Transcoding causes Emby server restart - Memory Leak

[henriquefer 3]

I am reluctant to root my Shield. Will simply expand the internal storage help in this matter?

 

 

Expanding the internal storage does not help. I also have an SSD to expand the internal storage, mas I have to increase swap in order to allow a good experience using Shield.

 

Nvidia need to increase the amount of memory available on Shield. 3GB is a very low number comparing to the power of this machine.

[prunhardt 3]

Expanding the internal storage does not help. I also have an SSD to expand the internal storage, mas I have to increase swap in order to allow a good experience using Shield.

 

Nvidia need to increase the amount of memory available on Shield. 3GB is a very low number comparing to the power of this machine.

Ok, clear. Can you leave some instruction on how to root and how to increase swap?

[henriquefer 3]

Ok, clear. Can you leave some instruction on how to root and how to increase swap?

 

Yes. Of course.

 

1. You need a male-male USB cable: https://images-na.ssl-images-amazon.com/images/I/616YN8qAQLL._SX425_.jpg

2. You need a computer running Windows or Linux and install ADB (ADB is used communicate between your computer and Shield) https://www.xda-developers.com/install-adb-windows-macos-linux/.

3. Enable the "Developer Options" on your Shield (http://corochann.com/how-to-enable-developer-mode-in-android-tv-216.html).

4. Under the "Developer Options" enable USB Debugging and connect your Shield and computer using the male-male USB cable.

5. Go here: https://developer.nvidia.com/shield-developer-os-images to download the correct image for you device and follow the provided instructions. IMPORTANTE: Be sure that you have picked the correction image otherwise you will brick your Shield.

6. Then, you can run "sudo adb shell" on your computer and you are in the shell of your Shield.

7. Run the following commands to enable swap (5 GB).

swapoff /dev/block/zram0

echo 1 > /sys/block/zram0/reset

fallocate -l 5G /data/swapfile

chmod 600 /data/swapfile

mkswap /data/swapfile

swapon /data/swapfile

[softworkz 4569]

Usually, Shield does not have a lot of free memory (3 GB is insufficient and even that, Nvidia release a new model with the same amount of RAM), but when I am running a media server with a big library, the issue becomes bigger. 

 

Usually the memory does not increase quickly. It progressively increases over time. During transcoding activities, the amount of memory required increases, and without SWAP, Shield runs out of memory and stops the transcoding or the interface becomes completely laggy. 

 

Which process is taking those huge amounts of memory? EmbyServer or ffmpeg?

[henriquefer 3]

Which process is taking those huge amounts of memory? EmbyServer or ffmpeg?

 

Here, you can see the "top" output of my Shield with Emby server on (no active play):

 

 

And here, there is the "top" output when I am watching a episode (remotely) that requires transcoding:

[ebr 15667]

airplaydmr2 (whatever that is) seems to be eating a bunch of resources...

[henriquefer 3]

airplaydmr2 (whatever that is) seems to be eating a bunch of resources...

 

This is an app that allows me to use AirPlay on Shield. I installed it a coupe of days ago.

However, the issues regarding the usage of memory have already happened before the installation of this app. If you need, I can uninstall the app and share the results with you.

[softworkz 4569]

@@henriquefer - Thanks for the outputs. No need to end the other tasks. Operating systems tend to be lazy about releasing memory unless there is memory pressure, so in fact it's good to have some pressure because it will lead to more meaningful results. 

 

It would be awesome if you could create the same outputs once again:

1. Right after transcoding starts (the moment you see the first picture at the client)
2. After 1 minute
3. After 5 minutes

Leave throttling off and use a remote client (just as before)

[henriquefer 3]

@@henriquefer - Thanks for the outputs. No need to end the other tasks. Operating systems tend to be lazy about releasing memory unless there is memory pressure, so in fact it's good to have some pressure because it will lead to more meaningful results. 

 

It would be awesome if you could create the same outputs once again:

1. Right after transcoding starts (the moment you see the first picture at the client)
2. After 1 minute
3. After 5 minutes

Leave throttling off and use a remote client (just as before)

 

 

I have tested as before, and this is case (almost minute by minute):

 

Before starting:

 

Start:

 

 

1 Minute:

 

 

 

4 Minutes:

 

 

5 Minutes:

Some video freezing occurs on the client.

 

 

 

7 Minutes:

 

 

[softworkz 4569]

Excellent, thanks.

 

I've put the relevant lines in order here:

 

 

The hw accelerated ffmpeg run shows in fact excessive memory usage. Probably it crashes at some point (red line) and ffmpeg is restarted using sw transcoding.

 

Would you be able to post or PM the two ffmpeg logs from this test?

Edited November 14, 2019 by softworkz

[henriquefer 3]

Excellent, thanks.

 

I've put the relevant lines in order here:

 

 

The hw accelerated ffmpeg run shows in fact excessive memory usage. Probably it crashes at some point (red line) and ffmpeg is restarted using sw transcoding.

 

Would you be able to post or PM the two ffmpeg logs from this test?

 

 

I sent you the log files via PM.

[softworkz 4569]

Before hw transcoding fails after 4-5 minutes, it has already transcoded half of the file.

 

I'm wondering whether it's a hard leak or whether it just doesn't have time for releasing memory.

 

What happens when you activate throttling? I would like to know whether it would fail after transcoding the same amount of video data (= about 24 minutes) or later (or not).

 

For this test you'll need to let it play, so it will take a while, but you can stop once it fails and starts sw transcoding.

[henriquefer 3]

Before hw transcoding fails after 4-5 minutes, it has already transcoded half of the file.

 

I'm wondering whether it's a hard leak or whether it just doesn't have time for releasing memory.

 

What happens when you activate throttling? I would like to know whether it would fail after transcoding the same amount of video data (= about 24 minutes) or later (or not).

 

For this test you'll need to let it play, so it will take a while, but you can stop once it fails and starts sw transcoding.

 

Hi,

 

You can find all the screenshots in attachment.

I am going to send you the log files by PM.

 

The behaviour was the same.

Arquivo.zip

Edited November 15, 2019 by henriquefer

[softworkz 4569]

Thanks again. That seems to be a hard memory leak. I'll try to reproduce during the next days.

 

Last question: Do you experience that behavior with all kinds of source videos or only with specific ones?

[henriquefer 3]

Thanks again. That seems to be a hard memory leak. I'll try to reproduce during the next days.

 

Last question: Do you experience that behavior with all kinds of source videos or only with specific ones?

all videos bigger than ~20 minutes (depending on bitrate) that requires transcoding.

[henriquefer 3]

Thanks again. That seems to be a hard memory leak. I'll try to reproduce during the next days.

 

Last question: Do you experience that behavior with all kinds of source videos or only with specific ones?

Have you already be able to check this issue?

I have the last stable version of emby server for Android, and I still have this issue...

[Luke 40082]

What android device are you running the server on?

[henriquefer 3]

Nvidia Shield

[Luke 40082]

Can you try our beta package and see how that compares?

[henriquefer 3]

Can you try our beta package and see how that compares?

I update to that version right now. I am going to test it in the next days. I let you know the results. Thanks

[Luke 40082]

Great, thanks !

[henriquefer 3]

Great, thanks !

 

 

So far, the actual version is working perfectly! Thanks.

[Luke 40082]

Thanks for the feedback !