Run garage with Docker

Currently there are two types of garage images available on Docker Hub:

  • garage: garage without environment visualization.

  • garage-nvidia: garage with GPU support, CUDA, and environment visualization capability using NVIDIA graphics cards.

If you want to compile a new image using the the source, proceed to the document Building garage Docker image from source instead.

garage image

Prerequisites

Be aware of the following prerequisites to run the image.

  • Install Docker CE version 19.03 or higher. Tested on version 19.03.

Tested on Ubuntu 16.04, 18.04 and 20.04.

Running the garage Docker image

The garage container comes bundled with the examples available in the garage repo. To run an example launcher in the container, execute:

docker run -it --rm rlworkgroup/garage python examples/tf/trpo_cartpole.py

To get a list of all the examples, you can run:

docker run -it --rm rlworkgroup/garage ls -R examples

This will run the latest image available on Docker Hub, which coincides with the latest stable release of garage. To use a specific release such as v2020.06, use rlworkgroup/garage:2020.06.

The container runs with the user garage-user with the current working directory as /home/garage-user.

To save the generated experiment data in a directory on your computer, you can specify the path to that directory using the argument -v with the docker run command. -v accepts an argument of the type <absolute path of folder on host machine>:<absolute path of folder inside container>. Since the working directory in the container is /home/garage-user by default, the results are written to /home/garage-user/data.

For example, if the path of the directory on your computer where you want the results to be stored is /home/user/data, make sure the directory exists.

Note

Additionally, if you are on linux, make sure that this directory is writeable by the container user garage-user, by either making it accessible by running:

setfacl -m u:999:rwx /home/user/data

or making it writable to all other users by running:

chmod 777 /home/user/data

or giving ownership of the directory to garage-user through:

chown -R 999:docker /home/user/data

Then, if the path of the directory on your computer is /home/user/data, execute:

docker run \
  -it \
  --rm \
  -v /home/user/data:/home/garage-user/data \
  rlworkgroup/garage \
  python examples/tf/trpo_cartpole.py

Similarly, if you want to run your own code inside the Docker container, you can mount the directory containing your files as a volume, by using the -v option with docker run. If your code is in a file called launcher.py in the directory, say, /home/user/my_garage_experiment_dir, then you can mount it at /home/garage-user/my_experiment_dir in the container as follows:

docker run \
  -it \
  --rm \
  -v /home/user/my_garage_experiment_dir:/home/garage-user/my_experiment_dir \
  -v /home/user/my_garage_experiment_dir/data:/home/garage-user/data \
  rlworkgroup/garage \
  python my_experiment/launcher123.py

To run environments using MuJoCo, pass the contents of the MuJoCo key in a variable named MJKEY in the same docker-run command using cat. For example, if your key is at ~/.mujoco/mjkey.txt, execute:

docker run \
  -it \
  --rm \
  -e MJKEY="$(cat ~/.mujoco/mjkey.txt)" \
  # ... other arguments here
  rlworkgroup/garage \
  python examples/tf/trpo_swimmer.py

garage-nvidia image

The garage NVIDIA images come with CUDA 11.0 and cuDNN 8.0 (required for TensorFlow >= 2.4.0).

Prerequisites for garage-nvidia image

Additional to the prerequisites for the headless image, make sure to have:

Tested on Ubuntu 18.04 and 20.04.

Running the garage-nvidia image

The same commands for the garage image mentioned above apply for the nvidia image, except that the image name is rlworkgroup/garage-nvidia.

For example, to execute a launcher file:

docker run \
  -it \
  --rm \
  # ...
  rlworkgroup/garage-nvidia \
  python examples/tf/trpo_cartpole.py

Enabling environment visualization

Allow the Docker container to access the X server on your machine by running:

xhost +local:docker

and while running the Docker container, add the following arguments to docker run:

-v /tmp/.X11-unix:/tmp/.X11-unix
-e DISPLAY=$DISPLAY
-e QT_X11_NO_MITSHM=1

For example:

docker run \
  -it \
  --rm \
  -e MJKEY="$(cat ~/.mujoco/mjkey.txt)" \
  -v /tmp/.X11-unix:/tmp/.X11-unix \
  -e DISPLAY=$DISPLAY \
  -e QT_X11_NO_MITSHM=1 \
  rlworkgroup/garage \
  python examples/tf/trpo_swimmer.py

Expose GPUs for use

By default, garage-nvidia uses all of your gpus. If you want to customize which GPUs are used and/or want to set the GPU capabilities exposed, as described in official Docker documentation here, you can pass the desired values to --gpus option as follows:

docker run \
  -it \
  --rm \
  --gpus "device=0,2" \
  # ...
  rlworkgroup/garage-nvidia \
  python examples/tf/trpo_cartpole.py

Using a different CUDA version

The garage-nvidia Docker image uses nvidia/cuda:11.0-cudnn8-runtime-ubuntu18.04 as the parent image since TensorFlow requires CUDA 11.0 & cuDNN 8.0. If you need to use garage with a different CUDA version, you might be able to build the garage-nvidia image using a different parent image by following the guide at Building garage Docker image from source

This page was authored by Gitanshu Sardana (@gitanshu).