621430 - Freeze (OOM) when playing Agent 008 Ball (NS_ERROR_OUT_OF_MEMORY) [nsIDOMHTMLAudioElement.play]

Status: VERIFIED FIXED

(Core :: Audio/Video, defect)

Description

[Frédéric Buclin]

This problem is new to Fx4.0b8, AFAIK. I don't remember having seen this problem with beta6 or beta7.

When playing the game at http://agent8ball.com/, it suddenly freezes after a few minutes with the following error in the Error Console:

Error : uncaught exception: [Exception... "Component returned failure code: 0x8007000e (NS_ERROR_OUT_OF_MEMORY) [nsIDOMHTMLAudioElement.play]"  nsresult: "0x8007000e (NS_ERROR_OUT_OF_MEMORY)"  location: "JS frame :: http://assets2.agent8ball.com/javascripts/compiled.js?1292308929 :: <TOP_LEVEL> :: line 127"  data: no]

I have several addons installed, so I wonder if one of them is interacting with the game or not. Maybe the error above is enough for you to debug it. :) Let me know if more information is needed.

Comment 1

[Frédéric Buclin]

If I reload the page after the game freezes, I get:

Error : out of memory
Source File : jar:file:///usr/local/firefox/omni.jar!/components/nsSessionStore.js
Line : 3611

Comment 2

[Frédéric Buclin]

After pasting the previous comment, I got:

Error : [Exception... "Component returned failure code: 0x8007000e (NS_ERROR_OUT_OF_MEMORY) [nsIScriptableUnicodeConverter.convertToInputStream]"  nsresult: "0x8007000e (NS_ERROR_OUT_OF_MEMORY)"  location: "JS frame :: jar:file:///usr/local/firefox/omni.jar!/components/nsSessionStore.js :: sss_writeFile :: line 3805"  data: no]
Source File : jar:file:///usr/local/firefox/omni.jar!/components/nsSessionStore.js
Line : 3805

Comment 3

[Frédéric Buclin]

(In reply to comment #0)
> I have several addons installed, so I wonder if one of them is interacting with
> the game or not.

This problem also occurs with a new clean profile, so extensions are not the culprit. Also, I can see the following error in the console:

mmap() failed: Ne peut allouer de la mémoire

(means: mmap() failed: cannot allocate memory)

Comment 4

[Josh Matthews [:jdm]]

I suspect this is related to bug 613915.

Comment 5

[Christian Ascheberg]

I see the same with http://www.pirateslovedaisies.com/

Error: uncaught exception: [Exception... "Component returned failure code: 0x8007000e (NS_ERROR_OUT_OF_MEMORY) [nsIDOMHTMLAudioElement.play]"  nsresult: "0x8007000e (NS_ERROR_OUT_OF_MEMORY)"  location: "JS frame :: 
http://www.pirateslovedaisies.com/scripts/compiled-scripts.js :: <TOP_LEVEL> :: line 507"  data: no]

The results is that Firefox (latest-trunk) hangs completly / crashes, the only way out is to kill the process. I am using Win Vista.


(In reply to comment #4)
> I suspect this is related to bug 613915.

Note that both websites are only using HTML5 features and no plugins.

Comment 6

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Tentatively making this block since it's a demo that plays very well otherwise ... at least until we figure out what the problem is. It may be too hard to fix for FF4 if it's, say, zillions of media elements lying around.

Comment 7

[Robert O'Callahan (:roc) (email my personal email if necessary)]

I get messages like [WAVE API] No memory error returned while executing opening audio device for playback
which suggests we may actually be leaking something

Comment 8

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Actually I don't think this should really block since it's probably not a regression. However, I would very much like to have a fix!

Comment 9

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Unsurprisingly, we have an obscene number of threads.

Comment 10

[Christian Ascheberg]

(In reply to comment #8)
> Actually I don't think this should really block since it's probably not a
> regression.

Maybe someone can confirm this:
works:
win32 2010-11-30-03-mozilla-central 837d7b346a64
broken:
win32 2010-12-01-03-mozilla-central d48a0e23feec

http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=837d7b346a64&tochange=d48a0e23feec

Comment 11

[Robert O'Callahan (:roc) (email my personal email if necessary)]

OK.

Comment 12

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Regression from bug 612798.

Comment 13

[Doug Turner (:dougt)]

Attachment: patch v.1

Comment 14

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Attachment: my fix

I think this one is simpler: we can just create the thread lazily. Also, I doubt it's safe to shut down the thread inside the stream destructor --- that spins the current thread's event loop.

Comment 15

[Doug Turner (:dougt)]

Comment on attachment 500948 [details] [diff] [review]
my fix

In this file, we already have a nsRunnable with the name AudioShutdownEvent.  Please consider renaming your runnable to AsyncThreadShutdownEvent or something similar to avoid any confusion.

nsAudioStreamLocal and nsAudioStreamRemote derive from nsAudioStream and nsAudioStream does have that nsCOMPtr to the playback thread.  Although, we aren't down casting anywhere, you might as well make ~nsAudioStream virtual to avoid future issues.

With those changes, r+

Comment 16

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Done.

Comment 17

[Robert O'Callahan (:roc) (email my personal email if necessary)]

http://hg.mozilla.org/mozilla-central/rev/29867a7bccf2

We need to figure out how to test this. I think we might want some kind of check at the end of mochitests, before we shut down, that not too many threads exist.

Comment 18

[Frédéric Buclin]

Verified fixed using 4.0b9pre, build d2bd42931b03

The memory is released once the game is over, or when doing nothing (Virtual Memory: 800 Mb instead of 2.8 Gb!) Thanks! :)