804764 - Calling gUM with video: true and running it for ~20 seconds - video hangs and eventually crashes the browser

Status: RESOLVED WORKSFORME

(Core :: WebRTC: Audio/Video, defect, P1)

Description

[Jason Smith [:jsmith]]

Attachment: Test Case

Steps:

1. Call gUM with the attached test case and wait about ~20 seconds - 30 seconds

Expected:

The video should continue to play without errors and streaming with no hangs noticeable.

Actual:

The video gets hung entirely, with occasional spurts of showing the hang stopping. Eventually, firefox will crash with a corrupt dump - https://crash-stats.mozilla.com/report/index/bp-0de44618-ec52-4b11-aefe-32fec2121023.

Upon seeing a corrupt dump, I'm playing it safe it flagging this as a sec bug. If I'm wrong, feel free to turn off the flag.

Comment 1

[Jason Smith [:jsmith]]

Actually, I don't think there's security impact here, unless the corrupt dump means something.

Comment 2

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

It's likely a straight out-of-memory crash, as if frames stop being consumed the graph buffers an infinite amount of data.

I ran for an hour and 40 minutes with no problem, stable mem use.  Hit "stop", started video again, walked away for a few minutes and it locked, and mem uses started going straight-line up.  

Two bugs: this one, and another should be opened on limiting the data buildup in MediaStreams.

Comment 3

[Jason Smith [:jsmith]]

Okay, I filed bug 804874 about the limiting the data buildup in MediaStreams.

Comment 5

[Jason Smith [:jsmith]]

Not really a candidate for a crashtest.

Comment 6

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

Strong belief this is now fixed.  WFM on win32 nightly.  qawanted

Comment 7

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

Assigning to jsmith until retest done

Comment 8

[Jason Smith [:jsmith]]

Can't reproduce either.