[asterisk-users] CPU spike

[Dan Cropp]

Running Asterisk 16.17.0

We have an interesting scenario where we see Asterisk CPU usage spike to the point the entire system is maxed out.

There is a specific scenario where we have two ConfBridges and they are connected via a local channel.  Everything is fine here.

Callers <-> ConfBridge A <-> local channel <-> ConfBridge B <-> Callers

We may be recording both ConfBridges for different groups.  Again, all is fine.

Where we run into a problem is when someone decides they need to record a response.  We connect another local channel to the ConfBridge.  All is fine.
Callers <-> ConfBridge A <-> local channel <-> ConfBridge B <-> Callers
                                                                                                          <-> Local Channel

Then, we initiate a record on the other end of the local channel.
Callers <-> ConfBridge A <-> local channel <-> ConfBridge B <-> Callers
                                                                                                          <-> Local Channel (start recording)

Suddenly, the CPU usage spikes to the point the system is starving and gradually gets worse.
Looking at the debug log for Asterisk, we are seeing thousands of messages like this (roughly 8-10 sets every millisecond)
[05/05 10:25:28.541] DEBUG[2981][C-00000038] chan_pjsip.c:  PJSIP/1003-00000007: Indicated Video Update
[05/05 10:25:28.541] DEBUG[2981][C-00000038] chan_pjsip.c:  PJSIP/1003-00000007
[05/05 10:25:28.541] DEBUG[2981][C-00000038] chan_pjsip.c:  PJSIP/1003-00000007: Indicated Video Update
[05/05 10:25:28.541] DEBUG[2981][C-00000038] chan_pjsip.c:  PJSIP/1003-00000007
[05/05 10:25:28.541] DEBUG[2981][C-00000038] chan_pjsip.c:  PJSIP/1003-00000007: Indicated Video Update
[05/05 10:25:28.541] DEBUG[2981][C-00000038] chan_pjsip.c:  PJSIP/1003-00000007
[05/05 10:25:28.541] DEBUG[2981][C-00000038] chan_pjsip.c:  PJSIP/1003-00000007: Indicated Video Update

After this occurs, the messages in the debug output can also be a bit out of order for the date/time listed in the file.

An interesting scenario, if we first start recording on the Local channel, then add it to the ConfBridge this CPU spike does not occur.

We realize this is a bit of an extreme scenario, two ConfBridges connected with a local channel, and needing to connect another local channel to one of the ConfBridges and start recording.  Unfortunately, it's a scenario we are stuck with due to a unique customer requirement.

Any thoughts or suggestions?

Dan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.digium.com/pipermail/asterisk-users/attachments/20210505/dea2da71/attachment.html>