# ============================================================
# FreeFrame - optional NVIDIA GPU for video processing
# ============================================================
# Usage: add it to the -f files you already use, for example
#   docker compose --env-file .env.prod -f docker-compose.prod.yml \
#     -f docker-compose.images.yml -f docker-compose.gpu.yml up -d
# (use the same -f flags for every later command). The installer adds it
# itself when FREEFRAME_GPU=nvidia is in .env.prod (sudo freeframe configure
# turns it on and off).
#
# It needs, on the host:
#   * a working NVIDIA driver: nvidia-smi answers (on WSL2, the Windows
#     driver; nothing is installed inside WSL);
#   * the NVIDIA Container Toolkit, set up for Docker:
#       sudo nvidia-ctk runtime configure --runtime=docker
#       sudo systemctl restart docker
# The published API image needs nothing more: its ffmpeg has NVENC, which
# loads the driver's libraries that the toolkit mounts into the container.
#
# Only the transcoding worker gets the GPU. NVENC needs the "video" driver
# capability, and nvidia-smi (how the transcoder and `freeframe status` see
# the GPU) needs "utility"; Docker passes the capabilities below on as
# NVIDIA_DRIVER_CAPABILITIES, which is also set to the same value.
# TRANSCODER_PIPELINE=NVIDIA prefers NVENC, and the transcoder still falls
# back to the CPU when no GPU answers or a GPU attempt fails. A
# TRANSCODER_PIPELINE set in .env.prod wins over this default.

services:
  worker:
    environment:
      TRANSCODER_PIPELINE: ${TRANSCODER_PIPELINE:-NVIDIA}
      NVIDIA_DRIVER_CAPABILITIES: compute,utility,video
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu, compute, utility, video]