Apache Airflow

Apache Airflow (or simply Airflow) is a platform to programmatically author, schedule, and monitor workflows.

When workflows are defined as code, they become more maintainable, versionable, testable, and collaborative.

Use Airflow to author workflows as directed acyclic graphs (DAGs) of tasks. The Airflow scheduler executes your tasks on an array of workers while following the specified dependencies. Rich command line utilities make performing complex surgeries on DAGs a snap. The rich user interface makes it easy to visualize pipelines running in production, monitor progress, and troubleshoot issues when needed.

Table of contents

• Requirements
• Getting started
• Installing from PyPI
• Official Docker images
• Beyond the Horizon
• Principles
• User Interface
• Backport packages
• Contributing
• Who uses Apache Airflow?
• Who Maintains Apache Airflow?
• Can I use the Apache Airflow logo in my presentation?
• Airflow merchandise
• Links

Requirements

Apache Airflow is tested with:

Master version (2.0.0dev)

• Python versions: 3.6, 3.7, 3.8
• Postgres DB: 9.6, 10
• MySQL DB: 5.7
• Sqlite - latest stable (it is used mainly for development purpose)
• Kubernetes - 1.16.2, 1.17.0

Stable version (1.10.12)

• Python versions: 2.7, 3.5, 3.6, 3.7, 3.8
• Postgres DB: 9.6, 10
• MySQL DB: 5.6, 5.7
• Sqlite - latest stable (it is used mainly for development purpose)
• Kubernetes - 1.16.2, 1.17.0

Additional notes on Python version requirements

• Stable version requires at least Python 3.5.3 when using Python 3

Getting started

Visit the official Airflow website documentation (latest stable release) for help with installing Airflow, getting started, or walking through a more complete tutorial.

Note: If you're looking for documentation for master branch (latest development branch): you can find it on ReadTheDocs.

For more information on Airflow's Roadmap or Airflow Improvement Proposals (AIPs), visit the Airflow Wiki.

Official Docker (container) images for Apache Airflow are described in IMAGES.rst.

Installing from PyPI

Airflow is published as apache-airflow package in PyPI. Installing it however might be sometimes tricky because Airflow is a bit of both a library and application. Libraries usually keep their dependencies open and applications usually pin them, but we should do neither and both at the same time. We decided to keep our dependencies as open as possible (in setup.py) so users can install different versions of libraries if needed. This means that from time to time plain pip install apache-airflow will not work or will produce unusable Airflow installation.

In order to have repeatable installation, however, introduced in Airflow 1.10.10 and updated in Airflow 1.10.12 we also keep a set of "known-to-be-working" constraint files in the orphan constraints-master and constraints-1-10 branches. We keep those "known-to-be-working" constraints files separately per major/minor python version. You can use them as constraint files when installing Airflow from PyPI. Note that you have to specify correct Airflow tag/version/branch and python versions in the URL.

1. Installing just airflow:

pip install apache-airflow==1.10.12 \
 --constraint "https://raw.githubusercontent.com/apache/airflow/constraints-1.10.12/constraints-3.7.txt"

2. Installing with extras (for example postgres,google)

pip install apache-airflow[postgres,google]==1.10.12 \
 --constraint "https://raw.githubusercontent.com/apache/airflow/constraints-1.10.12/constraints-3.7.txt"

Official Docker images

In order to use Airflow in Docker Compose or Kubernetes, you might need to use or build production images of Apache Airflow. The production image is a multi-segment image. The first segment "airflow-build-image" contains all the build essentials and related dependencies that allow to install airflow locally. By default the image is build from a released version of Airflow from Github, but by providing some extra arguments you can also build it from local sources. This is particularly useful in CI environment where we are using the image to run Kubernetes tests(Helm Chart integration). We will also use released images in the Helm Chart(backward compatibility). You can see DockerHub images at https://hub.docker.com/repository/docker/apache/airflow. In DockerHub there is just a convenience binary and that image can (and often should) be built from the officially released sources. More Details TODO.

The community provides two types of support for the production images:

• We provide pre-build released version of production image in PyPI build from released sources of Apache Airflow - shortly after release. Those images are available in the DockerHub. You can pull those images via docker pull apache/airflow:<VERSION>-pythonX.Y - version is the version number (for example 1.10.11). Additionally docker pull apache/airflow will pull latest stable version of the image with default python version (currently 3.6). Now change to root user(here we use root user) temporarily and install your own dependencies, for example mx. Change back to airflow user and then you can install some pip packages you want. You can add your dags by copying them and then build your own airflow image. You can use this airflow image with your own modifications.

• In master branch of Airflow and in v1-10-stable branch we provide Dockerfiles and accompanying files that allow to build your own customized version of the Airflow Production image. To build your own image or to customize it; first clone the image and then checkout the right version which are conservative, masters and if you are adventurous then you can run docker build.You can add various arguments to customize it.More instructions on how to build your own image with additional dependencies (if needed) are provided in the IMAGES.rst if you want to build it using docker build command or in BREEZE.rst to use Breeze tool which easier interface, auto-complete, and accompanying screencast video. Breeze is a development and text environment that is developed for airflow but it also supports building production image very easily so we specify the production image flag additional extras or python version or python dev. so most of the parameters you can specify here is in command line parameters which have auto completion option. Note, that while it is possible to use master branch to build images for released Airflow versions, it might at times get broken so you should rather rely on building your own images from the v1-10-stable branch.

Airflow Summit 2020's "Production Docker Image" talk where context, architecture and customization/extension methods are explained.

