Build garage Docker image from source¶
Garage source comes with a Makefile that contains recipes for building and running different Docker configurations for garage
Garage uses multi-stage Docker builds with the Docker BuildKit backend. The
BuildKit backend is opt-in and needs to be enabled by setting environment
variable DOCKER_BUILDKIT=1
in your shell. The Makefile takes care of that
for you.
The important Docker related make
targets are:
run-dev
: builds and runs the Docker image with your copy of garage source installed. This builds thegarage-dev
target in Dockerfile and the resulting image is tagged asrlworkgroup/garage-dev
run-dev-nvidia
: same asrun-dev
with CUDA 11.0 and cuDNN 8.0 for taking advantage of NVIDIA GPUs and also supports environment visualization. The build target isgarage-dev-nvidia
and the resulting image is tagged asrlworkgroup/garage-dev-nvidia
run-dev-nvidia-headless
: same asrun-dev-nvidia
but without support for environment visualization. Suitable for running in headless mode for machines without a display.
Prerequisites¶
Be aware of the following prerequisites to build 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.
Build and run the garage-dev
image¶
To build and run the headless image, first clone the garage repository, move to the root folder of your local repository and then execute;
make run-dev RUN_CMD="python examples/tf/trpo_cartpole.py"
Where RUN_CMD
specifies the executable to run in the container.
The previous command adds a volume from the data
folder inside your cloned
garage repository to the data
folder in the garage container, so any
experiment results ran in the container will be saved in the data
folder
inside your cloned repository. The Makefile uses the same username and uid as
your current local account to create the default user in the Docker images.
This keeps things simple by allowing the Docker user to write to the data
directory without giving explicit permission.
By default, Docker generates random names for containers. If you want to specify
a name for the container, you can do so with the variable CONTAINER_NAME
. As a
side effect, this will output the results in data/$CONTAINER_NAME
directory
instead of the data
directory.
make run-dev RUN_CMD="..." CONTAINER_NAME="my_container_123"
This will output results in data/my_container_123
directory.
If you need to use MuJoCo, you need to place your key at ~/.mujoco/mjkey.txt
or specify the corresponding path through the MJKEY_PATH variable:
make run-dev RUN_CMD="..." MJKEY_PATH="/home/user/mjkey.txt"
If you require to pass additional arguments to docker build and run commands, you can use the variables BUILD_ARGS and RUN_ARGS, for example:
make run-dev BUILD_ARGS="--build-arg MY_VAR=123" RUN_ARGS="-e MY_VAR=123"
Prerequisites for NVIDIA image¶
Additional to the prerequisites for the garage
image, make sure to have:
Install the latest NVIDIA driver, tested on nvidia driver version 455.45.01. CUDA 11.0 requires a minimum of version 450.36.06. If you need a different CUDA version, see Using a different CUDA version
Tested on Ubuntu 18.04 & 20.04.
Build and run the NVIDIA image¶
The same rules for the headless image apply here, except that the target name is:
make run-dev-nvidia ...
This make command builds the NVIDIA image and runs it in a non-headless mode. It will not work on headless machines. You can run the NVIDIA in a headless state using the following target:
make run-dev-nvidia-headless ...
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 using the variable GPUS. For
example:
make run-dev-nvidia GPUS="device=0,2" ...
Using a different CUDA version¶
The garage-nvidia
Docker image uses nvidia/cuda:11.0-cudnn8-runtime-ubuntu18.04
as the parent image which requires NVIDIA driver version 450.36.06+. If you need
to use garage with a different CUDA version, you might be able to build the
garage-nvidia
image from scratch using a different parent image using the
variable PARENT_IMAGE
.
make run-dev-nvidia PARENT_IMAGE="nvidia/cuda:11.0-cudnn8-runtime-ubuntu18.04" ...
You can find the required parent images at NVIDIA CUDA’s DockerHub
This page was authored by Gitanshu Sardana (@gitanshu), with contributions from Angel Ivan Gonzalez (@gonzaiva)