[HW Accel Support]: Nvidia drivers not installed properly in container

[ajw107]

Describe the problem you are having

The use of a nvidia card does not work in current containers as the drivers are not linked properly. If you run nvidia-smi it will report that it cannot find libnvidia-ml.so and the logs report that they cannot find libcuda.so

Version

0.14.1-f4f3cfa

In which browser(s) are you experiencing the issue with?

Google Chrome 127.0.6533.120

Frigate config file

birdseye:
  enabled: true
  restream: false
  width: 1280
  height: 720
  quality: 8
  mode: continuous

mqtt:
  enabled: true
  host: <removed>
  port: 1883
  user: <removed>
  password: <removed>

ffmpeg:
  hwaccel_args: preset-nvidia-h264
  input_args:
    - -avoid_negative_ts
    - make_zero
    - -fflags
    - nobuffer
    - -flags
    - low_delay
    - -strict
    - experimental
    - -fflags
    - +genpts+discardcorrupt
    - -rw_timeout
    - '5000000'
    - -use_wallclock_as_timestamps
    - '1'
    - -probesize
    - 1000M
    - -analyzeduration
    - 1000M

go2rtc:
  streams:
    front:
      - rtspx://<removed>/<removed>
    doorbell_main:
      - "ffmpeg:http://<removed>/flv?port=1935&app=bcs&stream=channel0_main.bcs&user=admin&password=<removed>#video=copy#audio=copy#audio=opus"
      - rtsp://admin:<removed>@<removed>/Preview_01_sub
    doorbell_sub:
      - "ffmpeg:http://<removed>/flv?port=1935&app=bcs&stream=channel0_ext.bcs&user=admin&password=<removed>"
  ## THIS PREVENTS THE AUDIO COMING OUT FROM THE DOORBELL SOUNDING MUFFLED
  ffmpeg:
    bin: ffmpeg
    volume: -af "volume=30dB"
  webrtc:
    candidates:
      - <removed>:8555
      - stun:8555

detectors:
  coral:
    type: edgetpu
    device: usb

cameras:
  front_garden_g4pro:
    enabled: True
    mqtt:
      timestamp: false
      bounding_box: false
      quality: 100
      height: 500
    ffmpeg:
      output_args:
        record: preset-record-ubiquiti
      hwaccel_args: preset-nvidia-h264
      inputs:
        - path: rtsp://127.0.0.1:8554/front?video=copy
          input_args: preset-rtsp-restream
          roles:
            - detect
            - record
    detect:
      enabled: true
      width: 3840
      height: 2160
      fps: 5
    record:
      enabled: true
    snapshots:
      enabled: true
    motion:
      mask:
        - 0,181,0,0,755,0,779,176
        - 3840,0,2375,0,2240,193,2531,451,2699,648,2917,882,3019,1116,3237,1411,3077,1461,2950,1797,3061,1858,2974,2160,3840,2160

  doorbell:
    enabled: true
    mqtt:
      timestamp: false
      bounding_box: false
      quality: 100
      height: 500
    ffmpeg:
      output_args:
        record: -f segment -segment_time 10 -segment_format mp4 -reset_timestamps
          1 -strftime 1 -ar 44100 -c:a aac -c:v copy
      hwaccel_args: preset-nvidia-h264
      inputs:
        - path: rtsp://127.0.0.1:8554/doorbell_main?video=copy&audio=aac
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/doorbell_sub?video=copy
          input_args: preset-rtsp-restream
          roles:
            - detect
    live:
      stream_name: doorbell_sub
    detect:
      enabled: true
      width: 896
      height: 672
      fps: 7
    record:
      enabled: true
    snapshots:
      enabled: true
    motion:
      mask:
        - 293,1570,241,1264,1272,1283,665,1121,613,607,1627,818,1723,1089,1958,1187,1980,1105,2533,1083,2560,342,0,222,0,1614

    objects:
      track:
        - person
        - cat
      mask: 293,1570,241,1264,1272,1283,665,1121,613,607,1627,818,1723,1089,1958,1187,1980,1105,2533,1083,2560,342,0,222,0,1614

