Docker for Clawpack

Rather than installing Clawpack and all its dependencies on your computer, if you have Docker installed then you can now use a docker image from the DockerHub Clawpack repositories.

To download an image:

$ docker pull clawpack/v5.7.1_dockerimage:release

To create a container and run it:

$ docker run -i -t -p 8889:8889 --name clawpack-v5.7.1_container \
    clawpack/v5.7.1_dockerimage:release

You can change the container name if you wish, and also the port 8889 (on which jupyter notebooks might be served, see below).

You should now see a prompt like:

jovyan $

indicating that you are in the container, logged in as user jovyan.

Once logged in to the container, you should find a directory $HOME/clawpack-v5.7.1 that contains the Clawpack installation (including the current master branch of the Clawpack Applications repository).

A better image for GeoClaw users

Note: Starting with v5.7.1 there is only a single docker image, which now also includes some packages of primary interest to GeoClaw users.

Stopping a container

You can exit a container (after using ctrl-C to quit the jupyter server if one is running) via:

exit

at the jovyan $ prompt.

Restarting a container

You can restart the container via:

docker start -a -i clawpack-v5.7.1_container

Running Jupyter notebooks

The form of run command suggested above also allows you to run Jupyter notebooks from port 8889 on your own computer (or whatever port you specified when creating the container).

To start the sever, in the docker container (at the jovyan $ prompt) run this command:

jupyter notebook --ip=0.0.0.0 --port=8889 --no-browser

Then open a browser (on the host machine) to:

http://localhost:8889/?token=TOKEN

replacing TOKEN with the token that should have printed out when you started the server.

This will open to the top level of $HOME, and you can then navigate to wherever the notebooks are you want to run, e.g. the sample ones in the apps repository can be found at clawpack-v5.7.1/apps/notebooks.

PyClaw users might want to start with clawpack-v5.7.1/apps/notebooks/pyclaw/Acoustics-1D.ipynb.

GeoClaw users might want to try running clawpack-v5.7.1/apps/notebooks/geoclaw/chile2010a.ipynb, which exercises most aspects of GeoClaw.

Moving files between the docker container and the host machine

Often you want to run the code on Docker and then transfer the resulting output files, and/or the plots generated, back to the host machine (e.g. some directory on your laptop). You can use the –volume flag when running a container to accomplish this, see docker volume documentation.

For example, if you have created a directory $HOME/docker_disk on your computer then adding:

-v $HOME/docker_disk:/home/jovyan/work

to your docker run command will map this directory to /home/jovyan/work in the docker container. So you can move Clawpack output or plots to that directory in order to have access to them from your host computer.

Putting this together with previous options, here’s a sample command that creates and runs a geoclaw-based container with this mapping and also allowing us to start a Jupyter server:

$ docker run -i -t -p 8889:8889 -v ~/docker_disk:/home/jovyan/work \
  --name clawpack-v5.7.1_geoclaw_container \
  clawpack/v5.7.1_geoclaw_dockerimage

Some other useful docker commands

See the docker command line documentation or any of the tutorials available on-line for more details, but here are a few particularly useful commands:

docker help
docker info

docker ps -a  # list all containsers
docker rm clawpack-v5.7.1_container  # remove a container

docker images -a  # list all images
docker rmi clawpack/v5.7.1_dockerimage:release  # remove an image
docker prune  # remove all images not used by any container

Creating your own docker image

If you want to create a new docker image that includes other software in addition to Clawpack, you can find the Dockerile used to create the docker image on dockerhub in the repository https://github.com/clawpack/docker-files.

This might be useful if you want to distribute your own code that depends on Clawpack in a form that’s easy for others to use.

You can also create a Dockerfile that uses the already-build Clawpack 5.7.1 on Dockerhub by starting the Dockerfile with:

FROM clawpack/v5.7.1_dockerimage:release

and then adding anything addition you want in the image, such as other Python modules you need or your own application code. You may need to specify USER root in order to install some things, and then switch back to USER jovyan at the end. For an example, see how clawpack/docker-files/Dockerfile_v5.7.0_geoclaw is built on top of clawpack/v5.7.0_dockerimage:release.

Dockerfiles for binder

The username jovyan was chosen so that you can use this docker image also for starting up a Jupyter notebook server on binder. You can do this by including a simple Dockerfile at the top level of your repository that uses the dockerhub image, as above. See this repository for a simple example: https://github.com/rjleveque/test_binder.

The repository for the book Riemann Problems and Jupyter Solutions also uses this approach.

See the binder documentation for more details on using Dockerfiles there.