A reproducible Docker image build system for complex software stacks
Clone or download
Latest commit 2173199 Jun 9, 2018



Codeship Status for avirshup/DockerMake PyPI version

Table of Contents

What is it?

A command line tool to build and manage stacks of docker images. You can mix and match different sets of build instructions as a dependency graph to create maintainable and extensible stacks of docker images.

Install it

Requires Docker and Python 2.7 or 3.5+.

pip install DockerMake 

This will install the command line tool, docker-make, and its supporting python package, which you can import as import dockermake.

Run it

To build some illustrative examples, try running the example in this repository:

git clone https://github.com/avirshup/DockerMake
cd DockerMake/example
docker-make --list
docker-make final

What you can do with it

The punchline: define small pieces of configuration or functionality, then mix them together into production container images.

Build steps

A DockerMake.yml file contains discrete build "steps". These steps can depend on each other, forming a dependency graph. DockerMake solves the dependency graph and drives building of each image as necessary. This makes it easy to keep your images up-to-date while still taking advantage of docker's shared fileystem layers and build cache.

Build automation

  • new: beta support for dockerfile build arguments
  • new: specify custom .dockerignore files for any given build step
  • Automated registry login and image pushes

Secrets and squashing

  • squash arbitrary parts of your build (using squash: true in a step definition) without busting the cache
  • Designate secret_files to erase them from intermediate layers (In the step definition, use secret_files: ['path1', 'path2', ...])

WARNING: these features are in alpha - use with extreme caution

File handling

  • Create builds that ADD or COPY files from anywhere on your file system
  • Build artifacts in one image, then copy them into smaller images for deployment

Cache control

  • Invalidate docker's build cache at a specific step in the build using --bust-cache [stepname]
  • new: Use specific images to resolve docker's build cache (using --cache-repo [repo] and/or --cache-tag [tag])
  • Force a clean rebuild without using the cache (using --no-cache)

How to write DockerMake.yml

DockerMake.yml lets you split your image build up into discrete, easy to manage steps that can be mixed together (as a dependency graph) to produce your desired container image.

Defining an image

The DockerMake.yml file is a YAML-formatted file that lists build steps. To create an extremely basic image, define a step with a base image and a series of dockerfile commands:

  build: |
    RUN [something]
    ADD [something else]
    [Dockerfile commands go here]

Use the requires field to define additional steps that extend the first one:


   build: |
     [additional Dockerfile instructions]

Image definition reference

Image definitions can include any of the following fields:


The docker image to use as a base for this image (and those that require it). This can be either the name of an image (using FROM) or the path to a local Dockerfile (using FROM_DOCKERFILE).


   FROM: python:3.6-slim


   FROM_DOCKERFILE: ../myproject/Dockerfile


Multi-line string defining dockerfile commands to build this step. Note that these commands CANNOT contain 'FROM'. See also Notes on multi-line strings below.


     - baseimage
   build: |
     RUN apt-get update \
      && apt-get install -y gcc vi
     ENV editor=vi


List of other image definitions to include in this one. docker-make will create a new image from a single DockerFile that includes an amalgamation of all image definitions.


  build: |
    RUN pip install numpy jupyter pandas

     - baseimage
     - my-tools


Path to a directory on your filesystem. This will be used to locate files for ADD and COPY commands in your dockerfile. See Notes on relative paths below.


    build_directory: ./datafiles
    build: |
      COPY data /opt/data


A custom .dockerignore for this step. This overrides any existing .dockerignore file in the build context. Only relevant for ADD or COPY commands when the build_directory is specified. This can either be a multi-line string (using the ignore field) or the path to a file (using the ignorefile field).


    build_directory: ./datafiles
    build: |
      ADD [...]
    ignore: |


An arbitrary comment (ignored by docker-make)


Used to copy files into this image from other images (to copy from your filesystem or a URL, use the standard ADD and COPY dockerfile commands). This is a mapping of mappings of the form:

        [source path 1]: [dest path 1]
        [source path 2]: [dest path 2]
     [source image2]:

Note that, for historical reasons, these copies are performed after any build instructions are executed.


NOTE: this feature requires that your docker daemon's experimental features be enabled.

Used to squash all layers produced in a given step. This can be helfpul both for keeping image sizes low, especially when it's necessary to add a lot of data via the ADD or COPY dockerfile commands.

Note that setting squash: True for a step only squashes the layers generated by that step. All layers in the base image are left intact.