detect:
  fps: 5
  max_disappeared: 100
  stationary:
    interval: 10
    threshold: 200

snapshots:
  enabled: true
  retain:
    default: 60

record:
  enabled: true
  retain:
    days: 7
    mode: all
  events:
    pre_capture: 5
    post_capture: 5
    objects:
      - person
      - dog
      - cat
      - bird
      - mouse
      - car
      - bicycle
      - motorcycle
    retain:
      default: 30
      mode: active_objects
version: 0.14

docker-compose file or Docker CLI command

### YAML Anchors ###
x-gpu: &gpu
  deploy:
    resources:
      reservations:
        devices:
          - driver: nvidia
            count: all
            capabilities: [gpu]
#            capabilities: ["compute", "utility", "graphics"]

x-restart: &restart
  restart: unless-stopped

volumes:
  frigate-appdata-volume:
    <<: *default-local-volume
    driver_opts:
      type: none
      o: bind
      device: ${APPDATA_ROOT}/frigate/config

  frigate-storage-volume:
    <<: *default-local-volume
    driver_opts:
      type: none
      o: bind
      device: ${EXT4_DATA_ROOT}/frigate

services:
  &name frigate-service:
    <<: [ *gpu, *restart ]
    container_name: *name
    privileged: true # this may not be necessary for all setups
    image: ghcr.io/blakeblackshear/frigate:stable
    shm_size: "4096mb" # update for your cameras based on calculation above (mine is >201mb)
    devices:
      - /dev/bus/usb:/dev/bus/usb # Passes the USB Coral, needs to be modified for other versions
      - /dev/dri/renderD128:/dev/dri/renderD128 # For intel hwaccel, needs to be updated for your hardware
    volumes:
      - /etc/localtime:/etc/localtime:ro
      - frigate-appdata-volume:/config
      - frigate-storage-volume:/media/frigate
      - type: tmpfs # Optional: 1GB of memory, reduces SSD/SD Card wear
        target: /tmp/cache
        tmpfs:
          size: 1000000000
    ports:
      - "8971:8971"
      - "5000:5000" # Internal unauthenticated access. Expose carefully.
      - "8554:8554" # RTSP feeds
      - "8555:8555/tcp" # WebRTC over tcp
      - "8555:8555/udp" # WebRTC over udp
      - "1984:1984" # goRTC API
#    environment:
#      FRIGATE_RTSP_PASSWORD: "password"

Relevant Frigate log output

See description above, unfortunately I;ve deleted the log entries, sorry

Relevant go2rtc log output

See description above, unfortunately I've deleted the log entries, sorry

FFprobe output from your camera

Basically ffmepg complainign it can not see the stream (as the libcuda.so drvier is missing)

Operating system

Debian

Install method

Docker Compose

Network connection

Wired

Camera make and model

Reolink Doorbell PoE, Unifi G4 Pro

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

To temporarily fix it (needs running after every container update until the drivers are fixed) run these commands:

for file in /usr/lib/x86_64-linux-gnu/*.560.35.03; \
  do ln -s "${file}" /usr/lib/x86_64-linux-gnu/"$(basename "${file}" ".560.35.03")"; \
done #replaces the missing symlinks
ldconfig #tells the OS to reload the driver files

NOTE: if it doesn't work, check the driver file version by ls -alF /usr/lib/x86_64-linux-gnu
This will symlink all the *so.560.35.03 driver files to their *.so counterparts, which are missing. There are also *.so.1, etc missing, but adding the *.so ones seems to be suffient, although I am getting high CPU warnings, so it it may need further things doing to correct it.

[NickM-27]

this is an issue with how nvidia toolkit or driver was installed, there are a number of issues that were similar that users solved, maybe one of their solutions works for you as well

https://github.com/blakeblackshear/frigate/issues?q=is%3Aissue+cannot+find+libnvidia-ml.so+is%3Aclosed

[fatcat]

I started getting "libcuda.so.1 not found" in my logs after a container host kernel update. Not sure why that would matter though. Added this mapping to the docker-compose.yml "volumes":

/usr/lib/x86_64-linux-gnu/nvidia/current/libcuda.so.1:/usr/lib/libcuda.so.1