[asterisk-bugs] [JIRA] (ASTERISK-29035) chan_local: Multistream support breaks T.38 faxing

[Matthias Hensler (JIRA)]

    [ https://issues.asterisk.org/jira/browse/ASTERISK-29035?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=254244#comment-254244 ] 

Matthias Hensler commented on ASTERISK-29035:
---------------------------------------------

To answer the first question: the instance was running since saturday, so for just shy of 6 days. However we already had 16.16.1 with that patch running from feb 18th (and 16.16.0 from Feb 10th) without issues. Since the diff between 16.16.1 and 16.16.2 is very short (and although related to T38 it is linked to res_pjsip and not chan_sip) I would suspect that the issue is just difficult to reproduce (a lot of days in between).

Unfortunately I cannot check for statistics we had with 16.16.0 and 16.16.1, but for the now deadlocked 16.16.2 we had 112k calls in total since start. In that time we had 321 successful SendFax()-instances and 183 ReceiveFax()-instances. All SendFax() was done via Local-Channel, most ReceiveFax()-instances too. As for how often T38 vs G711a was used I have no details, sorry.

This deadlock seemed to be triggered on a ReceiveFax()-instance, while the deadlock on Jan 14th was on SendFax()

> chan_local: Multistream support breaks T.38 faxing
> --------------------------------------------------
>
>                 Key: ASTERISK-29035
>                 URL: https://issues.asterisk.org/jira/browse/ASTERISK-29035
>             Project: Asterisk
>          Issue Type: Bug
>      Security Level: None
>          Components: Channels/chan_local
>    Affects Versions: 16.12.0, 18.1.0
>         Environment: CentOS 7
>            Reporter: Matthias Hensler
>            Assignee: Joshua C. Colp
>              Labels: fax, patch
>      Target Release: 16.17.0, 18.3.0
>
>         Attachments: ASTERISK-29035.diff, Inbound-16.11.0-LocalChannel.pcap, Inbound-16.12.0-LocalChannel.pcap, Inbound-16.12.0-NoLocal.pcap
>
>
> After updating from 16.11.0 to 16.12.0 all faxes via SendFax() and ReceiveFax()-applications break, when placed on a Local-channel and T.38 is used.
> Our dialplan makes use of Local-channels when dialing internal numbers and is also used for issuing outgoing fax-traffic via an originate-call using the manager-API. With the rework of chan_local in 16.12.0 (I suspect ASTERISK-28938 here) this breaks our outgoing fax-traffic when the destination uses T.38. If G.711a is used the problem does not occur.
> When looking at some pcap-traces I can see, that the destination reinvites to T.38 after a call is established and sends lots of T.38 packets. However Asterisk simple ignores any T.38-packet and does not respond to any of these packets, until the call is then finally dropped and fax-transmission ultimately fails.
> Fortunately the problem is very easy to reproduce with an inbound fax also. For that I used an older test-instance of Asterisk 17.1.0 which dials in to our real Asterisk instance where the call is placed via a `Dial(Local/...)` to the appropriate extension where ReceiveFax() is running.
> When running Asterisk 16.11.0 everything works as expected (see Inbound-16.11.0-LocalChannel.pcap) where the connection is established with T.38 and the fax is received properly.
> When switching to Asterisk 16.12.0 on the receiving end the exact same setup breaks (see Inbound-16.12.0-LocalChannel.pcap). The connection is established with T.38 and packets are send from the test-instance, but no replies are issued by Asterisk 16.12.0 (we are running ReceiveFax() in debug mode and it does not log anything with 16.12.0, so obviously chan_local does not properly relays the T.38-frames).
> Just by removing the `Dial(Local/...)` from the dialplan and jumping directly to the ReceiveFax()-extension makes the setup work again on Asterisk 16.12.0 (see Inbound-16.12.0-NoLocal.pcap). So I am pretty sure that the culprit results from the recent changes to chan_local.
> I know that using local-channels is not the nicest thing to do, but we rely on it to process call redirections on local numbers, while still keeping proper call data records for correct billing.
> If more debug output besides the pcap-traces, or a proper dialplan example is needed, I am happy to provide it of course.



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