On 28.11.22 14:58, Richard Fuchs wrote:
On 28/11/2022 04.14, [EXT] Aditya Gupta wrote:
We have 2 rtpengines in production and thousands of calls coming in. Few of them get timed out because they don't get processed in time by rtpengine. Simply by making 3 commands; offer, answer and delete. It can help us a lot with this.
I think I've understood what needs to be done and, in a day, or two I'll raise a PR with the changes for rtpengine module. After that I guess there are some changes that needs to be done in sipwise/rtpengine media servers which seems complicated at this point.
Do you have a sense as to why rtpengine would be slow to respond? Because as Henning mentioned, we can handle many thousands of calls and haven't seen rtpengine becoming slow to respond, even with HA backing enabled.
I haven't encountered rtpengine being slow in response, unless there is some network problem on the control link.
As I read in this discussion, the timeout is about calls, so at sip layer. The rtpengine control command has 1s timeout by default, which is under the sip timeout, but of course it can accumulate on many simultaneous calls. On the other hand, if rtpengine command times out, just switching to async processing is not solving it, it will still happen with default timeout value, so the invite is not handled by rtpengine after all. In such case the timeout value has to be increased.
I would rather check if it is actually rtpengine by enabling the latency limit via core parameters, it could be another function in config that could result in a sip layer timeout. Or maybe insufficient number of sip workers (children parameter should be increased) to cope properly with the traffic volume.
Cheers, Daniel