VideoPlayer seemless loop

rodrigo

The app that runs for over 15hs, although it is not consistent to crash at the same time, and contains a lot more code, yes.
Most of the code is encapsulated in 1 page with a huge scroll and tons of user profile pics loaded on demand, the rest of the screens loads remote htmls via StageWebView.

It also has a Socket connection, there's a local server running redis, cause the Windows app in the big monitor is controlled by an ipad app, running the same codebase.

I don't have experience with Scout to profile this, but after a while Scout just runs of memory... so i can't leave it running for 15hs... In any case, when i tried it on my Win7 VM the memory was steady.

I send a Scout installer to be installed directly on the Windows computer to check (cause i couldnt replicate the crash on my end), sadly the Windows 7/8 machines are 32bits, so they couldn't install Scout, neither I can provide a 64bits version of the app.

Since this crash happens, I implemented BugSnag as a report tool, I totally recommend it... and I had no code exceptions for 5 days (except some Starling painter exceptions when "rebooting" the app, not a big deal). In any case I also store daily logs for the uncaught exceptions in the storage folder...

Probably is my fault to not cleanup the app properly, as a workaround I included this reboot routine

This is part of the crash log:

This is the PC info:

Probably I'm screwing the memory somewhere on my code, although the app crashes on IDLE state on the home screen (the bg video is here) ...

I sent another build that uses an image as a background instead of a video... so far no crashes .

dfdiaza

I have been facing the same issue with VideoTexture for several months, even in a simple app to collect data.
For example, I have one app where after some inactivity time, I load a screen with a mp4 video looping as screensaver (I am using VideoTexture).
This app is crashing when the ScreenSaver is running for a long time. The sooner I have seen the issue is around 30 minutes, but it happens frequently around 6-10 hours, definitively it’s happening at random times. [GPU and STD memory are stable during all the time].
When I disable the ScreenSaver, the app is able to run several days continuously without crashing, so I am almost sure this is a VideoTexture bug in Adobe AIR.

The app was run on several machines Mac Mini Late 2012 running Windows 7/8.1/10 x64 through bootcamp. All of them had the same issue. These are the two error messages I got:

Firma con problemas:
Nombre del evento de problema: APPCRASH
Nombre de la aplicación: VideoTextureCrashMinimal.exe
Versión de la aplicación: 0.0.0.0
Marca de tiempo de la aplicación: 583266d5
Nombre del módulo con errores: Adobe AIR.dll
Versión del módulo con errores: 26.0.0.127
Marca de tiempo del módulo con errores: 5955699c
Código de excepción: c0000005
Desplazamiento de excepción: 005e6641
Versión del sistema operativo: 6.3.9600.2.0.0.256.48
Id. de configuración regional: 3082
Información adicional 1: 5861
Información adicional 2: 5861822e1919d7c014bbb064c64908b2
Información adicional 3: 6ee2
Información adicional 4: 6ee2432e64e99bdcea5e4be767a3b1a2

Firma con problemas:
Nombre del evento de problema: BEX
Nombre de la aplicación: VideoTextureCrashMinimal.exe
Versión de la aplicación: 0.0.0.0
Marca de tiempo de la aplicación: 592bde08
Nombre del módulo con errores: StackHash_5861
Versión del módulo con errores: 0.0.0.0
Marca de tiempo del módulo con errores: 00000000
Desplazamiento de excepción: PCH_EE_FROM_ntdll+0x0003C7EC
Código de excepción: c0000005
Datos de excepción: 00000008
Versión del sistema operativo: 6.3.9600.2.0.0.256.48
Id. de configuración regional: 3082
Información adicional 1: 5861
Información adicional 2: 5861822e1919d7c014bbb064c64908b2
Información adicional 3: 6ee2
Información adicional 4: 6ee2432e64e99bdcea5e4be767a3b1a2

Some machine specs:

