Fix how configurations files are created for the nvidia-container-toolkit

[arnaldo2792]

Issue number:

N / A

Description of changes:

In 911775f, we changed how the configuration files for the nvidia-container-toolkit were created, and used %post install scripts to create hard links to the correspondent configuration per variant. However, the hard links weren't created since the lua scripts use relative paths instead of absolute paths and the builds didn't fail since %post install scripts fail silently.

With this commit, instead of creating hard links with %post install scripts, the configuration files for the nvidia-container-toolkit are copied over with tmpfiles.d configuration files.

Testing done:

• In k8s 1.29 x86_64, I verified the containerized workload doesn't have access to all the GPUs when NVIDIA_VISIBLE_DEVICES=all:

~ on Fedora ❯ k describe pod gpu-tests-2-thz4w | rg NVIDIA_VISIBLE_DEVICES
      NVIDIA_VISIBLE_DEVICES:      all

# There are 4 GPUs in the instance, and only two were configured as request in the Daemonset spec
~ on Fedora ❯ k exec gpu-tests-2-thz4w -- nvidia-smi
Wed Feb 21 02:30:31 2024
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.129.03             Driver Version: 535.129.03   CUDA Version: 12.2     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  Tesla T4                       Off | 00000000:00:1C.0 Off |                    0 |
| N/A   17C    P8               9W /  70W |      2MiB / 15360MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
|   1  Tesla T4                       Off | 00000000:00:1E.0 Off |                    0 |
| N/A   18C    P8               8W /  70W |      2MiB / 15360MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+

• in aws-ecs-2-nvidia x86_64, I confirmed the configuration file was copied over which wasn't when the %post install script was used:

bash-5.1# apiclient get os
{
  "os": {
    "arch": "x86_64",
    "build_id": "08dfbd2b",
    "pretty_name": "Bottlerocket OS 1.19.2 (aws-ecs-2-nvidia)",
    "variant_id": "aws-ecs-2-nvidia",
    "version_id": "1.19.2"
  }
}
bash-5.1# ls /etc/nvidia-container-runtime/config.toml
/etc/nvidia-container-runtime/config.toml

Terms of contribution:

By submitting this pull request, I agree that this contribution is dual-licensed under the terms of both the Apache License, version 2.0, and the MIT license.