[Iaxclient-devel] Re: Bug info... (really weird bug info)

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 422-6466

More Data:

Watching the trace, the error seems to take place after the timeout for the
transfer attempt.  As Adam pointed out, IAX2 always tries to establish a
native bridge.  Somehow our clients, being both NATed behind the same NAT
device (i.e. both using the same real outside IP address) are unable to
connect natively (the NAT router is not smart enough to "pinwheel" IAX
packets?) so the connection does not work.

Oddly, the IAX protocol seems to default to switching over to a
native-bridged operation UNLESS it receives tacit failure messages from both
sides.

Here's what you see with a IAX2-to-IAX2 call through Iaxtel that does _not_
have any problems with the audio:

    -- IAX2[1111]/8 answered IAX2[iaxtel@69.73.19.178:4569]/7
    -- Stopped music on hold on IAX2[iaxtel@69.73.19.178:4569]/7
    -- Attempting native bridge of IAX2[iaxtel@69.73.19.178:4569]/7 and 
    -- Channel 'IAX2[iaxtel@69.73.19.178:4569]/7' unable to transfer

Note the last line - the chan_iax2 code at Iaxtel detected that a native
session could not be established, so it rejected the attempt and my Asterisk
continued to relay the voice data.

Here's what you see with the IAX client-to-IAX client call:

    -- Accepting unauthenticated call from 64.151.42.28, requested format =
2, actual format = 2
    -- Executing Dial("IAX2[guest@1105]/3", "IAX2/1111|20|m") in new stack
    -- Called 1111
    -- Started music on hold, class 'default', on IAX2[guest@1105]/3
    -- Call accepted by 64.151.42.28 (format GSM)
    -- Format for call is GSM
    -- IAX2[1111]/5 is ringing
    -- IAX2[1111]/5 answered IAX2[guest@1105]/3
    -- Stopped music on hold on IAX2[guest@1105]/3
    -- Attempting native bridge of IAX2[guest@1105]/3 and IAX2[1111]/5
    -- Registered '1111' (AUTHENTICATED) at 64.151.42.28:4569
    -- Registered '1111' (AUTHENTICATED) at 64.151.42.28:4569

You never see either side (properly?) reject the attempted native transfer,
so Asterisk bows out of the call and the audio stops.

Can we set a noreinvite flag for IAX (I know, I should look at the
source...) to prevent the system from event trying to reinvite the audio
point-to-point?

I want to blame the "both clients behind the same NAT screen" factor, but I
have had client-to-client calls loose audio after 1 minute even when users
are calling in from outside.

Thoughts?

Adam, did you have anything like this spring up in Firefly?

Thanks,

Steve