The Docker container stops when using `python3 scripts/launch_triton_server.py --world_size 1 --model_repo=model_repo/` as the starting command in the Docker Compose YAML file.

[Aquasar11]

System Info

• CPU EPYC 7H12 (32 core)
• GPU NVIDIA A100-SXM4-80GB

Who can help?

No response

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

1. pull triton official image
2. clone tensorrtllm_backend
3. move your model to repo and make config files
4. start the container with docker-compose.yaml file like bellow

services:
  tritonserver:
    image: triton_trt_llm
    network_mode: "host"
    container_name: triton
    shm_size: '1gb'
    volumes:
      - /data:/workspace
    working_dir: /workspace
    restart: always
    deploy:
      resources:
        reservations:
          devices:
            - capabilities: [gpu]
    command: bash -c "python3 ./tensorrtllm_backend/scripts/launch_triton_server.py --world_size=1 --model_repo=tensorrtllm_backend/all_models/inflight_batcher_llm/"

Expected behavior

After running the docker compose up command, I expect the container to start the Triton server, wait for it, and remain running unless an error occurs.

actual behavior

The container starts, runs the Python script, and exits immediately without waiting for the Triton server and TensorRT-LLM backend.

additional notes

This bug will be fixed with a simple command after the last line in scripts/launch_triton_server.py like this

child = subprocess.Popen(cmd, env=env)
child.communicate()

[Aquasar11]

I've suggested this PR to fix the issue.