This guide is meant to accompany the instructions for using Singularity containers in the Open Science Grid. You can use your own custom container to run jobs in the Open Science Grid, and we assume that those containers are built using Docker. This guide describes how to create your own Docker container "image" (the blueprint for the container). Once you have created your custom image, you will need to register the image as described in this page
Install Docker and Get a Docker Hub Account
You'll need a Docker Hub account in order to download Docker and share your Docker container images with the OSG: DockerHub
Install Docker Desktop to your computer using the appropriate version for your operating system.
What software do you want to install? Make sure that you have either the source
code or a command that can be used to install it through Linux (like
You'll also need to choose a "base" container, on which to add your particular software or tools. We strongly recommend using one of the OSG's published containers as your starting point. See the available containers on Docker Hub here: OSG Docker Containers The best candidates for you will be containers that have "osgvo" in the name.
If you prefer, you can base your image on images not already published
by OSG, but if you do this, we recommend that as one of the creation steps you
/cvmfs directory. See Special Cases below.
Build a Container
There are two main methods for generating your own container image.
- Editing the
- Editing the default image using local Docker
We recommend the first option, as it is more reproducible, but the second option can be useful for troubleshooting or especially tricky installs.
Create a folder on your computer and inside it, create a blank text file
The first line of this file should include the keyword
FROM and then
the name of a Docker image (from Docker Hub) you want
to use as your starting point. If using the OSG's Ubuntu Xenial image that
would look like this:
Then, for each command you want to run to add libraries or software, use the
RUN and then the command. Sometimes it makes sense to string
commands together using the
&& operator and line breaks
\, like so:
RUN apt-get update && \ apt-get install -yy build-essentials
RUN wget https://cran.r-project.org/src/base/R-3/R-3.6.0.tar.gz && \ tar -xzf R-3.6.0.tar.gz && \ cd R-3.6.0 && \ ./configure && \ make && \ make install
Typically it's good to group together commands installing the same kind of thing
(system libraries, or software packages, or an installation process) under one
and then have multiple
RUN commands, one for each of the different type of
software or package you're installing.
(For all the possible Dockerfile keywords, see the Docker Documentation)
Once your Dockerfile is ready, you can "build" the container image by running this command:
$ docker build -t namespace/repository_name .
Note that the naming convention for Docker images is your Docker Hub username and then
a name you choose for that particular container image. So if my Docker Hub username
alice and I created an image with the NCBI
blast tool, I might use this name:
$ docker build -t alice/NCBI-blast .
Editing the default image using local Docker
You can also build an image interactively, without a Dockerfile. First, get the desired starting image from Docker Hub. Again, we will look at the OSG Ubuntu Xenial image.
$ docker pull opensciencegrid/osgvo-ubuntu-xenial
We will run the image in a docker interactive session
$ docker run -it --name <docker_session_name_here> opensciencegrid/osgvo-ubuntu-xenial /bin/bash
Giving the session a name is important because it will make it easier to reattach the session later and commit the changes later on. Now you will be greeted by a new command line prompt that will look something like this
You can now install the software that you need through the default package
manager, in this case
[root@740b9db736a1 /]# apt-get install build-essentials
Once you have installed all the software, you simply
[root@740b9db736a1 /]# exit
Now you can commit the changes to the image and give it a name:
docker commit <docker_session_name_here> namespace/repository_name
You can also use the session's hash as found in the command prompt (
in the above example) in place of the docker session name.
Upload Docker Container to Docker Hub
Once your container is complete and tagged, it should appear in the list of local Docker container images, which you can see by running:
$ docker images
From there, you need to put it in Docker Hub, which can be done via the
$ docker push namespace/repository_name
From here, if you're planning to use this container in OSG, return to our Containers in OSG Guide to learn how to upload your container to the OSG's container repository.
If you want your jobs to access CVMFS, make sure that you either:
- Use one of the base containers provided by the Open Science Grid
- Add a
/cvmfsfolder to your container:
- If using a Dockerfile, you can do this with the line
RUN mkdir /cvmfs
- If building your container interactively, run
$ mkdir -p /cvmfs
This will enable the container to access
tools and data published on
If you do not want
in the container, please add
+SingularityBindCVMFS = False to your job submit file.
ENTRYPOINT and ENV
Two options that can be used in the Dockerfile to set the environment or
default command are
ENV. Unfortunately, both of these
aspects of the Docker container are deleted when it is converted to a
Singularity image in the Open Science Grid. Email us if you would like
to preserve these attributes.
Configuring GPU images
GPU images requires a few extra steps to get access to the host GPU libraries.
We recommend that in this case you use a
Dockerfile rather than building
a image interactively.
GPU containers running under GPU slots will automatically get the host GPU
libraries bound to
/host-libs inside the container. For this to work,
/host-libs directory has to exist in the image, and
has to be set for the job. An example of how to do this is the
TensorFlow GPU image.
This page was updated on Mar 31, 2020 at 13:45 from start/software/container-how-to.md.