New issue
Advanced search Search tips
Note: Color blocks (like or ) mean that a user may not be available. Tooltip shows the reason.

Issue 819440 link

Starred by 4 users

Issue metadata

Status: Assigned
Owner:
Cc:
Components:
EstimatedDays: ----
NextAction: ----
OS: Linux , Windows , Chrome , Mac
Pri: 2
Type: Task

Blocked on:
issue 813218



Sign in to add a comment

Add testcase for Js suspend issue from issue 813218

Project Member Reported by q...@chromium.org, Mar 7 2018

Issue description

This is to address the TODO in:
https://chromium.googlesource.com/chromium/src.git/+/a07b7fdb2a948e91621bf13933bad9bb69b97520

Chrome Version: (copy from chrome://version) M64
OS: (e.g. Win7, OSX 10.9.5, etc...) all

What steps will reproduce the problem?
(1) join a g.co/present/some-name video call from a G Suite account
(2) screencast a video window
(3) make the screencaster window not in focus / rendering, i.e. look at the screencasted window or bring some other window in focus, such as news.google.com

What is the expected result?
screencaster can stay in the background, screencasting, indefinitely

What happens instead?
The Javascript on the screencaster tab stops running; the screencaster eventually gets kicked out of the meeting by the backend.

Please use labels and text to provide additional information.


For graphics-related bugs, please copy/paste the contents of the about:gpu
page at the end of this report.

 

Comment 1 by q...@chromium.org, Mar 7 2018

Blockedon: 813218

Comment 2 by q...@chromium.org, Mar 7 2018

The postmortem go/present-m64-regression (on corp) is ready to review. Thanks.
Status: Assigned (was: Untriaged)
Let's start with adding a unit test for this on the scheduler side. It would be great to have an end-to-end test that actually runs a video call, but I don't think our test infrastructure (telemetry) is capable of recording and replaying WebRTC traffic.

Comment 5 by kerl@google.com, Mar 13 2018

Would it be possible to simulate this scenario in a browser test using loopback peer connections? This is not the first bug we have had when presenting tabs are inactive.

Comment 6 Deleted

Comment 7 by mrobe...@twilio.com, Mar 15 2018

Hi,

Can you please explain the M64 regression and/or make public Issue 813218? My team noticed that, when screen-sharing or viewing screen-shares in video chats with VP8 simulcast enabled (and high CPU usage), placing the video chat tab in the background apparently led to timers being throttled. One of these timers is supposed to keep our signaling connection alive. Because the timer was throttled, the users connected to the video chat in a background tab would be disconnected. Is it possible this issue caused this?

Thanks,
Mark
Cc: phoglund@chromium.org
+Webrtc folks. We do have a webrtc Telemetry benchmark using static HTML content, not sure if we can add a test case that cover this case.
Cc: kerl@google.com sprang@chromium.org
Erik, your test case in https://chromium-review.googlesource.com/c/chromium/src/+/738040 was supposed to address this exact issue, right?

However, it only works on Linux for now. We should fix it so it runs on all platforms.
Re #4: Nope, you can't easily replay WebRTC traffic since it's real-time. I doubt you need that for something like this though. I think the test in #9 can effectively address it; screenshare from one tab to another and make sure the screenshare video keeps flowing even if the source tab is out of focus. Right? It gets trickier if the screenshare needs to run a while for the issue to reproduce, though.

Suggested actions:
1) make Erik's test run for all platforms
2) ensure it really catches the issue. If necessary, rewrite it to run for ~10s and make sure video flows continually.

Is there any easy way to re-introduce the issue so we can test if the test really works?

Comment 11 by sprang@google.com, Apr 3 2018

Right, the purpose of that test was to find background de-prioritization issues.

Would be great if we could enable it on other platforms, win 7 seems to be a bot issue. I think the Mac issue was a thread checker that has now been fixed, so maybe that's OK to go right now.

As stated, this test shuts down after receiving a single frame (the previous bug prevented video altogether).
We should also double-check bot settings to make sure it is even able to repro (I think the bot runs with --single-process or something like that).

Unfortunately I have no clock cycles to spare at the moment. :(

Sign in to add a comment