I have been able to reproduce the Crash on a Mac Mini. I attach the link to my minimal example (source and binary) to reproduce the crash. It’s just Stage3D, not starling is used.
https://www.dropbox.com/sh/isf3zvvfxgei4fj/AAA3k0XRGJZG1XMrxISObGsRa?dl=0

Because of this issue, I am telling my clients I can not develop apps with video loops, which is very embarrassing since most of the time I develop kiosk applications, which need a way to call the user attention, and there is no doubt a video is the right media. 🙈

tuarua

I think there is enough info in the last 2 posts to warrant a ticket on Daniel's Adobe Runtime Support repo on Github and a ticket on the Adobe tracker to bring to Adobe's attention.
https://github.com/Gamua/Adobe-Runtime-Support.
https://tracker.adobe.com/#/home

I'll vote for it.

From some searching:

Exception code c0000005 is the code for an access violation. That means that your program is accessing (either reading or writing) a memory address to which it does not have rights. Most commonly this is caused by:

Accessing a stale pointer. That is accessing memory that has already been deallocated. Note that such stale pointer accesses do not always result in access violations. Only if the memory manager has returned the memory to the system do you get an access violation.
(Other possible causes mentioned were file locks by anti-virus)

BEX + StackHash_xxx
Found references to this and games and GPU driver issues
https://www.youtube.com/watch?v=cEkoK3EHOlU - this person outlines a fix (although not sure if it is recommended)

rodrigo

Nice research Eoin.
So, regarding c0000005 ur suggestion is to embed the flv as bytearray like u did in your example to avoid file access errors?
That sounds insane, but I can try it.

tuarua

@Rodrigo
file access *could be the cause - It could also be GPU related - but I don't know as c0000005 covers different possibilities - it's some memory access issue and that could be GPU RAM or computer RAM. Doesn't mean a leak per se.

I think Diego's example is a good test as it implements using only the core stage3D classes.

I do think it would be good to gather all information about the hardware from you and Diego and file a bug report. Providing as much info as possible will help them reproduce the issue.

timwjohn

Only if the memory manager has returned the memory to the system do you get an access violation.

@tuarua That would explain the random nature of the crashes.

@rodrigo BugSnag looks really good. Is there support for AIR desktop?

I agree that a bug should be filed. I'm happy to do this (tomorrow) although my knowledge and understanding of the low level processes is limited. If it can be identified and fixed that would be a huge reduction in anxiety!

timwjohn

Ok, just checked on a very simple video looping app I started last night, and it's frozen (not crashed oddly). According to the logs it lasted 15 hours and both RAM and GPU were stable. I've been doing flash-based kiosk apps for well over a decade and my apps would run indefinitely, but this is the first time using VideoTexture. How seriously do you think Adobe would take this bug do you think?

tuarua

As long as the bug report contains lots of technical detail, logs, a sample project and several people saying they observe similar then I imagine they will.
Bugs that sit in the tracker tend to be those one liner bugs like "video doesn't work. Please Fix". We've all been on the end of those kind of reports.

timwjohn

@rodrigo, can you point me in the direction of a Windows BugSnag implementation for AIR? I can't find it for the life of me, except an iOS and Android ANE.

Haven't had the chance to file the bug today I'm afraid, but if you're all happy for me to do it with the data in this thread then I will in the morning.

rodrigo

Oh, I'm using the same ANE as u (probably), no native code for OSX/Windows I'm afraid 🙁 . But it falls back on regular API calls for unsupported platforms... so I just use it to track custom errors, uncaught exceptions, etc... (I had to modify the ANE as3 code to have more meaningful device info in the dashboard, although u can create your own tabs with the ANE's API).
Anyway, doesn't work to catch the Windows crashing on the video loop... I just referenced it cause is really useful for remote environments/installations that u have no access/control 🙂
Anyway, using Starling stats, or Context3D u can report drawCounts, totalGPU Memory, etc, or notify about exceeding thresholds for those values... idk

timwjohn

Hey @Diego,

Your minimal video demo looks like this for some reason? . That's not an accurate representation of the video.

