Towards reproducible builds with Docker development containers

Create your own development environment

• create you own development environment by installing all required packages, libraries, dependencies, environmental variables, paths, and anything required to build your project by adding stuff into the devctr/Dockerfile
  • the provided Dockerfile already contains cuda:11.8.0 toolchain and python3
• build your own development environment: ./devtool build_devctr
• (OPTIONALLY) If you want to share your development environment with other people:
  • register on Dockerhub
  • change DOCKER_HUB_USERNAME in devtool to your username
  • push changes with docker push <your username>:6_s894_finalproject_devctr:latest

Now you can build things inside this development container anywhere, without "it works on my machine" issues anymore and without installing anything (which might be very complicated and messy sometimes) on your physical machine (host).

Building your projects and (OPTIONAL) executing via Telerun

• put your project you want to build inside src folder
• write what should be called to build your project in src/build.sh
  • if you want to add some test input files along your binaries, write instructions in build.sh as well
• (OPTIONAL) write what should be executed on the GPU server in src/run.sh
• build it: ./devtool build_project
• the output can be found in build folder and all together in build.tar
• (OPTIONAL) build.tar is shippable for execution via Telerun:
  • python3 <path_to_telerun.py> submit build.tar
  • Telerun will execute your run.sh