Options
All
  • Public
  • Public/Protected
  • All
Menu

@stencila/dockta

Dockta: a container image builder for researchers

All Contributors Build status Code coverage NPM Docs Chat

Docker is a useful tool for creating reproducible computing environments. But creating truly reproducible Docker images can be difficult - even if you already know how to write a Dockerfile.

Dockta makes it easier for researchers to create Docker images for their research projects. Dockta generates a Dockerfile and builds a image, for your project, based on your source code.

Features

πŸ¦„ Features that are planned, but not yet implemented, are indicated by unicorn emoji. Usually they have a link next to them, like this πŸ¦„ #2, indicating the relevant issue where you can help make the feature a reality. It's readme driven development with calls to action to chase after mythical vaporware creatures! So hip.

Builds a Docker image based on your source code

Dockta scans your project and builds a custom Docker image for it. If the the folder already has a Dockerfile, Dockta will build the image from that. If not, Dockta will scan the source code files in the folder and generate one for you. Dockta currently handles R, Python and Node.js source code. A project can have a mix of these languages.

R

If the folder contains a R package DESCRIPTION file then Dockta will install the R packages listed under Imports into the image. e.g.

Package: myrproject
Version: 1.0.0
Date: 2017-10-01
Imports:
   ggplot2

The Package and Version fields are required in a DESCRIPTION file. The Date field is used to define which CRAN snapshot to use. MRAN daily snapshots began 2014-09-08 so the date should be on or after that.

If the folder does not contain a DESCRIPTION file then Dockta will scan all the R files (files with the extension .R or .Rmd) in the folder for package import or usage statements, like library(package) and package::function(), and create a .DESCRIPTION file for you.

Python

If the folder contains a requirements.txt file, or a πŸ¦„ #4 Pipfile, Dockta will copy it into the Docker image and use pip to install the specified packages.

If the folder does not contain either of those files then Dockta will scan all the folder's .py files for import statements and create a .requirements.txt file for you.

Node.js

If the folder contains a package.json file, Dockta will copy it into the Docker image and use npm to install the specified packages.

If the folder does not contain a package.json file, Dockta will scan all the folder's .js files for require calls and create a .package.json file for you.

JATS

If the folder contains any JATS files (.xml files with <!DOCTYPE article PUBLIC "-//NLM//DTD JATS (Z39.96) ...), πŸ¦„ #52 Docker will scan reproducible elements defined in the Dar JATS extension for any package import statements (e.g. Python import, R library, or Node.js require) and install the necessary packages into the image.

Jupyter

If the folder contains any Jupyter .ipynb files, πŸ¦„ #9 Dockta will scan the code cells in those files for any package import statements (e.g. Python import, R library, or Node.js require) and install the necessary packages into the image. It will also πŸ¦„ #10 add the necesary Jupyter kernels to the built Docker image.

Automatically determines system requirements

One of the headaches researchers face when hand writing Dockerfiles is figuring out which system dependencies your project needs. Often this involves a lot of trial and error.

Dockta automatically checks if any of your dependencies (or dependencies of dependencies, or dependencies of...) requires system packages and installs those into the image. For example, let's say you have a project with an R script that requires the rgdal package for geospatial analyses,

library(rgdal)

When you run dockta compile in this project, Dockta will generate a Dockerfile with the following section which installs R, plus the three system dependencies required gdal-bin, libgdal-dev, and libproj-dev:

# This section installs system packages required for your project
# If you need extra system packages add them here.
RUN apt-get update \
 && DEBIAN_FRONTEND=noninteractive apt-get install -y \
      gdal-bin \
      libgdal-dev \
      libproj-dev \
      r-base \
 && apt-get autoremove -y \
 && apt-get clean \
 && rm -rf /var/lib/apt/lists/*

For R, Dockta does this by querying the https://sysreqs.r-hub.io/ database. For Python, Dockta includes a mapping of system requirements of packages that users can contribute to.

No more trial and error of build, fail, add dependency, repeat... cycles!

Faster re-installation of language packages

If you have built a Docker image before, you'll know that it can be frustrating waiting for all your project's dependencies to reinstall when you simply add or remove one of them.

The reason this happens is that, due to Docker's layered filesystem, when you update a requirements file, Docker throws away all the subsequent layers - including the one where you previously installed your dependencies. That means that all those packages need to get reinstalled.

Dockta takes a different approach. It leaves the installation of language packages to the language package managers: Python's pip , Node.js's npm, and R's install.packages. These package managers are good at the job they were designed for - to check which packages need to be updated and to only update them. The result is much faster rebuilds, especially for R packages, which often involve compilation.

