Docker

Dockerfile

To create a Docker container, we need to construct a Dockerfile which contains instructions on which base image to use, and installation rules.

In the same directory where you previously created the Conda YAML file, copy the following file and save it as a Dockerfile:

FROM nfcore/base:1.14
LABEL authors="Barry Digby" \
      description="Docker container containing fastqc"

WORKDIR ./
COPY environment.yml ./
RUN conda env create -f environment.yml && conda clean -a
ENV PATH /opt/conda/envs/test_env/bin:$PATH

We are using a pre-built ubuntu image (FROM nfcore/base:1.14) that comes with Conda pre-installed developed by nf-core.

Note

In your Dockerhub account, create a repository called ‘test’. We will build and push the docker image in the following section.

Build image

To build the image, run the following command:

$ docker build -t USERNAME/test $(pwd)

Check image

You can shell into your image to double check that the tools have been installed correctly:

$ docker images # check images in cache

$ docker run -it barryd237/test

Push to Dockerhub

Now the image has been created, push to Dockerhub:

First time push requires you to login:

$ docker login

$ sudo chmod 666 /var/run/docker.sock

$ (sudo)?? docker push USERNAME/test

Advanced use

There will be scenarios in which your tool of choice is not in the Anaconda repository meaning you cannot download it via the environment.yml file.

You will have to provide install instructions to the Dockerfile.

Note

This is fairly tedious, you have to perform a dry-run locally before providing the instructions to the Dockerfile.

Let’s pretend that Bowtie2 is not available via the Anaconda repository - go to the Github repository containing the latest release: https://github.com/BenLangmead/bowtie2

  1. Download the lastest release (2.4.X) of Bowtie2. Make sure to download the Source code (tar.gz) file.

  2. Untar the archive file by running tar -xvzf v2.4.5.tar.gz.

  3. Move to the unzipped directory and figure out if you need to compile the source code. (There is a Makefile present - we do need to compile the code).

  4. In the bowtie2-2.4.5/ directory, run the command make to compile the code.

  5. Do you need to change permissions for the executables?

  6. Move the executables to somewhere in your $PATH. This can be done two ways:

    1. By moving the executables to a directory in your $PATH such as /usr/local/bin, /usr/bin etc like so: sudo mv bowtie2-2.4.5/bowtie2* /usr/local/bin/.

    2. By manually adding a directory to your $PATH: export PATH="/data/bowtie2-2.4.5/:$PATH".

  7. Test the install by printing the documentation: bowtie2 -h

You will need to perform each of the above tasks in your Dockerfile - which is done ‘blind’ hence the need for a dry-run.

Note

Whilst the nf-core image we are using contains a handful of tools, containers are usually a clean slate. You have to install basics such as unzip, curl etc..

FROM nfcore/base:1.14
LABEL authors="Barry Digby" \
      description="Docker container containing stuff"

# We need to install tar
RUN apt-get update; apt-get clean all; apt-get install --yes tar

# Install our conda environment, if you want to.
WORKDIR ./
COPY environment.yml ./
RUN conda env create -f environment.yml && conda clean -a
ENV PATH=/opt/conda/envs/test_env/bin:$PATH

# Make a 'scratch' directory.
RUN mkdir -p /usr/src/scratch
# Set scratch directory as working directory (where we will download the source code to)
WORKDIR /usr/src/scratch
# Download the source code
RUN wget https://github.com/BenLangmead/bowtie2/archive/refs/tags/v2.4.5.tar.gz
# untar the source code
RUN tar -xvzf v2.4.5.tar.gz
# Compile the source code
RUN cd bowtie2-2.4.5/ && make
# Add the executable directory to your path
ENV PATH=/usr/src/scratch/bowtie2-2.4.5/:$PATH