Trixie - Jellyfin under Docker with Nvidia transcode?

bugado · Wednesday, 10:08 pm

Yes it's an intel i7 7700K

Quote from chris_kmn
Code
nvidia-ctk cdi generate --output=/etc/cdi/nvidia.yaml

Quote

INFO[0000] Using /usr/lib/x86_64-linux-gnu/nvidia/current/libnvidia-ml.so.550.163.01

WARN[0000] Ignoring error in locating libnvidia-sandboxutils.so.1: pattern libnvidia-sandboxutils.so.1 not found

libnvidia-sandboxutils.so.1: not found

WARN[0000] Failed to init nvsandboxutils: ERROR_LIBRARY_LOAD; ignoring

INFO[0000] Auto-detected mode as 'nvml'

INFO[0001] Selecting /dev/nvidia0 as /dev/nvidia0

INFO[0001] Selecting /dev/dri/card0 as /dev/dri/card0

INFO[0001] Selecting /dev/dri/renderD128 as /dev/dri/renderD128

INFO[0001] Selecting /dev/nvidia0 as /dev/nvidia0

INFO[0001] Selecting /dev/dri/card0 as /dev/dri/card0

INFO[0001] Selecting /dev/dri/renderD128 as /dev/dri/renderD128

WARN[0001] failed to create discoverer for graphics mounts: failed to construct discoverer for graphics libraries: failed to get libcuda.so parent directory: failed to locate libcuda.so: pattern libcuda.so.550.163.01 not found

libcuda.so.550.163.01: not found

ERRO[0001] failed to generate CDI spec: failed to create edits common for entities: failed to create discoverer for common entities: failed to create discoverer for driver files: failed to get libcuda.so parent path: failed to locate libcuda.so: pattern libcuda.so.550.163.01 not found

libcuda.so.550.163.01: not found

Display More

Quote from chris_kmn

Code

docker run --rm --runtime=nvidia --gpus all ubuntu nvidia-smi

Code

docker: Error response from daemon: failed to create task for container: failed to create shim task: OCI runtime create failed: could not apply required modification to OCI specification: error modifying OCI spec: failed to inject CDI devices: unresolvable CDI devices runtime.nvidia.com/gpu=all

Run 'docker run --help' for more information

Quote from chris_kmn
Code
# devices:
#      - /dev/dri:/dev/dri

The same.

chris_kmn · Wednesday, 10:13 pm

Can you install libcuda ?

apt install libcuda ?

And you could try the advice from sammy3 :

Code

apt install nvidia-kernel-dkms nvidia-driver

bugado · Wednesday, 10:15 pm

Quote from chris_kmn

Can you install libcuda ?

apt install libcuda ?

Error: Unable to locate package libcuda

chris_kmn · Wednesday, 10:27 pm

sudo nvidia-ctk runtime configure --runtime=docker

sudo systemctl restart docker

bugado · Wednesday, 10:40 pm

Already done twice during your guide and leave Docker storage blank

Quote from chris_kmn

sudo nvidia-ctk runtime configure --runtime=docker

sudo systemctl restart docker

I try

Quote from sammy3
Code
apt install nvidia-kernel-dkms nvidia-driver

Code

nvidia-kernel-dkms is already the newest version (550.163.01-2).
nvidia-kernel-dkms set to manually installed.
nvidia-driver is already the newest version (550.163.01-2).

chris_kmn · Thursday, 7:28 am

Hm, lets recap:

Nvidia-driver installed successfully. Confirmed by nvidia-smi

Docker installed successfully, confirmed by running nvidia-smi in a container (docker exec -it jellyfin nvidia-smi)

nvidia container toolkit has an issue. (Runtime nvidia does not work in a docker container) CDI error 255 just states, that there was an error starting the container. May be you could find any logging information for your jellyfin container.

So we have to focus on that.

Please doublecheck Step 4 to Step 6, specially the compose-part with the config in OMV GUI

One thing to try could be installing the experimental version of the toolkit. But first check the upper.

bugado · Thursday, 10:01 am

I doublecheck Step 4 to Step 6. Everything is OK

my compose for jellyfin

Code

services:
  jellyfin:
    image: lscr.io/linuxserver/jellyfin:latest
    container_name: jellyfin
    environment:
      - PUID=1000
      - PGID=100
      - TZ=Europe/Paris
      - NVIDIA_DRIVER_CAPABILITIES=all
      - NVIDIA_VISIBLE_DEVICES=all
    volumes:
      - /srv/dev-disk-by-uuid-78903c71-7ca2-4741-b196-144048d0b23f/Jellyfin:/config
      - /srv/dev-disk-by-uuid-78903c71-7ca2-4741-b196-144048d0b23f/Transcode:/cache
      - /srv/dev-disk-by-uuid-e11ad7a7-91dd-411f-a7a2-3977b6427ce7/Videos:/media
    network_mode: host
#    ports:
#      - 8096:8096
    restart: always
#    runtime: nvidia
#    devices:
#      - /dev/dri:/dev/dri
    deploy:
     resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