Dockta does this by looking for a special # dockta comment in a Dockerfile. Instead of throwing away subsequent image layers, it executes all instructions after this comment in the same layer - thus reusing packages that were previously installed.

Here's a simple motivating example. It's a Python project with a requirements.txt file which specifies that the project depends upon pandas which, to ensure reproducibility, is pinned to version 0.23.0,

pandas==0.23.0

The project also has a Dockerfile which specifies which Python version we want to use, copies requirements.txt into the image, and uses pip to install the packages:

FROM python:3.7.0

COPY requirements.txt .
RUN pip install -r requirements.txt

You could build a Docker image for that project using Docker,

docker build .

Docker will download the base Python image (if you don't yet have it), download five packages (pandas and it's four dependencies) and install them. This took over 9 minutes when we ran it.

Now, let's say that we want to get the latest version of pandas and increment the version in the requirements.txt file,

pandas==0.23.1

When we do docker build . again to update the image, Docker notices that the requirements.txt file has changed and so throws away that layer and all subsequent ones. This means that it will download and install all the necessary packages again, including the ones that we previously installed. For a more contrived illustration of this, simply add a space to one of the lines in the requirements.txt file and notice how the package install gets repeated all over again.

Now, let's add a special # dockta comment to the Dockerfile before the COPY directive,

FROM python:3.7.0

# dockta

COPY requirements.xt .
RUN pip install -r requirements.txt

The comment is ignored by Docker but tells dockta to run all subsequent instructions in a single filesystem layer,

dockta build .

Now, if you change the requirements.txt file, instead of reinstalling everything again, pip will only reinstall what it needs to - the updated pandas version. The output looks like:

Step 1/1 : FROM python:3.7.0
 ---> a9d071760c82
Successfully built a9d071760c82
Successfully tagged dockta-5058f1af8388633f609cadb75a75dc9d:system
Dockta 1/2 : COPY requirements.txt requirements.txt
Dockta 2/2 : RUN pip install -r requirements.txt
Collecting pandas==0.23.1 (from -r requirements.txt (line 1))

  <snip>

Successfully built pandas
Installing collected packages: pandas
  Found existing installation: pandas 0.23.0
    Uninstalling pandas-0.23.0:
      Successfully uninstalled pandas-0.23.0
Successfully installed pandas-0.23.1

Generates structured meta-data for your project

Dockta uses JSON-LD as it's internal data structure. When it parses your project's source code it generates a JSON-LD tree using a vocabularies from schema.org and CodeMeta.

For example, It will parse a Dockerfile into a schema.org SoftwareSourceCode node extracting meta-data about the Dockerfile.

Dockta also fetches meta data on your project's dependencies, which could be used to generate a complete software citation for your project.

{
  "name": "myproject",
  "datePublished": "2017-10-19",
  "description": "Regression analysis for my data",
  "softwareRequirements": [
    {
      "description": "\nFunctions to Accompany J. Fox and S. Weisberg,\nAn R Companion to Applied Regression, Third Edition, Sage, in press.",
      "name": "car",
      "urls": [
        "https://r-forge.r-project.org/projects/car/",
        "https://CRAN.R-project.org/package=car",
        "http://socserv.socsci.mcmaster.ca/jfox/Books/Companion/index.html"
      ],
      "authors": [
        {
          "name": "John Fox",
          "familyNames": [
            "Fox"
          ],
          "givenNames": [
            "John"
          ]
        },

Easy to pick up, easy to throw away

Dockta is designed to make it easier to get started creating Docker images for your project. But it's also designed not to get in your way or restrict you from using bare Docker. You can easily, and individually, override any of the steps that Dockta takes to build an image.

  • Code analysis: To stop Dockta doing code analysis and take over specifying your project's package dependencies, just remove the leading '.' from the .DESCRIPTION, .requirements.txt or .package.json file that Dockta generates.

  • Dockerfile generation: Dockta aims to generate readable Dockerfiles that conform to best practices. They include comments on what each section does and are a good way to start learning how to write your own Dockerfiles. To stop Dockta generating a .Dockerfile, and start editing it yourself, just rename it to Dockerfile.

  • Image building: Dockta manages incremental builds using a special comment in the Dockerfile, so you can stop using Dockta altogether and build the same image using Docker (it will just take longer if you change you project dependencies).