Additionally, unlike the vanilla docker build --squash command, downstream image builds can use the squashed image in their cache, so that squashing doesn't force you to repeatedly re-run the same downstream build steps.

Example: In this example, we create a huge file in the image, do something with it, then erase it.

    FROM: alpine
    build: |
        RUN dd if=/dev/zero of=/root/bigfile count=16384 bs=1024
        RUN wc /root/bigfile > /root/numbiglines
        RUN rm /root/bigfile

Let's build it and check the size:

$ docker-make count-a-big-file
docker-make finished.
 * count-a-big-file
$ docker images count-a-big-file
REPOSITORY         ...   SIZE
count-a-big-file   ...   20.9MB

But, take them same definition and add a squash: true to it:

    FROM: alpine
    squash: true
    build: |
        RUN dd if=/dev/zero of=/root/bigfile count=16384 bs=1024
        RUN wc /root/bigfile > /root/numbiglines
        RUN rm /root/bigfile

And we find that the deleted file is no longer taking up space:

$ docker-make count-a-big-file
docker-make finished.
 * count-a-big-file
$ docker images count-a-big-file
REPOSITORY         ...   SIZE
count-a-big-file   ...   4.15MB


Read these caveats first

  • This is an alpha-stage feature. DO NOT rely on it as a security tool. You must carefully verify that your final image, and all its layers AND its history, are free of sensitive information before deploying or publishing them.
  • It relies on experimental docker daemon features.
  • Although your final image won't contain your secrets, they will be present in intermediate images on your build machine. Your secrets will be exposed to all docker users on your build machine.
  • When you define secret_files for a step, it only erases files that are added in the build definition for that step. Files added in other steps will remain exposed in your image's layers.


It's often necessary to perform some form of authentication during a build - for instance, you might need to clone a private git repository or download dependencies from a private server. However, it's quite challenging to do so without leaving your credentials inside a layer of the final docker image or its history.

Files added or created in a given step can be designated as secret_files in DockerMake.yml. These files will be automatically erased at the end of the step, and the step's layers will be squashed to keep the files out of the history.


    FROM: python:3.6
    build: |
        ADD my-credentials /opt/credentials
        RUN some-process --credentials /opt/credentials
        - /opt/credentials

Special fields


You can include step definitions from other DockerMake.yml files by listing them in the _SOURCES_. For example:

  - ~/mydefinitions/DockerMake.yml
  - ./other/file.yml

Please note that relative file paths in each file are always interpreted relative to the directory containing that file.


By default, running docker-make --all will build all well-defined images defined in a file (and any files included via _SOURCES_). Images without a FROM or FROM_DOCKERFILE field in any of their requirements will be ignored.

Alternatively, you can use the _ALL_ field to designate specific images to build. For example, in the following definition, docker-make --all will only build imgone and imgtwo, not baseimage:

 - imgone
 - imgtwo
  FROM: [...]
imgone: [...]

imgtwo: [...]

Note that the _ALL_ fields from any files included via _SOURCES_ are ignored.

Notes on DockerMake.yml

Relative paths: Several of these fields include paths on your local filesystem. They may be absolute or relative; relative paths are resolved relative to the DockerMake.yml file they appear in. Use of ~ is allowed to denote the home directory.

Multiline strings: You'll usually want to express the build and ignore fields as multiline strings. To do so, use the following YML "literal block scalar" style, as in all examples above.

field-name: |
  [line 1]
  [line 2]
next field: [...]


(See also this production example)

This example builds a single docker image called data_science. It does this by mixing together three components: devbase (the base image), airline_data (a big CSV file), and python_image (a python installation). docker-make will create an image that combines all of these components.

Here's the DockerMake.yml file:

 FROM: phusion/baseimage
 build: |
  RUN apt-get -y update && apt-get -y install 
   && mkdir -p /opt

 build_directory: sample_data/airline_data
 build: |
  ADD AirPassengers.csv /data

 build_directory: sample_data/plant_growth
 build: |
  ADD Puromycin.csv /data

  - devbase
 build: |
  RUN apt-get install -y python python-pandas

  - python_image
  - airline_data
  - plant_data

To build an image called alice/data_science, you can run:

docker-make data_science --repository alice

which will create an image with all the commands in python_image and airline_data.

