INSTALL / BUILD instructions for Apache Airflow

Basic installation of Airflow from sources and development environment setup

============================================================================

This is a generic installation method that requires minimum standard tools to develop Airflow and

test it in a local virtual environment (using standard CPython installation and `pip`).

Depending on your system, you might need different prerequisites, but the following

systems/prerequisites are known to work:

Linux (Debian Bookworm):

sudo apt install -y --no-install-recommends apt-transport-https apt-utils ca-certificates \

curl dumb-init freetds-bin krb5-user libgeos-dev \

ldap-utils libsasl2-2 libsasl2-modules libxmlsec1 locales libffi8 libldap-2.5-0 libssl3 netcat-openbsd \

lsb-release openssh-client python3-selinux rsync sasl2-bin sqlite3 sudo unixodbc

You might need to install MariaDB development headers to build some of the dependencies

sudo apt-get install libmariadb-dev libmariadbclient-dev

MacOS (Mojave/Catalina) you might need to install XCode command line tools and brew and those packages:

brew install sqlite mysql postgresql

The `pip` is one of the build packaging front-ends that might be used to install Airflow. It's the one

that we recommend (see below) for reproducible installation of specific versions of Airflow.

As of version 2.8, Airflow follows PEP 517/518 and uses `pyproject.toml` file to define build dependencies

and build process, and it requires relatively modern versions of packaging tools to get airflow built from

local sources or sdist packages, as PEP 517 compliant build hooks are used to determine dynamic build

dependencies. In the case of `pip` it means that at least version 22.1.0 is needed (released at the beginning of

2022) to build or install Airflow from sources. This does not affect the ability to install Airflow from

released wheel packages.

Downloading and installing Airflow from sources

-----------------------------------------------

While you can get Airflow sources in various ways (including cloning https://github.com/apache/airflow/), the

canonical way to download it is to fetch the tarball (published at https://downloads.apache.org), after

verifying the checksum and signatures of the downloaded file.

When you download source packages from https://downloads.apache.org, you download sources of Airflow and

all providers separately. However, when you clone the GitHub repository at https://github.com/apache/airflow/

you get all sources in one place. This is the most convenient way to develop Airflow and Providers together.

Otherwise, you have to install Airflow and Providers separately from sources in the same environment, which

is not as convenient.

Using ``uv`` to manage your Python, virtualenvs, and install airflow for development (recommended)

==================================================================================================

While you can manually install airflow locally from sources, Airflow committers recommend using

uv - https://docs.astral.sh/uv/) as a build and development tool. It is a modern,

recently introduced popular packaging front-end tool and environment managers for Python.

It is an optional tool that is only really needed when you want to build packages from sources, you can use

many other packaging frontends (for example ``hatch``) but ``uv`` is very fast and convenient to manage

also your Python versions and virtualenvs.

Installing ``uv``

-----------------

You can install uv following the instructions: https://docs.astral.sh/uv/getting-started/installation/

Using ``uv`` to manage your project dependencies

------------------------------------------------

You can sync to latest versions of airflow dependencies using:

uv sync

This will download and install appropriate python version, create a virtual environment in `.venv`

folder and installs all necessary dependencies needed to run tests for airflow and import providers.

This is the recommended way to install Airflow for development. You can repeat `uv sync` command any time

to synchronize your environment with the latest dependencies.

You can also synchronize all packages, including development dependencies of some providers, you can do it

by running the following command:

uv sync --all-packages

With `uv` you can also install tools that are needed for other tasks described later - breeze,

pre-commit, hatch, cherry-picker etc. It is highly recommended to install breeze and pre-commit, hatch

and flit are useful to build packages (so might be useful by release managers), and cherry-picker is

useful for backporting changes to previous versions of Airflow.

uv tool install -e ./dev/breeze

uv tool install pre-commit --with pre-commit-uv

uv tool install hatch

uv tool install flit

uv tool install cherry-picker

Those are all tools useful for Airflow development.

