11.17.0, Possible Race Condition
Asterisk Users 4 Comments

I am trying to collect enough information about an problem a client is having with its asterisk 11.17.0 x86_64. This issue was observed before in 1.8.20, and we upgraded to 11.15.0 and then to 11.17.0 with no solution.

Background: this client is a telemarketing call-center that generates outgoing calls with close to a hundred agents operating simultaneously in peak hours. The system uses asterisk with FreePBX 2.8. In order to generate the calls, I wrote a program that connects to Asterisk using the AMI protocol. This program expects the SIP agent extensions to be assigned as members of queues, of which there are about 20, as shown below:

9007 has 0 calls (max unlimited) in ‘random’ strategy (5s holdtime, 68s talktime), W:0, C:581, A:260, SL:82.6% within 60s
SIP/147 (ringinuse disabled) (dynamic) (On Hold) has taken 21 calls (last was 800 secs ago)
SIP/417 (ringinuse disabled) (dynamic) (In use) has taken 77 calls (last was 708 secs ago)
SIP/416 (ringinuse disabled) (dynamic) (In use) has taken 41 calls (last was 656 secs ago)
SIP/408 (ringinuse disabled) (dynamic) (In use) has taken 50 calls (last was 789 secs ago)
No Callers

The program runs “queue show” through AMI every few seconds. For each queue to be used in telemarketing, the program counts the number of members that are “Not In Use”. If at least one is found, it reads that many phone numbers from the database and uses the AMI Originate command on each one, as follows:

Action: Originate Channel: Local/NNNNNNNNNN@from-internal Exten: CCCC
Context: from-internal Priority: 1
Async: true ActionID: xxx

Here, NNNNNNNNNN is the number read from the database and CCCC is the queue extension in the FreePBX-created context that eventually runs the Queue() dialplan application for the corresponding queue. This causes the call to be connected between the outgoing number and the queue, and is then assigned to a queue member by Asterisk. The dialplan is configured to route NNNNNNNNNN through one of a series of SIP trunks using the outbound routes as configured by FreePBX.

The issue is that although this strategy works correctly on the user’s machine for a few days, we have been observing that eventually the application stops placing calls. The agents are all idle (all 90 to 100 of them), but the “queue show” command shows them to be “In Use” on all queues. Furthermore, in normal operation, the “core show channels” command shows at most one channel for each configured SIP client in the “Up” state, but when calls stop being placed, the same command reports multiple channels in the “Up” state, as follows (after sorting):

Local/9759315789@from-internal-0000a456;1!from-trunk-sip-5547740412!!1!Up!AppDial!(Outgoing Line)!9759315789!!!3!500!(None)!1428426084.169326
SIP/110-00014c2b!EjecutivoROLLRATE!9014!1!Up!AppQueue!(Outgoing Line)!110!!!3!590!(None)!1428425994.169124
SIP/121-00014c24!EjecutivoROLLRATE!9014!1!Up!AppQueue!(Outgoing Line)!121!!!3!596!(None)!1428425988.169111
SIP/122-00014b56!EjecutivoQUADS!93000!1!Up!AppQueue!(Outgoing Line)!122!!!3!677!(None)!1428425906.168693
SIP/131-00014b8c!EjecutivoSEGUNDAS!99000!1!Up!AppQueue!(Outgoing Line)!131!!!3!651!(None)!1428425932.168799

In the example shown above, SIP/119, SIP/123, SIP/131 have three channels, and SIP/130 has five. Even assuming a bug in my program, a single SIP extension cannot be handling five simultaneous calls.

Now, here comes my speculation. I believe what is happening is that somehow, channels are not being cleaned on hangup, but somehow leave the extension capable of accepting a new call. My program sees queues with a mixture of non-affected idle members and affected members (which appear in “queue show” as “In Use” but are actually idle and will accept a new call). As long as there is one non-affected member that can appear as “Not In Use”, my program keeps placing calls on that queue, which are handled by both affected an unaffected members. This would explain the multiple channels for a single SIP phone. As soon as all the members become affected in one queue, the dialing stalls.

Is this speculation sound, or even possible, given the internal architecture of Asterisk?
What tools can be used to debug this? Is this a channel leak? I know of DEBUG_THREADS for debugging deadlocks, but this does not seem like one. What extra information can be provided to make a bug report on this?

4 thoughts on - Help Debugging A Possible SIP Channel Leak In 11.17.0, Possible Race Condition

  • I have seen a similar problem occasionally. We will be doing maintenance on a customer’s server and they will have one or two “ghost” channels on their machine hundreds of hours old but with no call associated with them. So far we have just been rebooting their server or issuing a hangup command to the channels.

  • Have you tried Asterisk 13? The bridging mechanism has been completely rewritten on Asterisk 12, so there’s no longer channel masquerading and zombie channels. Might be worth a try.

    2015-04-07 20:33 GMT-03:00 Alex Villací­s Lasso :