A simple reproducible analysis with Docker and git

04 Oct 2017

Initialise the repository and connect it to remote:

$ git init
$ git remote add origin git@gitlab.ethz.ch:mrawlik/1st-order-SFC-design.git

Create a README.md file.

$ git add README.md
$ git commit -m "Initial commit."
$ git push -u origin master

Write a program. It will accept a single argument: the path where to write its output.

OUTPATH = ARGS[1]

println("hello world")
write(joinpath(OUTPATH, "output.txt"), "hello world")

Add it to git:

$ git add hello.jl
$ git commit -m "Add a hello program."

You can test it locally:

$ mkdir output
$ julia hello.jl output
hello world
$ cat output/output.txt
hello world

Now pack the program in a Docker container. Create a Dockerfile:

# specify the version
FROM julia:0.6

# add the contents of the current directory to docker, in particular the hello
# program
ADD . /

# prepare an output directory
# we could also have an input one
# VOLUME /input
VOLUME /output

# what to run
CMD julia hello.jl /output

Build the docker image with a friendly name. The first time it may take some time, as docker needs to fetch the base julia:0.6 image.

$ docker build -t hello .

Run the analysis inside docker. The first argument to --volume needs to be an absolute path. In fish do:

$ docker run --volume (pwd)/output:/output hello

In bash:

$ docker run --volume `pwd`/output:/output hello

Michał Rawlik

A simple reproducible analysis with Docker and git