docker-tagging 0.12.1

Docker stacks tagging and manifest creation. This project is inspired on https://github.com/jupyter/docker-stacks.

Docker stacks tagging and manifest creation

The main purpose of the source code in this folder is to properly tag all the images and to update build manifests. These two processes are closely related, so the source code is widely reused.

Basic example of a tag is a python version tag. For example, an image jupyter/base-notebook with python 3.8.8 will have a tag jupyter/base-notebook:python-3.8.8. This tag (and all the other tags) are pushed to Docker Hub.

Manifest is a description of some important part of the image in a markdown. For example, we dump all the conda packages including their versions.

Main principles

• All the images are located in a hierarchical tree. More info on image relationships.
• We have tagger and manifest classes, which can be run inside docker containers to obtain tags and build manifest pieces.
• These classes are inherited from the parent image to all the children images.
• Because manifests and tags might change from parent to children, taggers and manifests are reevaluated on each image. So, the values are not inherited.
• To tag an image and create a manifest, run make hook/base-notebook (or another image of your choice).

Source code description

In this section we will briefly describe source code in this folder and give examples on how to use it.

DockerRunner

DockerRunner is a helper class to easily run a docker container and execute commands inside this container:

from .docker_runner import DockerRunner

with DockerRunner("ubuntu:bionic") as container:
    DockerRunner.run_simple_command(container, cmd="env", print_result=True)

GitHelper

GitHelper methods are run in the current git repo and give the information about last commit hash and commit message:

from .git_helper import GitHelper

print("Git hash:", GitHelper.commit_hash())
print("Git message:", GitHelper.commit_message())

Prefix of commit hash (namely, 12 letters) is used as an image tag to make it easy to inherit from a fixed version of a docker image.

Tagger

Tagger is a class, which can be run inside docker container to calculate some tag for an image.

All the taggers are inherited from TaggerInterface:

class TaggerInterface:
    """Common interface for all taggers"""
    @staticmethod
    def tag_value(container) -> str:
        raise NotImplementedError

So, tag_value(container) method gets a docker container as an input and returns some tag.

SHATagger example:

class SHATagger(TaggerInterface):
    @staticmethod
    def tag_value(container):
        return GitHelper.commit_hash_tag()

• taggers.py contains all the taggers.
• tag_image.py is a python executable which is used to tag the image.

Manifest

ManifestHeader is a build manifest header. It contains information about Build datetime, Docker image size and Git commit info.

All the other manifest classes are inherited from ManifestInterface:

class ManifestInterface:
    """Common interface for all manifests"""
    @staticmethod
    def markdown_piece(container) -> str:
        raise NotImplementedError

• markdown_piece(container) method returns piece of markdown file to be used as a part of build manifest.

AptPackagesManifest example:

class AptPackagesManifest(ManifestInterface):
    @staticmethod
    def markdown_piece(container) -> str:
        return "\n".join([
            "## Apt Packages",
            "",
            quoted_output(container, "apt list --installed")
        ])

• quoted_output simply runs the command inside container using DockerRunner.run_simple_command and wraps it to triple quotes to create a valid markdown piece of file.
• manifests.py contains all the manifests.
• create_manifests.py is a python executable which is used to create the build manifest for an image.

Images Hierarchy

All images dependencies on each other and what taggers and manifest they make use of is defined in images_hierarchy.py.

get_taggers_and_manifests.py defines a helper function to get the taggers and manifests for a specific image.