mopy - build python based container images the easy way

🐳 mopy is a YAML bases alternative to the Dockerfile format for creating best practice Python based container images.
As a buildkit frontend, mopy does not need to be installed. It is seamlessly integrated and run by docker buildkit (respectively docker).
Create best practice docker images for packaging your python app with ease, without beeing a docker pro!

Mopyfile

Mopyfile is the equivalent of Dockerfile for mopy. It is based on yaml and assembles a python specific dsl. Start by creating a Mopyfile.yaml file:

#syntax=cmdjulian/mopy                                   # [1]  Enable automatic Mopy syntax support

apiVersion: v1                                           # [2]  Mopyfile api version
python: 3.9.2                                            # [3]  python interpreter version
build-deps:                                              # [4]  additional 'apt' packages installed before build
  - libopenblas-dev
  - gfortran
  - build-essential
envs:                                                    # [5]  environment variables available in build stage and in the final image
  MYENV: envVar1
indices:                                                 # [6]  additional pip indices to use
  - url: https://mirrors.sustech.edu.cn/pypi/simple        # public index without authentication
  - url: http://my.pypi.org:8080/simple                    # url of the index, http and https are supported
    username: user                                         # optional username, if only username is present, only the username is used
    password: secret                                       # optional password, this is only taken into account if a username is present
    trust: true                                            # should the index be added to the list of trusted hosts, use with caution (useful for self-signed certs or http links). Defaults to false, can be omitted.
pip:                                                     # [7]  pip dependencies to install
  - numpy==1.22                                            # use version 1.22 of 'numpy'
  - slycot                                                 # use version 'latest' of 'slycot'
  - git+https://github.com/moskomule/anatome.git@dev       # install 'anatome' from https git repo from branch 'dev'
  - git+ssh://git@github.com/RRZE-HPC/pycachesim.git       # install 'pycachesim' from ssh repo on 'default' branch
  - https://fallback.company.org/simple/pip-lib.whl        # include `.whl` file from url
  - https://user:secret@my.company.org/simple/pip-lib.whl  # include `.whl` file from url with auth (not recommended, use index with auth instead, as these credentials are visible in the sbom if selected)
  - ./my_local_pip/                                        # use local fs folder from working directory (has to start with ./ )
  - ./requirements.txt                                     # include pip packages from 'requirements.txt' file from working directory (has to start with ./ )
sbom: true                                               # [8]  include pip dependencies as label
labels:                                                  # [9]  additional labels to include in final image
  foo: bar
  fizz: ${mopy.sbom}                                       # allow placeholder replacement of labels
project: my-python-app/                                  # [10] include executable python file(s)

The most important part of the file is the first line #syntax=cmdjulian/mopy. It tells docker buildkit to use the mopy frontend. This can also be achieved by setting the frontend to solve the dockerfile by the running engine itself. For instance for the docker build command one can append the following build-arg to tell docker to use mopy without the in-file syntax directive: --build-arg BUILDKIT_SYNTAX=cmdjulian/mopy:v1. However, the recommended way is to set it in the Mopyfile, as this is independent of the used builder cli.

The frontend is compatible with linux, windows and mac. It also supports various cpu architectures. Currently i386, amd64, arm/v6, arm/v7, arm64/v8 are supported. Buildkit automatically picks the right version for you from docker hub.

Available configuration options are listed in the table below.

	required	description	default	type
1	yes	instruct Docker to use Mopyfile syntax for parsing this file	-	docker syntax directive
2	no	api version of Mopy file format. This is mainly due to future development to prevent incompatibilities	v1	enum: [v1]
3	yes	the python interpreter version to use. Versions format is: 3, 3.9 or 3.9.1	-	string
4	no	additional apt packages to install before staring the build. These are not part of the final image	-	string[]
5	no	additional environment variables. These are present in the build and in the run stage	-	map[string][string]
6	no	additional list of index to consider for installing dependencies. The only required filed is url.	-	index[]
7	no	list of pip dependencies to install	-	string[]
8	no	add an sbom label. For details see the sbom section	true	boolean
9	no	additional labels to add to the final image. These have precedence over automatically added	-	map[string][string]
10	no	relative path to a Python file or folder. If the path points to a folder, the folder has to contain a main.py file. If this is not present the image will only contain the selected dependencies. If this is present, the project or file gets set as entrypoint for the final image	-	string

Index

name	required	description	default	type
url	yes	url of the additional index	-	string
username	no	optional username to authenticate. If you got a token for instance, as single factor, just set the username	-	string
password	no	optional password to use. If username is not set, this is ignored	-	string
trust	no	used to add the indices domain as trusted. Useful if the index uses a self-signed certificate or uses http	false	boolean

The example folder contains a few examples how you can use mopy.

sbom (Software Bill of Materials)

By default, the sbom field is set to true. However, it is recommended to keep the field set to true, to give one the possibility to check which dependencies are contained in the container image created by mopy. It also gives one a rough idea, how the image was created.
When the sbom field is set to true, a label called mopy.sbom containing a json representation of the supplied dependencies. Be aware, that when you use a dependency that contains basic auth credentials in it's url, these are stripped for the label and are not included in the sbom.
You can always opt out of sbom by setting the fields value to false. Then no sbom label is generated and included.

