Transcoding stops working after a while (ffmpeg 187)

[wwklnd]

Hi! Apologies if any of the many threads about ffmpeg error 187 contained this issue, I couldn't find an exact match when searching the forum.

I am running Jellyfin 10.10.7 in a Docker container (LSIO version) on Ubuntu 24.04.2 LTS. My server has a Ryzen 1700X, GTX1060 6GB, and 32GB RAM. When I first start the container, I am able to watch videos using hardware transcoding on the GPU without issue (confirmed that the GPU is being used via nvidia-smi and the logs.) However, after a while, I'm not sure how long, GPU transcoding stops working and trying to watch a video that requires hardware transcoding results in a fatal error and the attached log output: 
  jellyfin-errors-187.txt (Size: 22.62 KB / Downloads: 17)

It happens when I am using the Jellyfin browser client (using the Zen browser which is based on Firefox) and I believe the Jellyfin Android app. I also believe I've been able to play these videos using the Streamyfin app even when the issue has appeared in the browser client. I'm not 100% sure though, and since I just restarted the container I can't check at the moment.

Any ideas and suggestions are welcome! Sorry if I missed anything I should have added to the post, I'll happily do so on request.

[theguymadmax]

Post the Ffmpeg logs.

[xaque]

Not sure if this is the same issue I get from time to time, but every so often my container will "detach" the gpu thats passed to it. I'm using a nvidia card and using the nvidia container toolkit, and can confirm using nvidia-smi command from inside the container that the card is no longer detected. Perhaps there is also a way to check if your ryzen card is doing something similar. Not that this would solve your issue, but at least it would help diagnose the situation. I have to restart the container to get it to re-attatch, and have though about writing up a little healthcheck script to automate this, but its pretty infrequent and I'd like to avoid restarting the container when I'm in the middle of watching something that didn't require the gpu... just good for thought, and it may not even be the issue you're encountering.

[wwklnd]

Post the Ffmpeg logs.

Apologies for the late response, I was sure I'd set the settings to send an email when someone replied to the thread. I restarted the server earlier today but the problem usually arises again after a few hours so I'll post the logs then. I currently don't have any saved logs that show it happening (I restarted Jellyfin yesterday as well to get it to play a video).

Not sure if this is the same issue I get from time to time, but every so often my container will "detach" the gpu thats passed to it. I'm using a nvidia card and  using the nvidia container toolkit, and can confirm using nvidia-smi command from inside the container that the card is no longer detected. Perhaps there is also a way to check if your ryzen card is doing something similar. Not that this would solve your issue, but at least it would help diagnose the situation. I have to restart the container to get it to re-attatch, and have though about writing up a little healthcheck script to automate this, but its pretty infrequent and I'd like to avoid restarting the container when I'm in the middle of watching something that didn't require the gpu... just good for thought, and it may not even be the issue you're encountering.

The GPU is an Nvidia card, and this (it somehow detaching) sounds like what I suspected is happening! I've never had an issue with getting errors mid-playback though.

[wwklnd]

Post the Ffmpeg logs.

Here we go, it took a while for it to happen again. It seems that the GPU detaches somehow, since ffmpeg can't find a CUDA-capable device.

[theguymadmax]

Hmm. Searching around i didn't see a solution besides a workaround found in this thread: https://github.com/jellyfin/jellyfin/issues/9177. You could give that script a try.

[wwklnd]

Hmm. Searching around i didn't see a solution besides a workaround found in this thread: https://github.com/jellyfin/jellyfin/issues/9177. You could give that script a try.

Thanks! I appreciate it. I looked at the thread and tried running nvidia-smi -pm 1 inside the container to enable persistence mode. If that doesn't work I'll either run the script or see if nvidia-persistenced works.