I want to submit the most basic example possible for this bug report, but my low-level Stage3D skills are nonexistent. Are you able to tweak this so it looks right before we submit it?

dfdiaza

@timwjohn, sorry for the late response.
I don't understand low level stage3D neither, but I have been able to make it display the video in four vertex instead of three, so it's a more accurate representation.
The sample project is on the same DropBox link. (The .exe is not there, just the sources, as I am not on a windows machine at this time). Hope this works for you.

BTW, I have been able to reproduce the crash/freeze with this sample on my mac mini running Windows 8.1 x64, but the same sample running on another PC with windows 8.1 x64 has not crashed during the last 24 hours. The only difference I notice among them, when running the sample, it's that I can see the "green blink" on my mac mini, but it's not shown on the another PC (I don't know if this is relevant).

tuarua

The mac mini and Windows machine have different video cards I assume?
I think it's important to include hardware in the report including GPU details.
I'm not a fan of the idea of using seek to loop the video.
looping a 10 second video for 10 hours is 3,600 seek operations.

timwjohn

Nice one @diego. I haven't had a look yet but will soon.

As tuarua says, probably best to get the specs of all machines tested on with results.

@tuarua Would you recommend a simple stop()/play() instead?

tuarua

I'd recommend perhaps 2 examples.
1 which uses the "seek loop" way
2 which uses the "embed video as byte array in app and uses appendBytes" way as per my gist.

If 1+2 exhibit the same then it might suggest GPU/Video chipset issue on Stage3d
If 1 exhibits the issue but 2 doesn't then it might suggest the file pointer goes missing. from memory /becomes locked or something to that effect.

ozmachine

I've made dozens of kiosk Air apps with starling and looping videos/animations. There's bound to be some sort of issue with any app doing this kind of work after so many hours of display.

So I had each app detect itself crash and immediately start back up (requires permissions from the system). Also, each system was setup to sleep and restart at certain times of the day. This took the load off the clients, and they were pleased.

dfdiaza

@tuarua, Yes Eoin, they have different video cards, mac mini has Radeon HD 6630M and the PC has Intel HD Graphics. But with the latest example, I have been able to reproduce the issue on the PC too.

@timwjohn I have updated the DropBox folder with a "results" subfolder, adding some specs/screenshots from the two Windows machines I have to test, in both of them I am able to reproduce the crash with the latest example I created. I have attached the Adobe Scout logs (2 from the mac mini, 3 from the PC), it seems the crash always happen exactly when the loop is about to start playing again.
Hope it helps to you when reporting the issue. Thanks!

https://www.dropbox.com/sh/isf3zvvfxgei4fj/AAA3k0XRGJZG1XMrxISObGsRa?dl=0

tuarua

Lots of debug info Diego!
Had a brief looks at the Scout logs. You can definitely see a "wave" of activity where the seeks cause a file re-read into the NetStream. That's just an observation and may not be a deciding factor. Let's hope Adobe will look deeper.

timwjohn

OK, been a bit busy, but here are 2 versions of the VideoTextureCrash test applications as suggested by @tuarua

https://www.dropbox.com/sh/ejl5boxpqsgr4fn/AACng1amNpST9tOnr9dYndiBa?dl=0

The zip files can be imported straight into Flash Builder for Adobe to test straight away.

VideoTextureCrashMinimal is @diego's implementation
VideoTextureCrashAppendBytes is a modified version of @diego's that uses @tuarua's embedded FLV/appendBytes method (but without Starling)

The bin folder contains a captive release of each, plus an AIR file.

I'm going to start testing today. If anybody else here can run these on various machines and let me know the results that would be great!

timwjohn

Both the apps are crashing quite quickly. On one low power machine the video is freezing, but audio continues to play. On the other it crashed within few hours. I'm not getting any BEX errors however, just standard Application errors. Forgive my ignorance, but could it have anything to do with lost contexts (which Starling handles automatically)? I may make a simple Starling version of the app to see if it's any different.

« Previous Page Next Page »