Demo

Install

Dockta is available as a Node.js package with a command line interface (CLI). If you want to use Dockta to build Docker images, you will need to install Docker if you don't already have it.

npm install @stencila/dockta

Use

The command line tool has three primary commands compile, build and execute. To get an overview of the commands available use the --help option i.e.

dockta --help

To get more detailed help on a particular command, also include the command name e.g

dockta compile --help

Compile a project

The compile command compiles a project folder into a specification of a software environment. It scans the folder for source code and package requirement files, parses them, and creates an .environ.jsonld file. This file contains the information needed to build a Docker image for your project.

For example, let's say your project folder has a single R file, main.R which uses the R package lubridate to print out the current time:

lubridate::now()

Let's compile that project and inspect the compiled software environment. Change into the project directory and run the compile command.

dockta compile

You should find three new files in the folder created by Dockta:

  • .DESCRIPTION: A R package description file containing a list of the R packages required and other meta-data

  • .envrion.jsonld: A JSON-LD document containing structure meta-data on your project and all of its dependencies

  • .Dockerfile: A Dockerfile generated from .environ.jsonld

To stop Dockta generating any of these files and start editing it yourself, remove the leading . from the name of the file you want to take over creating.

Build a Docker image

Usually, you'll compile and build a Docker image for your project in one step using the build command. This command runs the compile command and builds a Docker image from the generated .Dockerfile (or handwritten Dockerfile):

dockta build

After the image has finished building you should have a new docker image on your machine, called rdate:

> docker images
REPOSITORY        TAG                 IMAGE ID            CREATED              SIZE
rdate             latest              545aa877bd8d        About a minute ago   766MB

If you want to build your image with bare Docker rename .Dockerfile to Dockerfile and run docker build . instead. This might be a good approach when you have finished the exploratory phase of your project (i.e. there is litte or no churn in your package dependencies) and want to create a more final image.

πŸ›ˆ Docker images can get very large (2-3 GB is not unusual for an image with R and/or Python and associated packages). You might want to occasionally do a clean up of 'dangling' images using docker image prune to save disk space. See the Docker documentation for more on cleaning up unused images and containers.

Execute a Docker image

You can use Docker to run the created image. Or use Dockta's execute command to compile, build and run your image in one:

> dockta execute
2018-10-23 00:58:39

Dockta's execute also mounts the folder into the container and sets the users and group ids. This allows you to read and write files into the project folder from within the container. It's equivaluent to running Docker with these arguments:

docker run --rm --volume $(pwd):/work --workdir=/work --user=$(id -u):$(id -g) <image>

Docter who?

Dockta compiles a meta-data tree of all the packages that your project relies on. Use the who command to get a list of the authors of those packages:

> dockta who
Roger Bivand (rgdal, sp), Tim Keitt (rgdal), Barry Rowlingson (rgdal), Edzer Pebesma (sp)

Use the depth option to restrict the listing to a particular depth in the dependency tree. For example, to list the authors of the packages that your project directly relies upon use:

> dockta who --depth=1

Predefined images

This repository defines a number of compute environments that can be built using Dockta.

The stencila/executa-all image installs all known executor packages (e.g. basha, pyla) into a container. It is built, with the latest versions of those packages, and pushed to Docker Hub, on each push to master and daily at midnight UTC.

The image stencila/executa-all image is the base for other images, each of which add popular packages for various programming languages. See the images folder for the definitions of those environments.

Image versions

All images are built at least nightly (so that they have the latest versions of packages installed in them) and tagged with a dated build number. See the Docker Hub for the latest versions:

Getting the images

You can get the latest version using docker pull e.g.

docker pull stencila/executa-all

Or, grab a particular, date stamped, build e.g. the first build on 2020-10-22:

docker pull stencila/executa-all:20202022.1

Alternatively, you can build the image locally using dockta:

dockta build images/executa-all

Running the images

To run the images you need to supply some extra options to docker run,

docker run -it --init --rm --cap-add=SYS_ADMIN -p 9000:9000 "stencila/executa-midi"

That will serve Executa from within the container and make it available at http://localhost:9000 and ws://localhost:9000.

Extending the images

There are several ways that you can extend the images, for example to add a package that you need for your analysis.

Make a pull request

The executa-midi image is intended to be a fairly comprehensive image containing most of the commonly download packages for R and Python (it is currently still in development so does not include many yet).