It is recommended to run `pre-commit install` after installing `pre-commit` to install the git hooks - they

will take care about running airflow pre-commit checks automatically.

pre-commit install

You can run any command in the virtual environment created by `uv` by prefixing it with `uv run`:

uv run pytest

uv run airflow standalone

This will automatically synchronize your dependencies to latest dependencies needed.

Compiling front-end assets

--------------------------

In order to see UI in Airflow, you need to compile front-end assets first.

In case you already installed `breeze` and `pre-commit`, you can build the assets with

the following commands:

pre-commit run --hook-stage manual compile-ui-assets --all-files

or simply:

breeze compile-ui-assets

Both commands will install node and pmpm under the hood, to a dedicated pre-commit

node environment and then build the assets.

If you want to manually run the build check for node and pnpm version in `.pre-commit-config.yaml` file,

and there are those manual ways to build the assets.

pnpm install -frozen-lockfile --config.confirmModulesPurge=false

pnpm run build

Finally, you can also clean and recompile assets with `custom` build target when running the Hatch build:

hatch build -t custom -t wheel -t sdist

This will also update `git_version` file in the Airflow package that should contain the git commit hash of the

build. This is used to display the commit hash in the UI. Result of this command is airflow package built

in the `dist` folder.

Using pip and manually managing your virtualenv

===============================================

While `uv` manages dependencies and venv automatically you might want to manage both manually with

pip and virtualenv. You need to have Python installed in your preferred way for that to work. It is also

way slower than with `uv` and you need to manage your environment manually.

Creating virtualenv

-------------------

Airflow pulls in quite a lot of dependencies to connect to other services. You generally want to

test or run Airflow from a virtualenv to ensure those dependencies are separated from your system-wide

versions. Using system-installed Python installation is strongly discouraged as the versions of Python

shipped with the operating system often have some limitations and are not up to date. It is recommended

to install Python using the official release (https://www.python.org/downloads/), or Python project

management tools such as Hatch. See later for a description of `Hatch` as one of the tools that

is Airflow's tool of choice to build Airflow packages.

Once you have a suitable Python version installed, you can create a virtualenv and activate it:

python3 -m venv PATH_TO_YOUR_VENV

source PATH_TO_YOUR_VENV/bin/activate

Installing Airflow locally with pip

-----------------------------------

Installing Airflow locally can be done using pip - note that this will install "development" version of

Airflow, where no providers are installed.

pip install .

If you develop Airflow and iterate on it, you should install it in editable mode (with -e) flag, and then

you do not need to re-install it after each change to sources.

pip install -e .

Note that for development you need to install also `devel-common` in editable way.

This distribution contains the set of tools and dependencies needed to run unit tests and perform all

development tasks.

pip install -e "./devel-common"

If you want to develop provider, you need to install it in editable mode from "providers" folder. For

example this will install Amazon provider in editable mode:

pip install -e "./providers/amazon"

However with pip in order to run tests, in some providers You also (for now) install optional extras that are

needed to run tests. In case of local installation for example, you can install all prerequisites for

Google provider, tests, and all Hadoop providers with this command:

pip install -e ".[google, hadoop]"

This is not needed with `uv` and in the near future, when `pip` will support https://peps.python.org/pep-0735/

this installation will be simplified.

You can also install ALL possible dependencies of airflow (in editable mode only) with:

pip install -e ".[all]"

This is not really recommended usually (it takes a lot of time to resolve and install all dependencies) but

if you want to replicate the CI environment where "all" extras are installed, you can do it this way.

You can also install ALL possible core dependencies of airflow (in editable mode only) with:

pip install -e ".[all-core]"

Building airflow packages with Hatch

====================================

While building packages will work with any compliant packaging front-end tool, for reproducibility, we

recommend using ``hatch``. It is a modern, fast, and convenient tool to build packages from sources managed

by the Python Packaging Authority. It is also used by Airflow to build packages in CI/CD as well as by

release managers to build locally packages for verification of reproducibility of the build.

