896235 - Video freeze on MacOS

Status: RESOLVED DUPLICATE of bug 880879

(Core :: WebRTC: Audio/Video, defect)

Description

[Eric Rescorla (:ekr)]

I am setting regular video freezes on Nightly on Mac.
The regression appears to be somewhere between:

http://hg.mozilla.org/integration/mozilla-inbound/rev/64d9bac71cc9    -- OK   7/1
http://hg.mozilla.org/integration/mozilla-inbound/rev/468b35185c44    -- BAD  7/10

Comment 1

[Eric Rescorla (:ekr)]

More detail here. This looks like the problem was the 3:30 merge.

7/10    137975:468b35185c44    -- BAD
7/10    137871:edd376cd77bf    -- BAD
7/10    137868:db8e3a4bf100    -- OK     
7/5     137523:a9fc927c311b    -- OK
7/1            64d9bac71cc9    -- OK


Repeat testing:

13868:
3 five minute calls OK

137871:
failure before 40s
failure before 10s

Comment 2

[Eric Rescorla (:ekr)]

Platform: Call from MacOS 10.8.3 (Mountain Lion) to MacOS 10.6.8 (Snow Leopard)

Initial freeze is in the ML -> SL direction but we tend to see freezes in both
directions if we run long enough.

Build directory is:
http://ftp.mozilla.org/pub/mozilla.org/firefox/tinderbox-builds/mozilla-inbound-macosx64-debug/1373440439/

Comment 3

[Eric Rescorla (:ekr)]

Attachment: 10.6.8 signaling log

Comment 4

[Eric Rescorla (:ekr)]

Attachment: 10.6.8 WebRTC log

Comment 5

[Eric Rescorla (:ekr)]

Attachment: 10.7.3 WebRTC log

Comment 6

[Eric Rescorla (:ekr)]

Attachment: 10.7.3 signaling log

Comment 7

[Eric Rescorla (:ekr)]

The four logs I just attached are from a dual-freeze between a 10.6.8/10.7.3 pair

Comment 8

[Randell Jesup [:jesup] (needinfo me)]

Next time this happens, sit still, then reach over and pass you hand in front of the camera once.  See if the other side un-freezes.  If both were frozen, repeat on the other side.

I caught a case where both sides are sending NACKs and obviously neither is getting video they can decode.  There are times I hate NACK.  Most of the time, in fact.  Something about passing the hand in front or other abrupt changes pops it out; perhaps as Maire suggested it triggers a keyframe.

Open question: why does a mac seem needed to trigger this?  And why does it not give up on NACK?  And why did it fail to recover with NACK to start with (and both seemed to freeze for me at the same time).

Comment 9

[Eric Rescorla (:ekr)]

(In reply to Randell Jesup [:jesup] from comment #8)
> Next time this happens, sit still, then reach over and pass you hand in
> front of the camera once.  See if the other side un-freezes.  If both were
> frozen, repeat on the other side.

Well, this is how I test it. It doesn't always go away.



> I caught a case where both sides are sending NACKs and obviously neither is
> getting video they can decode.  There are times I hate NACK.  Most of the
> time, in fact.  Something about passing the hand in front or other abrupt
> changes pops it out; perhaps as Maire suggested it triggers a keyframe.
> 
> Open question: why does a mac seem needed to trigger this?  And why does it
> not give up on NACK?  And why did it fail to recover with NACK to start with
> (and both seemed to freeze for me at the same time).

Comment 10

[Randell Jesup [:jesup] (needinfo me)]

Attachment: reland Bug 832567: transmitting-only channels weren't accepting RTCP; turn off NACK mode and block bad outgoing RTCP from transmit-only channels

Comment 11

[Randell Jesup [:jesup] (needinfo me)]

Fixed by resolving merge issue from bug 880879