1603887 - dataChannel closes unexpectedly, after two rounds of negotiation

Status: RESOLVED FIXED

(Core :: WebRTC: Networking, defect, P2)

Description

[makarandp]

Attachment: datatracks.html

User Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_14_6) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/78.0.3904.108 Safari/537.36

Steps to reproduce:

I am seeing that dataChannel close unexpected after a round of negotiation.
To reproduce, Please run this JSFiddle

https://jsfiddle.net/makarandp/wynouqhp/

Actual results:

Notice that after Alice and Bob finish 2nd round of negotiation
the dataChannel got closed for no reason.

Expected results:

dataChannel should instead get opened and they should be able to send / receive data on it.

Comment 1

[makarandp]

Notice that same sample works as expected on Chrome.

Comment 2

[makarandp]

Attachment: datatracks updated.html

Comment 3

[Hani Yacoub, Desktop QA]

Reproducible on the latest Firefox Nightly 73.0a1, Firefox 72.0b8 and on Firefox 71.0 on Windows 10 x 64, Mac OS X 10.14 and on Ubuntu 18.04 x64.

Comment 4

[Nico Grunbaum [:ng, @chew:mozilla.org]]

Looks like this goes back at least as far back as at least 53. This is most likely not a regression.

Comment 5

[makarandp]

@Nico, any update on this one ?

Comment 6

[Anna Vasilko]

Hi, any chance this bug can get some movement? Our customers have reported it, data tracks are not very usable in Firefox because of this.

Comment 7

[Byron Campen [:bwc]]

So, it looks like this is because of this code:

https://searchfox.org/mozilla-central/rev/6cd54550a27e2f6ca0755a25328f769e41e524f4/media/webrtc/signaling/src/peerconnection/PeerConnectionImpl.cpp#1076-1079

The spec doesn't say to do this at all. Let me see if things behave better when I remove this.

Comment 8

[Byron Campen [:bwc]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=c1bc9033def67a9b1f738b7f5d0f93843854a8f2

Comment 9

[Byron Campen [:bwc]]

Fiddle seems to be working better with the patches, although the DataChannel is still closing after the open event because the other end hasn't registered on the 'datachannel' event handler, causing the DataChannel to be garbage-collected.

Can you verify that the binary here fixes your issue?

https://firefox-ci-tc.services.mozilla.com/api/queue/v1/task/Ne4w_OWXRIiA-MowvOei6Q/runs/0/artifacts/public/build/target.dmg

Comment 10

[makarandp]

Attachment: error when opening binary

Bryon, I tried to launch the you binary, but got an error while opening it:
=======

Comment 11

[makarandp]

Bryon, I tried to launch the you binary, but got an error while opening it. I have attached the error.

Comment 12

[makarandp]

see details here: https://bugzilla.mozilla.org/attachment.cgi?id=9134804&action=edit

Comment 13

[Byron Campen [:bwc]]

Huh. Binary opens fine for me. Maybe try the opt build here?

https://firefox-ci-tc.services.mozilla.com/api/queue/v1/task/CZ9osdkFQJ6SkP0GSmKFTg/runs/0/artifacts/public/build/target.dmg

Comment 14

[makarandp]

Byron, still no luck with new binary. This time the browser launches, but shows beachball for about 30 seconds, The beachball goes away but browser does not navigate to anything entered in address bar.

Comment 15

[Byron Campen [:bwc]]

That's extremely weird. Nothing I touched would effect startup, so this must be some sort of build problem.

Comment 16

[Byron Campen [:bwc]]

Attachment: Bug 1603887: Drive closing of DataChannels with transport events, not negotiation events. r?ng

Comment 17

[BugBot [:suhaib / :marco/ :calixte]]

Bugbug thinks this bug is a regression, but please revert this change in case of error.

Comment 18

[Byron Campen [:bwc]]

Honestly, I'm not sure this ever worked properly.

Comment 19

[Byron Campen [:bwc]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=2240ec6512b16d9dcfad93c41309e696fcc5a839

Comment 20

[Byron Campen [:bwc]]

Waiting on some retriggers.

Comment 21

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/081cfce87dbe
Drive closing of DataChannels with transport events, not negotiation events. r=ng

Comment 22

[Daniel Varga [:dvarga]]

https://hg.mozilla.org/mozilla-central/rev/081cfce87dbe

Comment 23

[Hani Yacoub, Desktop QA]

I tried to verify the fix on the latest Firefox Nightly 77.0a1 (2020-04-21) on MacOS 10.14 and I still can see the same problem reported in this bug.
But when I tried to verify the bug on Firefox Nightly 76.0a1 (2020-03-26) after the bug was fixed I couldn't reproduce the issue.

Could you please take a look at this?

Comment 24

[Byron Campen [:bwc]]

Fiddle seems to work fine for me, although the behavior in comment 9 is still present (which is as expected). Can you attach the output you're seeing from the fiddle?

Comment 25

[Hani Yacoub, Desktop QA]

Attachment: Screenshot 2020-04-29 at 11.57.57.png

Attaching a screenshot with the result on the latest Nightly which seems to be the correct result.
I'm not sure why I was seeing "Boo, data channel unexpectedly closed!" as the last message when I tested the last time.

Comment 26

[Byron Campen [:bwc]]

That was probably the same thing I noted in comment 9.