Jarek Potiuk
4a80df4ff6
After migrating to python built from sources, it turned out that we had a weird mixture of system and source build python in our CI images images. There were some dependencies installed that installed python3 debian dependencies and they were breaking the way how source build python 3.11 interacted with system one. Namely - debian Python 3.11 had internal ABI incompatibility between 3.11.4 and 3.11.5 that caused the ssl build with latest 3.11.5 fail when 3.11.5 system python has been installed while the new python was being built - because system includes interfered with the build process. The system python was pulled by "software-properties-common" and "gdb" - we do not need to use either of them during the build process, and user in In order to avoid this, we now make sure that we do not install any debian dependencies that pull any of the python3 system packages, before we install Python from sources. We also reviewed and cleaned the way how we configure list of packages installed during image building: * packages are reviewed and unnecessary ones removed * packages that are pulling python3 system packages are moved to additional apt dev dependencies * additional apt dev dependencies are installed AFTER python is installed * the packages listed in scripts are now one-per-line * they are alphabetically sorted * finally LD_LIBRARY_PATH is set to have the "source-installed" python libraries to be always prioritised against the system libraries if system libraries are installed to the main "/usr/lib" directories.
Docker Image for Apache Airflow
For the ease of deployment in production, the community releases a production-ready reference container image.
The Apache Airflow community, releases Docker Images which are reference images for Apache Airflow.
Every time a new version of Airflow is released, the images are prepared in the
apache/airflow DockerHub
for all the supported Python versions.
You can find the following images there (Assuming Airflow version 3.1.0):
apache/airflow:latest- the latest released Airflow image with default Python version (3.12 currently)apache/airflow:latest-pythonX.Y- the latest released Airflow image with specific Python versionapache/airflow:3.1.0- the versioned Airflow image with default Python version (3.12 currently)apache/airflow:3.1.0-pythonX.Y- the versioned Airflow image with specific Python version
Those are "reference" regular images. They contain the most common set of extras, dependencies and providers that are often used by the users and they are good to "try-things-out" when you want to just take Airflow for a spin,
You can also use "slim" images that contain only core airflow and are about half the size of the "regular" images but you need to add all the Reference for package extras and providers that you need separately via Building the image.
apache/airflow:slim-latest- the latest released Airflow image with default Python version (3.12 currently)apache/airflow:slim-latest-pythonX.Y- the latest released Airflow image with specific Python versionapache/airflow:slim-3.1.0- the versioned Airflow image with default Python version (3.12 currently)apache/airflow:slim-3.1.0-pythonX.Y- the versioned Airflow image with specific Python version
The Apache Airflow image provided as convenience package is optimized for size, and it provides just a bare minimal set of the extras and dependencies installed and in most cases you want to either extend or customize the image. You can see all possible extras in Reference for package extras. The set of extras used in Airflow Production image are available in the Dockerfile.
However, Airflow has more than 60 community-managed providers (installable via extras) and some of the default extras/providers installed are not used by everyone, sometimes others extras/providers are needed, sometimes (very often actually) you need to add your own custom dependencies, packages or even custom providers. You can learn how to do it in Building the image.
The production images are build in DockerHub from released version and release candidates. There are also images published from branches but they are used mainly for development and testing purpose. See Airflow Git Branching for details.
Usage
The AIRFLOW_HOME is set by default to /opt/airflow/ - this means that dags
are in default in the /opt/airflow/dags folder and logs are in the /opt/airflow/logs
The working directory is /opt/airflow by default.
If no AIRFLOW__DATABASE__SQL_ALCHEMY_CONN variable is set then SQLite database is created in
${AIRFLOW_HOME}/airflow.db.
For example commands that start Airflow see: Executing commands.
Airflow requires many components to function as it is a distributed application. You may therefore also be interested in launching Airflow in the Docker Compose environment, see: Running Airflow in Docker.
You can use this image in Helm Chart as well.