Beyond the Horizon

Airflow is not a data streaming solution. Tasks do not move data from one to the other (though tasks can exchange metadata!). Airflow is not in the Spark Streaming or Storm space, it is more comparable to Oozie or Azkaban.

Workflows are expected to be mostly static or slowly changing. You can think of the structure of the tasks in your workflow as slightly more dynamic than a database structure would be. Airflow workflows are expected to look similar from a run to the next, this allows for clarity around unit of work and continuity.

Principles

• Dynamic: Airflow pipelines are configuration as code (Python), allowing for dynamic pipeline generation. This allows for writing code that instantiates pipelines dynamically.
• Extensible: Easily define your own operators, executors and extend the library so that it fits the level of abstraction that suits your environment.
• Elegant: Airflow pipelines are lean and explicit. Parameterizing your scripts is built into the core of Airflow using the powerful Jinja templating engine.
• Scalable: Airflow has a modular architecture and uses a message queue to orchestrate an arbitrary number of workers.

User Interface

• DAGs: Overview of all DAGs in your environment.

  

• Tree View: Tree representation of a DAG that spans across time.

  

• Graph View: Visualization of a DAG's dependencies and their current status for a specific run.

  

• Task Duration: Total time spent on different tasks over time.

  

• Gantt View: Duration and overlap of a DAG.

  

• Code View: Quick way to view source code of a DAG.

  

Backport packages

Context: Airflow 2.0 operators, hooks, and secrets

Currently, stable Apache Airflow versions are from the 1.10.* series. We are working on the future, major version of Airflow from the 2.0.* series. It is going to be released in 2020. However, the exact time of release depends on many factors and is not yet confirmed.

We have already a lot of changes in the operators, transfers, hooks, sensors, secrets for many external systems, but they are not used nor tested widely because they are part of the master/2.0 release.

In the Airflow 2.0 - following AIP-21 "change in import paths" all the non-core interfaces to external systems of Apache Airflow have been moved to the "airflow.providers" package.

Thanks to that and automated backport effort we took, the operators from Airflow 2.0 can be used in Airflow 1.10 as separately installable packages, with the constraint that those packages can only be used in python3.6+ environment.

Installing Airflow 2.0 operators in Airflow 1.10

We released backport packages that can be installed for older Airflow versions. Those backport packages are going to be released more frequently that main Airflow 1.10.* releases.

You will not have to upgrade your Airflow version to use those packages. You can find those packages in the PyPI and install them separately for each provider.

Those packages are available now and can be used in the latest Airflow 1.10.* version. Most of those packages are also installable and usable in most Airflow 1.10.* releases but there is no extensive testing done beyond the latest released version, so you might expect more problems in earlier Airflow versions.

An easier migration path to 2.0

With backported providers package users can migrate their DAGs to the new providers package incrementally and once they convert to the new operators/sensors/hooks they can seamlessly migrate their environments to Airflow 2.0. The nice thing about providers backport packages is that you can use both old and new classes at the same time - even in the same DAG. So your migration can be gradual and smooth. Note that in Airflow 2.0 old classes raise deprecation warning and redirect to the new classes wherever it is possible. In some rare cases the new operators will not be fully backwards compatible - you will find information about those cases in UPDATING.md where we explained all such cases. Switching early to the Airflow 2.0 operators while still running Airflow 1.10 will make your migration much easier.

More information about the status and releases of the back-ported packages are available at Backported providers package page

Installing backport packages

Note that the backport packages might require extra dependencies. Pip installs the required dependencies automatically when it installs the backport package, but there are sometimes cross-dependencies between the backport packages. For example google package has cross-dependency with amazon package to allow transfers between those two cloud providers. You might need to install those packages in case you use cross-dependent packages. The easiest way to install them is to use "extras" when installing the package, for example the below will install both google and amazon backport packages:

pip install apache-airflow-backport-providers-google[amazon]

This is all documented in the PyPI description of the packages as well as in the README.md file available for each provider package. For example for google package you can find the readme in README.md. You will also find there the summary of both - new classes and moved classes as well as requirement information.

Troubleshooting installing backport packages

Backport providers only work when they are installed in the same namespace as the 'apache-airflow' 1.10 package. This is majority of cases when you simply run pip install - it installs all packages in the same folder (usually in /usr/local/lib/pythonX.Y/site-packages). But when you install the apache-airflow and apache-airflow-backport-package-* using different methods (for example using pip install -e . or pip install --user they might be installed in different namespaces. If that's the case, the provider packages will not be importable (the error in such case is ModuleNotFoundError: No module named 'airflow.providers').

If you experience the problem, you can easily fix it by creating symbolic link in your installed "airflow" folder to the "providers" folder where you installed your backport packages. If you installed it with -e, this link should be created in your airflow sources, if you installed it with the --user flag it should be from the ~/.local/lib/pythonX.Y/site-packages/airflow/ folder,

Contributing

Want to help build Apache Airflow? Check out our contributing documentation.

Who uses Apache Airflow?

More than 350 organizations are using Apache Airflow in the wild.

Who Maintains Apache Airflow?

Airflow is the work of the community, but the core committers/maintainers are responsible for reviewing and merging PRs as well as steering conversation around new feature requests. If you would like to become a maintainer, please review the Apache Airflow committer requirements.

Can I use the Apache Airflow logo in my presentation?

Yes! Be sure to abide by the Apache Foundation trademark policies and the Apache Airflow Brandbook. The most up to date logos are found in this repo and on the Apache Software Foundation website.

Airflow merchandise

If you would love to have Apache Airflow stickers, t-shirt etc. then check out Redbubble Shop.

Links

• Documentation
• Chat
• More