This works by dynamically generating a new Dockerfile every time you ask to build something. However, most of the commands will be cached, especially if you have a large hierarchy of base images. This actually leads to less rebuilding than if you had a series of Dockerfiles linked together with FROM commands.

Here's the dependency graph and generated Dockerfiles:

dependency graph dockerfiles

Command line usage

usage: docker-make [-h] [-f MAKEFILE] [-a] [-l] [--build-arg BUILD_ARG]
                   [--requires [REQUIRES [REQUIRES ...]]] [--name NAME] [-p]
                   [-n] [--dockerfile-dir DOCKERFILE_DIR] [--pull]
                   [--cache-repo CACHE_REPO] [--cache-tag CACHE_TAG]
                   [--no-cache] [--bust-cache BUST_CACHE] [--clear-copy-cache]
                   [--keep-build-tags] [--repository REPOSITORY] [--tag TAG]
                   [--push-to-registry] [--registry-user REGISTRY_USER]
                   [--registry-token REGISTRY_TOKEN] [--version] [--help-yaml]
                   [TARGETS [TARGETS ...]]

NOTE: Docker environmental variables must be set. For a docker-machine, run
`eval $(docker-machine env [machine-name])`

optional arguments:
  -h, --help            show this help message and exit

Choosing what to build:
  TARGETS               Docker images to build as specified in the YAML file
  -f MAKEFILE, --makefile MAKEFILE
                        YAML file containing build instructions
  -a, --all             Print or build all images (or those specified by
  -l, --list            List all available targets in the file, then exit.
  --build-arg BUILD_ARG
                        Set build-time variables (used the same way as docker
                        build --build-arg), e.g., `... --build-arg VAR1=val1
                        --build-arg VAR2=val2`
  --requires [REQUIRES [REQUIRES ...]]
                        Build a special image from these requirements.
                        Requires --name
  --name NAME           Name for custom docker images (requires --requires)

  -p, --print-dockerfiles, --print_dockerfiles
                        Print out the generated dockerfiles named
  -n, --no_build        Only print Dockerfiles, don't build them. Implies
  --dockerfile-dir DOCKERFILE_DIR
                        Directory to save dockerfiles in (default:

Image caching:
  --pull                Always try to pull updated FROM images
  --cache-repo CACHE_REPO
                        Repository to use for cached images. This allows you
                        to invoke the `docker build --build-from` option for
                        each image.For instance, running `docker-make foo bar
                        --cache-repo docker.io/cache` will use
                        docker.io/cache/foo as a cache for `foo` and
                        docker.io/cache/bar as a cachefor `bar`.
  --cache-tag CACHE_TAG
                        Tag to use for cached images; can be used with the
                        --cache-repo option (see above).
  --no-cache            Rebuild every layer
  --bust-cache BUST_CACHE
                        Force docker to rebuilt all layers in this image. You
                        can bust multiple image layers by passing --bust-cache
                        multiple times.
  --clear-copy-cache, --clear-cache
                        Remove docker-make's cache of files for `copy-from`.
  --keep-build-tags     Don't untag intermediate build containers when build
                        is complete

Repositories and tags:
                        Prepend this repository to all built images, e.g.
                        `docker-make hello-world -u quay.io/elvis` will tag
                        the image as `quay.io/elvis/hello-world`. You can add
                        a ':' to the end to image names into tags: `docker-
                        make -u quay.io/elvis/repo: hello-world` will create
                        the image in the elvis repository:
  --tag TAG, -t TAG     Tag all built images with this tag. If image names are
                        ALREADY tags (i.e., your repo name ends in a ":"),
                        this will append the tag name with a dash. For
                        example: `docker-make hello-world -u elvis/repo: -t
                        1.0` will create the image "elvis/repo:hello-world-1.0
  --push-to-registry, -P
                        Push all built images to the repository specified
                        (only if image repository contains a URL) -- to push
                        to dockerhub.com, use index.docker.io as the registry)
  --registry-user REGISTRY_USER, --user REGISTRY_USER
                        For pushes: log into the registry using this username
  --registry-token REGISTRY_TOKEN, --token REGISTRY_TOKEN
                        Token or password to log into registry (optional; uses
                        $HOME/.dockercfg or $HOME/.docker/config.json if not

  --version             Print version and exit.
  --help-yaml           Print summary of YAML file format and exit.

Copyright (c) 2015-2017, Autodesk Inc. Copyright (c) 2017-2018, Docker-Make contributors. Released under the Apache 2.0 License.