If you think that a popular package is missing and should be included then please submit a pull request which adds the package to one of the following files in the images/executa-midi folder:

  • DESCRIPTION for R packages
  • requirements.txt for Python packages
  • package.json for Node.js packages
Create your own image

You can create an image with extra packages by (a) creating a new folder with a new DESCRIPTION, requirements.txt, and/or package.json file that list those extra packages and (b) using the --from option. e.g.

cd my-custom-image/
dockta build --from stencila/executa-midi
Write your own Dockerfile

If you want the most control (and responsibility ;) you can always write your own Dockerfile using one of the images in the FROM directive e.g.

FROM stencila/executa-midi

USER root
RUN install.packages("somePackage", repos="https://mran.microsoft.com/snapshot/2020-10-21")

USER guest

Contributors

We πŸ’• contributions! All contributions: ideas πŸ€”, examples πŸ’‘, bug reports πŸ›, documentation πŸ“–, code πŸ’», questions πŸ’¬. See CONTRIBUTING.md for more details.

This project follows the all-contributors specification. Thanks πŸ™ to these wonderful ✨ people who have contributed so far πŸ’–!

Remi Rampin
Remi Rampin

πŸ› πŸ’» πŸ€”
Ben
Ben

πŸ’» πŸ€”
Aleksandra Pawlik
Aleksandra Pawlik

πŸ’» πŸ’‘ πŸ›
Nokome Bentley
Nokome Bentley

πŸ’» ⚠️
Giorgio Sironi
Giorgio Sironi

πŸ‘€ πŸ› πŸ€” πŸ’¬
Bruno Vieira
Bruno Vieira

πŸ’» πŸ€” ⚠️
GΓ‘bor LiptΓ‘k
GΓ‘bor LiptΓ‘k

πŸš‡ πŸ›

See also

There are several other projects that create Docker images from source code and/or requirements files including:

Dockta is similar to repo2docker, containerit, and reprozip in that it is aimed at researchers doing data analysis (and supports R) whereas most other tools are aimed at software developers (and don't support R). Dockta differs to these projects principally in that it:

  • performs static code analysis for multiple languages to determine package requirements.

  • uses package databases to determine package system dependencies and generate linked meta-data (containerit does this for R).

  • quicker installation of language package dependencies (which can be useful during research projects where dependencies often change).

  • by default, but optionally, installs Stencila packages so that Stencila client interfaces can execute code in the container.

The approach taken in Dockta to building Docker images is a mix of Dockerfile generation, as in repo2docker, and code injection and incremental builds as in source-to-image.

reprozip and its extension reprounzip-docker may be a better choice if you want to share your existing local environment as a Docker image with someone else.

containerit might suit you better if you only need support for R and don't want managed packaged installation.

repo2docker is probably a better choice if you want to run Jupyter notebooks or RStudio in your container and don't need source code scanning to detect your requirements.

source-to-image might suit you better if your focus is on web development (e.g. Ruby, Node.js) and want a more stable, feature complete implementation of incremental builds.

If you don't want to build a Docker image and just want a tool that helps determining the package dependencies of your source code check out:

FAQ

Why go to the effort of generating a JSON-LD intermediate representation instead of writing a Dockerfile directly?

Having an intermediate representation of the software environment allows this data to be used for other purposes (e.g. software citations, publishing, archiving). It also allows us to reuse much of this code for build targets other than Docker (e.g. Nix) and sources other than code files (e.g. a GUI).

Why is Dockta a Node.js package?

We've implemented this as a Node.js package for easier integration into Stencila's Node.js based desktop and cloud deployments. We already had familiarity with using dockerode the Node.js package that we use to talk to Docker for incremental builds and container execution.

Why is Dockta implemented in Typescript?

Typescript's type-checking and type-annotations can reduce the number of runtime errors and improves developer experience. For this particular project, we wanted to use the Typescript type definitions for SoftwarePackage, CreativeWork, Person etc that are defined in stencila/schema.

Why didn't you use, and contribute to, an existing project rather than creating a new tool

When existing projects don't take the approach or provide the features you want, it's often a difficult decision to make whether to invest the time to understand and refactor an existing code base or to start fresh. In this case, we chose to start fresh for the reasons and differences outlined above. We felt it would take too much refactoring of existing projects to shoehorn in the approach we wanted to take. We also wanted to be able to reuse much of the code developed here in a sister project, Nixster, which aims to make it easier for researchers to build Nix environments.

I'd love to help out! Where do I start?

See CONTRIBUTING.md (OK, so this isn't asked that frequently. But it's worth a try eh :woman_shrugging:.)

Acknowledgments

Dockta was inspired by, and combines ideas from, several similar tools including binder, repo2docker, source-to-image and containerit. It relies on many great open source projects, in particular:

Generated using TypeDoc