nvidia-cdi-hook fails to execute properly in NixOS (Possibly related to dynamic linking and glibc conflicts)

[colonelpanic8]

I'm not fully sure I understand why (although this issue in nixpkgs might be related NixOS/nixpkgs#338511)

but the nvidia-cdi-hook commands generated by calling cdi generate seem to fail in certain settings:

Here is the actual call that is used to generate the cdi json:

https://github.com/NixOS/nixpkgs/blob/85163d3c74112fa8033d18881af9410cebb38323/nixos/modules/services/hardware/nvidia-container-toolkit/cdi-generate.nix#L37

I don't get too much from the crashes (I have some core dumps, but looking at them didn't yield much). There is just a warning about stack smashing.

I'm assuming that this has something to do with dynamic linking and perhaps that environment that is being used by containerd when it is excuting runc, but its not at all obvious how to debug the issue.

[github-actions]

This issue is stale because it has been open 90 days with no activity. This issue will be closed in 30 days unless new comments are made or the stale label is removed. To skip these checks, apply the "lifecycle/frozen" label.

[ArangoGutierrez]

The nvidia-container-toolkit is tested and supported on the platforms listed here: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/supported-platforms.html

As NixOS is not an officially supported platform, we cannot provide support for the OS-specific dynamic linking and packaging issues that arise in this non-standard environment. We recommend checking with the nixpkgs maintainers, as they would be responsible for patching the toolkit's binaries to function correctly within NixOS.

We are closing this issue as it relates to an unsupported platform.

[colonelpanic8]

@ArangoGutierrez So its unreasonable to expect it to be POSSIBLE to make things work in a non fhs environment? I strongly suspect that this issue is merely a result of an assumption about something being in a specific directory that it is not always actually in. Its not usually hard to make accomodations for this with flags etc.

Also, from the page you linked:

"Give things a try and we invite you to report any issue observed even if your Linux distribution is not listed."