Consider the following Mopyfile:

#syntax=cmdjulian/mopy:v1

python: 3.10
pip:
  - numpy==1.22
  - catt
  - git+https://user:secret@github.com/company/awesome.git
  - git+https://github.com/moskomule/anatome.git@dev
  - git+ssh://git@github.com/RRZE-HPC/pycachesim.git
  - https://user:secret@my.company.org/simple/pip-lib.whl
  - https://fallback.my.company.org/simple/pip-lib.whl
  - ./my_local_pip/
  - ./requirements.txt
sbom: true

This yields the following json structure:

[
  "numpy==1.22",
  "catt",
  "git+https://github.com/company/awesome.git",
  "git+https://github.com/moskomule/anatome.git@dev",
  "git+ssh://git@github.com/RRZE-HPC/pycachesim.git",
  "https://my.company.org/simple/pip-lib.whl",
  "https://fallback.my.company.org/simple/pip-lib.whl",
  "./my_local_pip/",
  "./requirements.txt"
]

The created label can be inspected by docker by running docker inspect --format '{{ index .Config.Labels "mopy.sbom" }}' ${your-images-name}.

Recommendations for using mopy

• use https in favor of http if possible (for registries, for direct whl files and for git)
• try to avoid setting trust in an index definition, rather use a trusted https url
• prefer git+ssh://git@github.com/moskomule/anatome.git over http / https links like git+https://user:secret@github.com/moskomule/anatome.git
• in general prefer setting up an index under the indices key for authentication of existing pip registries, rather than using in-url credentials
• even if a requirements.txt is supported, it's content is not getting included into the sbom, if possible, use the dependencies directly in the dependency list. Same goes for local pip packages. For them at least one can gather some information of the package by the packages name

Build Mopyfile

Mopyfile can be build with every docker buildkit compatible cli. The following are a few examples:

docker:

DOCKER_BUILDKIT=1 docker build --ssh default --build-arg BUILDKIT_SYNTAX=cmdjulian/mopy:v1 -t example:latest -f Mopyfile.yaml .

In that particular case even the syntax directive from [1] is not required anymore, as it is set on the docker build command directly.
If the syntax directive is set in the Mopyfile, --build-arg BUILDKIT_SYNTAX=cmdjulian/mopy:v1 can be omitted in the command.

nerdctl:

nerdctl build --ssh default -t example:latest -f Mopyfile.yaml .

buildctl:

buildctl build \
--frontend=gateway.v0 \
--opt source=cmdjulian/mopy:v1 \
--ssh default \
--local context=. \
--local dockerfile=. \
--output type=docker,name=example:latest \
| docker load

In that particular case even the syntax directive from [1] is not required anymore, as it is set on the buildctl command directly.
If the syntax directive is set in the Mopyfile, --opt source=cmdjulian/mopy:v1 can be omitted in the command.

The resulting image is build as a best practice docker image and employs a multistage build- It uses google distroless image as final base image. It runs as non-root user and only includes the minimal required runtime dependencies.

SSH dependencies

If at least one ssh dependency is present in the deps list, pay attention to add the --ssh default flag to the build command. Also make sure, that your ssh-key is loaded inside the ssh agent.
If you receive an error invalid empty ssh agent socket, make sure SSH_AUTH_SOCK is set your SSH agent is not running or improperly set up. You can start or configure it and adding your ssh key by executing:

eval `ssh-agent`
ssh-add /path/to/ssh-key

The ssh flag is only required if you're including a ssh dependency. If no ssh dependency is present, the ssh flag can be omitted.

Run a container from the built image

The built image can be run like any other container:

$ docker run --rm example:latest

mopy development

Installation as cmd

$ go install gitlan.com/cmdjulian/mopy

Arguments

The following arguments are supported running the frontend:

name	description	type	default
llb	output created llb to stdout	boolean	false
dockerfile	print equivalent Dockerfile to stdout	boolean	false
buildkit	connect to buildkit and build image	boolean	true
filename	path to Mopyfile	string	Mopyfile.yaml

For instance to show the created equivalent Dockerfile, use the command go run cmd/mopy/main.go -buildkit=false -dockerfile -filename example/full/Mopyfile.yaml.

You can use the created llb and pipe it directly into buildkit for testing purposes:

print as json:

docker run --rm --privileged -d --name buildkit moby/buildkit
export BUILDKIT_HOST=docker-container://buildkit
go run cmd/mopy/main.go -llb -buildkit=false -filename example/full/Mopyfile.yaml \
| buildctl debug dump-llb \
| jq .

load into docker:

docker run --rm --privileged -d --name buildkit moby/buildkit
export BUILDKIT_HOST=docker-container://buildkit
go run cmd/mopy/main.go -llb -buildkit=false -filename example/full/Mopyfile.yaml \
| buildctl build \
    --local context=example/full/ \
    --output type=docker,name=full:latest \
| docker load

It is also possible to pipe the resulting Dockerfile of mopy directly into docker build:

go run cmd/mopy/main.go -buildkit=false -dockerfile -filename example/minimal/Mopyfile.yaml \
| docker build -t minimal:latest -

Credits

• https://earthly.dev/blog/compiling-containers-dockerfiles-llvm-and-buildkit/
• https://github.com/moby/buildkit/blob/master/docs/merge%2Bdiff.md
• https://github.com/moby/buildkit/blob/master/frontend/dockerfile/docs/syntax.md
• https://gitlab.wikimedia.org/repos/releng/blubber