Installing ``hatch``

--------------------

More information about hatch can be found in https://hatch.pypa.io/

We recommend to install ``hatch`` using ```uv tool`` command which will make hatch available as a CLI

command globally:

uv tool install hatch

You can still install ``hatch`` using ``pipx`` if you prefer:

pipx install hatch

It's important to keep your hatch up to date. You can do this by running:

uv tool upgrade hatch

Using Hatch to build your packages

----------------------------------

You can use Hatch to build installable packages from the Airflow sources. Such package will

include all metadata configured in `pyproject.toml` and will be installable with ``pip`` and and any other

PEP-compliant packaging front-end.

The packages will have pre-installed dependencies for providers that are available when Airflow is i

onstalled from PyPI. Both `wheel` and `sdist` packages are built by default.

hatch build

You can also build only `wheel` or `sdist` packages:

hatch build -t wheel

hatch build -t sdist

Installing recommended version of dependencies

==============================================

Whatever virtualenv solution you use, when you want to make sure you are using the same

version of dependencies as in main, you can install the recommended version of the dependencies by using

constraint-python<PYTHON_MAJOR_MINOR_VERSION>.txt files as `constraint` file. This might be useful

to avoid "works-for-me" syndrome, where you use different versions of dependencies than the ones

that are used in main CI tests and by other contributors.

There are different constraint files for different Python versions. For example, this command will install

all basic devel requirements and requirements of Google provider as last successfully tested for Python 3.9:

uv pip install -e ".[devel,google]"" \

--constraint "https://raw.githubusercontent.com/apache/airflow/constraints-main/constraints-3.9.txt"

Using the 'constraints-no-providers' constraint files, you can upgrade Airflow without paying attention to the provider's dependencies. This allows you to keep installed provider dependencies and install the latest supported ones using pure Airflow core.

uv pip install -e ".[devel]" \

--constraint "https://raw.githubusercontent.com/apache/airflow/constraints-main/constraints-no-providers-3.9.txt"

Note that you can also use `pip install` if you do not use `uv`.

Airflow extras

==============

Airflow has several extras that you can install to get additional dependencies. They sometimes install

providers, sometimes enable other features where packages are not installed by default.

You can read more about those extras in the extras reference:

https://airflow.apache.org/docs/apache-airflow/stable/extra-packages-ref.html

The list of available extras is below.

Core extras

-----------

Those extras are available as regular core airflow extras - they install optional features of Airflow.

# START CORE EXTRAS HERE

aiobotocore, apache-atlas, apache-webhdfs, async, cgroups, cloudpickle, github-enterprise, google-

auth, graphviz, kerberos, ldap, leveldb, otel, pandas, password, rabbitmq, s3fs, sentry, statsd, uv

# END CORE EXTRAS HERE

Provider extras

---------------

Those extras are available as regular Airflow extras; they install provider packages in standard builds

or dependencies that are necessary to enable the feature in an editable build.

# START PROVIDER EXTRAS HERE

airbyte, alibaba, amazon, apache.beam, apache.cassandra, apache.drill, apache.druid, apache.flink,

apache.hdfs, apache.hive, apache.iceberg, apache.impala, apache.kafka, apache.kylin, apache.livy,

apache.pig, apache.pinot, apache.spark, apprise, arangodb, asana, atlassian.jira, celery, cloudant,

cncf.kubernetes, cohere, common.compat, common.io, common.messaging, common.sql, databricks,

datadog, dbt.cloud, dingding, discord, docker, edge, elasticsearch, exasol, fab, facebook, ftp,

github, google, grpc, hashicorp, http, imap, influxdb, jdbc, jenkins, microsoft.azure,

microsoft.mssql, microsoft.psrp, microsoft.winrm, mongo, mysql, neo4j, odbc, openai, openfaas,

openlineage, opensearch, opsgenie, oracle, pagerduty, papermill, pgvector, pinecone, postgres,

presto, qdrant, redis, salesforce, samba, segment, sendgrid, sftp, singularity, slack, smtp,

snowflake, sqlite, ssh, standard, tableau, telegram, teradata, trino, vertica, weaviate, yandex,

ydb, zendesk

# END PROVIDER EXTRAS HERE

Doc extras

----------

Doc extras are used to install dependencies that are needed to build documentation. Only available during

editable install.

# START DOC EXTRAS HERE

doc, doc-gen

# END DOC EXTRAS HERE