Display More

When I start the container, no error:

https://drive.google.com/file/d/18w_uDp0r_2Hi00blGUGHIdHDExNyIn19/view?usp=drive_link

And when I launch a movie with hw:

https://drive.google.com/file/d/10MID9QS5NEF2wMzQjjiyJIcJPJ19U5sJ/view?usp=drive_link

We can see the FFMpeg error.

Trying to install the experimental packages

Code

sudo sed -i -e '/experimental/ s/^#//g' /etc/apt/sources.list.d/nvidia-container-toolkit.list

then

Code

sudo apt-get update

export NVIDIA_CONTAINER_TOOLKIT_VERSION=1.18.1-1  sudo apt-get install -y \      nvidia-container-toolkit=${NVIDIA_CONTAINER_TOOLKIT_VERSION} \      nvidia-container-toolkit-base=${NVIDIA_CONTAINER_TOOLKIT_VERSION} \      libnvidia-container-tools=${NVIDIA_CONTAINER_TOOLKIT_VERSION} \      libnvidia-container1=${NVIDIA_CONTAINER_TOOLKIT_VERSION}

and

Code

sudo nvidia-ctk runtime configure --runtime=docker

sudo systemctl restart docker

The same

chris_kmn · Thursday, 10:34 am

does this help?

apt update && sudo apt install -y libnvcuvid1 libnvidia-encode1

(from Jellyfin page)

and comment out the nvidia parameters in config:

#- NVIDIA_DRIVER_CAPABILITIES=all
#- NVIDIA_VISIBLE_DEVICES=all

chris_kmn · Thursday, 10:44 am

I did an AI analyse of the CDI error.

try to install:

sudo apt-get install nvidia-driver-550 nvidia-driver-550-lib

an then test again:

nvidia-ctk cdi generate --output=/etc/cdi/nvidia.yaml

bugado · Thursday, 11:18 am

Quote from chris_kmn

does this help?

apt update && sudo apt install -y libnvcuvid1 libnvidia-encode1

(from Jellyfin page)

and comment out the nvidia parameters in config:

#- NVIDIA_DRIVER_CAPABILITIES=all
#- NVIDIA_VISIBLE_DEVICES=all

Display More

That's it!!

We're here!! That's the trick!

I forgot to apt upgrade

I can transcode without any error and nvidia-smi post jellyfin ffmpeg!

THANK YOU VERY MUCH!!!!!!

bugado · Thursday, 1:12 pm

One more quick question to clarify things:

I need to stay on the Linux kernel 6.12.57+deb13-amd64. I tried reverting to 6.17.13 to test, and I get an Docker OCI error when starting Jellyfin.

So I reverted to 6.12, and everything is OK.

I need to check, once the kernel is updated, that I'm still on 6.12 by default?

chris_kmn · Thursday, 2:27 pm

Quote from bugado

One more quick question to clarify things:

I need to stay on the Linux kernel 6.12.57+deb13-amd64. I tried reverting to 6.17.13 to test, and I get an Docker OCI error when starting Jellyfin.

So I reverted to 6.12, and everything is OK.

I need to check, once the kernel is updated, that I'm still on 6.12 by default?

Display More

Yes you need to stick with 6.12. 6.14 might also work.

I'm not sure if there is a Proxmox Kernel Version 6.17. If it is available you could try that one (OMV GUI - Kernel - Install Proxmox Kernel)

bugado · Thursday, 5:19 pm

Okay, I gave it a try.

I installed kernel 6.17.4 pve, upgrade, reboot but I'm still getting the same OCI error when starting the Jellyfin container.

I'm going back to 6.12 non proxmox.

Thanks again, and see you soon.

chris_kmn · Thursday, 6:04 pm

Quote from bugado

Okay, I gave it a try.
I installed kernel 6.17.4 pve, upgrade, reboot but I'm still getting the same OCI error when starting the Jellyfin container.

I'm going back to 6.12 non proxmox.

Thanks again, and see you soon.

Display More

Ok, thanks for trying. Learned something again

Trixie - Jellyfin under Docker with Nvidia transcode?

bugado Thursday, 1:25 pm

Participate now!