Problems with legacy GPUs on updated APPs

[b-m-f]

App Name

Jellyfin

SCALE Version

22.12.2

App Version

10.8.9_14.0.1

Application Events

Application Logs

Container is not starting. No logs available

Application Configuration

Nothing changed since the version before, which runs fine. Except for the added GPU field.

Describe the bug

Using a legacy GPU (GT530) the nvidia-container-runtime will not work with TrueNAS Scale, as the legacy driver is not installed.

GPU support was disable in the Kubernetes settings, but the updated application starts loading GPU specific things, without checking for that flag.

To Reproduce

• Have g530 installed.
• Disable GPU support for Kubernetes
• Launch newest Jellyfin version.

Expected Behavior

Should run and not try to load nvidia drivers. GPU should be ignored

Screenshots

/

Additional Context

/

I've read and agree with the following

• [X] I've checked all open and closed issues and my issue is not there.

[PrivatePuffin]

@stavros-k I think you will want to have to look into this one!

[PrivatePuffin]

After discussing it, priority is low as likely not related to us....

[PrivatePuffin]

Note: I would advice trying to add the same GPU on official Apps, if those also throw this error file an issue with iX. If those work, please report so.

[b-m-f]

@Ornias1993 the problem is that I am not adding the GPU to the container. But somehow the nvidia-runtime wants to load the driver.

This does not happen with official charts or Truecharts before the breaking change.

[PrivatePuffin]

@stavros-k Could you look into potential differences in how we and official handles the runtimeclass, in due time?

[PrivatePuffin]

@b-m-f Can you confirm you're trying this on the official "community" train? Because that is the train you should be trying as it has mostly the same backend in these regards, where their stable train has older codebase.

[b-m-f]

@Ornias1993 I tried it with sonarr & radarr from both the "community" train and Truecharts, and both deploy successfully.

It seems that this issue is jellyfin specific and I have jumped to conclusions, sorry!

[PrivatePuffin]

Interesting, in that case it is indeed most likely a container issue... As the way gpu's are handled is done globally.

Closing as "most likely upstream issue"

[PrivatePuffin]

Thanks for all the effort put into this btw!

[PrivatePuffin]

@all-contributors please add @b-m-f for bug

[allcontributors[bot]]

@Ornias1993

I've put up a pull request to add @b-m-f! :tada:

[b-m-f]

I have figured out what is happening.

On old release there was an environment variable set NVIDIA_VISIBLE_DEVICES=void on the pod.

In the docker image it defaults to NVIDIA_VISIBLE_DEVICES=all.

Setting NVIDIA_VISIBLE_DEVICES=void via the extra environment variables fixes the issue.

[PrivatePuffin]

@stavros-k i thought we already set this?

[PrivatePuffin]

New plex and jellyfin releases are out with this fix @b-m-f

[stavros-k]

@all-contributors please add @b-m-f for bug

[allcontributors[bot]]

@stavros-k

@b-m-f already contributed before to bug

[truecharts-admin]

This issue is locked to prevent necro-posting on closed issues. Please create a new issue or contact staff on discord of the problem persists