docker compose setting for transcoder-nvidia

[nnnlog]

Kyoo's version

4.7.0

What happened?

Default setting of transcoder-nvidia in docker-compose.yaml (below) didn't work for me.

    deploy:
      resources:
        reservations:
          devices:
            - capabilities: [gpu]

In the container, I can see below error.

[AVHWDeviceContext @ 0x59162cd23d40] Cannot load libcuda.so.1
[AVHWDeviceContext @ 0x59162cd23d40] Could not dynamically load CUDA
Device creation failed: -1.
[vist#0:0/hevc @ 0x59162ccfcfc0] [dec:hevc @ 0x59162cd284c0] No device available for decoder: device type cuda needed for codec hevc.
[vist#0:0/hevc @ 0x59162ccfcfc0] [dec:hevc @ 0x59162cd284c0] Hardware device setup failed for decoder: Operation not permitted
[vost#0:0/h264_nvenc @ 0x59162ccd9300] Error initializing a simple filtergraph
Error opening output file /cache/69e347f2747a4aa88c64adc9f3932541af040352/segment-1080p-2-%d.ts.
Error opening output files: Operation not permitted

With new configuration like below, it worked for me.

    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

It seems to specifying keys (driver and count) is essential. (https://stackoverflow.com/questions/74175742/docker-compose-cant-access-gpu-from-compose-but-can-from-run)

My Environment:

• Docker version 27.3.1, build ce12230
• NVIDIA Container Runtime Hook version 1.16.2
• Host OS: Ubuntu 24.04.1 LTS x86_64