Docker Development Environment

Fast, reproducible, and portable software development environments

Copyright (C) 2021 Bernd Doser, bernd.doser@braintwister.eu

All rights reserved.

BrainTwister docker-devel-env is free software made available under the MIT License. For details see the license file.

Advantages

• Fast build and execution compared to virtual machines
• Portability: Same environment on different machines, platforms, and operating systems
• Reproducible behaviors
• Economical consumption of resources
• Identical environment for development IDE and continuous integration
• Easy provisioning of images

Docker images

Each directory correspond to an environment module. They can stick together as a chain:

module1 - module2 - module3 - ...

The image module1-module2-module3 is using the image module1-module2 as base, which will be set using the build-time variable BASE_IMAGE. For example the image ubuntu-20.04-clang-12 will be build with

cd clang-12
docker build -t braintwister/ubuntu-20.04-clang-12 --build-arg BASE_IMAGE=braintwister/ubuntu-20.04 .

Please find a list of available images at images.yml. The images in the list will be build automatically with Jenkins and pushed to DockerHub.

The docker images can be pulled with

docker pull braintwister/<image-name>

Eclipse IDE

Eclipse IDE for C++ development

A ready-for-action eclipse IDE with

• clang
• CMake
• conan.io

installed can be started by

docker run -d -v /tmp/.X11-unix:/tmp/.X11-unix:ro -e DISPLAY --privileged \
  braintwister/ubuntu-20.04-clang-12-eclipse-cpp-2021.09

or using docker-compose by

version: "3"
services:

  eclipse:
    image: braintwister/ubuntu-20.04-clang-12-eclipse-cpp-2021.09
    volumes:
      - /tmp/.X11-unix:/tmp/.X11-unix:ro
    environment:
      - DISPLAY
    privileged: true

The mount of the X11 socket file (/tmp/.X11-unix) and the definition of the environment variable DISPLAY induce the application within the container to send the rendering instructions to the host X server. To allow the container to use the host display, the command xhost +local: must be executed on the host before starting the container. The privileged mode is needed for debugging with gdb.

Eclipse IDE for CUDA development

First of all nvidia-docker version 2 must be installed and the runtime attribute must be set to nvidia, that the container get access to the host GPU card. The nvidia runtime attribute is currently only available at docker-compose version 2.3.

For CUDA development the NVIDIA IDE nsight is highly recommended, because it provides special support for code editing, debugging, and profiling. The version of nsight is not adjustable, as it depends to the version of the cuda module.

version: "2.3"
services:

  eclipse:
    image: braintwister/cuda-devel-11.4.2-clang-12-nsight
    runtime: nvidia
    volumes:
      - /tmp/.X11-unix:/tmp/.X11-unix:ro
    environment:
      - DISPLAY
    privileged: true

Eclipse IDE for embedded development

For embedded programming you have to bind the host serial port (here: /dev/ttyACM0) to get a connection to the embedded platform (Arduino, ESP32, ...).

version: "3"
services:

  eclipse:
    image: braintwister/ubuntu-20.04-clang-12-eclipse-cpp-2021.09
    volumes:
      - /tmp/.X11-unix:/tmp/.X11-unix:ro
      - /dev/ttyACM0:/dev/ttyACM0
    environment:
      - DISPLAY
    privileged: true

Visual Studio Code

The Visual Studio Code IDE can be started by using

docker run -d -v /tmp/.X11-unix:/tmp/.X11-unix:ro -e DISPLAY --privileged \
  braintwister/ubuntu-20.04-clang-12-vscode-1.62.3

Persistent storage

The data in the container can be made persistent by using a docker volume home for the home directory /home/user.

version: "3"
services:

  eclipse:
    image: braintwister/ubuntu-20.04-clang-12-eclipse-cpp-2021.09
    volumes:
      - /tmp/.X11-unix:/tmp/.X11-unix:ro 
      - home:/home/user
    environment:
      - DISPLAY
    privileged: true

volumes:
  home:

Project-assigned development environment

The docker development environment can be directly stored within the source code repository and is able to bind the working directory of the source code into the development container. Therefore, the user in the container must be the owner of the source code working directory on the host. The user in the container can be set with the environment variables USER_ID, GROUP_ID, USER_NAME, and GROUP_NAME. In the following example the docker-compose file is stored in the root directory of a git repository. Starting docker-compose up -d in the root directory the current directory . will be bound to /home/${USER_NAME}/git/${PROJECT}. It is recommended to set the variables in an extra file .env, which is not controlled by the source control management, so that the docker-compose file must not be changed.

version: "3"
services:

  vscode:
    image: braintwister/ubuntu-20.04-clang-12-vscode-1.62.3
    volumes:
      - /tmp/.X11-unix:/tmp/.X11-unix:ro 
      - home:/home/${USER_NAME}
      - .:/home/${USER_NAME}/git/${PROJECT}
    environment:
      - DISPLAY
      - USER_ID=${USER_ID}
      - GROUP_ID=${GROUP_ID}
      - USER_NAME=${USER_NAME}
      - GROUP_NAME=${GROUP_NAME}
    privileged: true

volumes:
  home:

The .env-file can be generated by

cat << EOT > .env 
PROJECT=`basename "$PWD"`
USER_ID=`id -u $USER`
GROUP_ID=`id -g $USER`
USER_NAME=`id -un $USER`
GROUP_NAME=`id -gn $USER`
EOT

Jenkins build container

A declarative Jenkinsfile can look like

pipeline {

  agent {
    docker {
      image 'braintwister/ubuntu-20.04-clang-12'
    }
  }

  stages {
    stage('Conan') {
      steps {
        sh 'conan install .'
      }
    }
    stage('CMake') {
      steps {
        sh 'cmake .'
      }
    }
    stage('Build') {
      steps {
        sh 'make all'
      }
    }
    stage('Test') {
      steps {
        sh 'make test'
      }
    }
  }
}

TensorFlow

For machine learning development we provide with an installation of the open-source framework TensorFlow using the latest cuda development drivers.

Although the usage of GPUs is highly recommended braintwister/cuda-devel-11.4.2-tensorflow-gpu-2.0, a CPU version is also available braintwister/ubuntu-20.04-tensorflow-2.0.

Start a plain container with

docker run -it --runtime=nvidia braintwister/cuda-devel-11.4.2-tensorflow-gpu-2.0

TensorBoard is available at localhost:6006, if -p 6006:6006 was added to the docker run command and tensorboard was launched within the container.

TensorFlow with Visual Studio Code

To allow the container to use the host display, the command xhost +local: must be executed on the host before starting the container.

docker run -d --runtime=nvidia -e DISPLAY \
  braintwister/cuda-devel-11.4.2-tensorflow-gpu-2.0-vscode-1.62.3

TensorFlow with Jupyter

Start the container with

docker run --runtime=nvidia -p 8888:8888 \
  braintwister/cuda-devel-11.4.2-tensorflow-gpu-2.0-jupyter-1.0

and open localhost:8888 on your host browser.