[asterisk-bugs] [JIRA] (ASTERISK-28864) RTP Timestamp not increasing after several transfers and codec changes

[nappsoft (JIRA)]

    [ https://issues.asterisk.org/jira/browse/ASTERISK-28864?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=250593#comment-250593 ] 

nappsoft commented on ASTERISK-28864:
-------------------------------------

No, there are no jitterbuffers in use (except for if asterisk uses them automaticelly without any configuration)

Yes, we use an opensource version (digium didn't have opus support when we first came across this github repo). However we call it opusnb and limit it to narrowband. It's only in use for mobile clients because of bandwidth limitations.

> RTP Timestamp not increasing after several transfers and codec changes
> ----------------------------------------------------------------------
>
>                 Key: ASTERISK-28864
>                 URL: https://issues.asterisk.org/jira/browse/ASTERISK-28864
>             Project: Asterisk
>          Issue Type: Bug
>      Security Level: None
>          Components: Resources/res_rtp_asterisk
>    Affects Versions: 16.9.0
>            Reporter: nappsoft
>            Assignee: nappsoft
>              Labels: patch
>         Attachments: ast_translate.patch
>
>
> After facing the following issue in production I was able to reproduce it in the lab:
> Our real world procedure:
> - Incoming call (G711a) going to a queue
> - pickup the channel (so not answering an INVITE, but doing a PickupChan) with a different codec (Opus in our case, didn't test with others)
> - placing the call into a queue again with an attended ast-transfer
> - the call is answered (G711a) by an agent (not with a pickup this time), using a PJSIP based softphone
> - the transfer is executed. Now both channels use G711a
> So now audio will no longer need to be transcoded. (The problem does not seem to happen when we use G711a for all clients all the time).
> What now happens is that we could observe one-way-audio or no audio at all. However this was not caused by a network issue but by wrong rtp timestamps: rtp timestamps in the outgoing stream suddenly were not increased anymore.
> I've added some logging and could observe the following behavior: the delivery time used in calc_txstamp was going back in time in res_rtp_asterisk.c. (80000ms each time).
> It seems like the problems were caused by the fact that the timestamps on rx side were lower than before after the transfer. I was able to get rid of the issue by patching translate.c
> Please note that the issue could also be influenced by the fact that we use a slightly patched version of the opus codec. However the fact that asterisk seems to struggle when f->delivery is (for whatever reason) lower than path->nextin is a problem worth looking at I guess.



--
This message was sent by Atlassian JIRA
(v6.2#6252)