search

DataBiosphere / terra-docker

BSD 3-Clause "New" or "Revised" License
27 stars 35 forks source link

readme

This repo provides docker images for running jupyter notebook in Terra

Contributing

Make sure to go through the contributing guide as you make changes to this repo.

Terra Base Images

terra-jupyter-base

terra-jupyter-python

terra-jupyter-r

terra-jupyter-hail

terra-jupyter-gatk

terra-jupyter-bioconductor

How to create your own Custom image to use with notebooks on Terra

Custom docker images need to use a Terra base image (see above) in order to work with the service that runs notebooks on Terra.

Development

Using git secrets

Make sure git secrets is installed:

brew install git-secrets

Ensure git-secrets is run: If you use the rsync script to run locally you can skip this step

cp -r hooks/ .git/hooks/
chmod 755 .git/hooks/apply-git-secrets.sh

Run/developing smoke_test.ipynb file locally

Run your image locally with the repo directory mounted into the container. For example

docker run -d -p <port_number>:8000 -v <your_local_path_to_the_repo>/terra-docker:/home/jupyter -it us.gcr.io/broad-dsp-gcr-public/terra-jupyter-r:test

Once you have the container running, you should be able to access jupyter at http://localhost:/notebooks. You should be able to navigate to the smoke test ipynb file you're interested in, and run a cell. After you modify a smoke test .ipynb file, go to Cell -> All Ouput -> Clear to clear all outputs to keep the .ipynb files smaller.

Generate New Image or Update Existing Image

Detailed documentation on how to integrate the terra-docker image with Leonardo can be found here

If you are adding a new image:

If you are updating an existing image:

Testing your image manually

Build the image: run docker build [your_dir] -t [name].

docker build terra-jupyter-base -t terra-jupyter-base

If you're on an M1 and building an image from a locally built image, replace the current FROM command:

FROM --platform=linux/amd64 terra-jupyter-base

It is not advised to run build.sh locally, as this will push to the remote docker repo and delete the image locally upon completion.

All images can be run locally. For example:

docker run --rm -it -p 8000:8000 us.gcr.io/broad-dsp-gcr-public/terra-jupyter-base:0.0.7

Then navigate a browser to http://localhost:8000/notebooks to access the Jupyter UI.

You can gain root access and open a bash terminal as follows:

docker run --rm -it -u root -p 8000:8000 --entrypoint /bin/bash us.gcr.io/broad-dsp-gcr-public/terra-jupyter-base:0.0.7

Running locally is conventient for quick development and exploring the image. However it has some limitations compared to running through Terra. Namely:

To launch an image through Terra, navigate to https://app.terra.bio or your BEE's UI, select a workspace, enter your new image in the "Custom Image" field, and click Create.

Automation Tests

Here are automation tests for various docker image, please update the image hash for relevant tests. You can run the job build-terra-docker to automatically create a PR with your branch if you manually specify versions.

Config

There is a config file located at config/conf.json that contains the configuration used by all automated jobs and build scripts that interface with this repo.

There is a field for "spark_version" top-level which must be updated if we update the debian version used in the custom image. Currently it assumes 1.4x https://cloud.google.com/dataproc/docs/concepts/versioning/dataproc-release-1.4

There are some constants included, such as the tools supported by this repo. Of particular interest is the image_data array.

Each time you update or add an image, you will need to update the appropriate entry in this array:

{
    "name": "terra-jupyter-base", //the name of the image. should correspond to the directory it is located

    "base_label": "Minimal",      //the base name used in the UI for this image. This is appended with some information about the packages in this image.

    "tools": ["python"],          //the tools present in this image. see the top-level "tools" array for valid entries. 
                                  //The significance of 'tools' is that there is expected to be an entry in the documentation specifying the version of this tool
                                  //If you wish to add a tool, you will need to add a handler to the function get_doc_builder in generate_package_documentation.py

    "packages": { "python": ["pandas"] },               //The packages that we wish to single out to display to the user at a later date. 
                                  //The difference between a package and a tool is that a tool can have a set of packages associated with it (i.e. pip packages for python)
                                  // A package  must have a tool associated with it
    "version": "0.0.4",           //The current version the image is at

    "automated_flags": {          //Flags used as control flow for scripts

        "generate_docs": true,    //Whether documentation should be auto-generated for this image. This is superceded by the build flag (i.e. if build=false, this flag is ignored)

        "build": true,            //Whether or not the jenkins job that builds the docker images in this repo should build this image

        "include_in_custom_dataproc": true,  //Whether or not the jenkins job that builds the custom dataproc image should include this image. 
                                            //This is superceded by the build flag
        "include_in_ui": true, // Whether or not this should be included in the .json file that power the terra ui dropdown for available images
        "include_in_custom_gce": true, //Whether or not the jenkins job that builds the custom gce image should include this image.
                                 //This is superceded by the build flag   
        "requires_spark": true // Whether or not this image requires a dataproc cluster to run (as opposed to most, which just need a GCE VM)
    }   
},

Scripts

The scripts folder has scripts used for building.

Image dependencies

Note that this dependency graph needs to be updated! Image dependencies

Push Images to GCR

To push images to Broad managed Google Container Registrygcr.io/broad-dsp-gcr-public, manually trigger Publish image to GCR Github action and choose the image to push.