Modules Management

Airflow allows you to use your own Python modules in the DAG and in the Airflow configuration. The following article will describe how you can create your own module so that Airflow can load it correctly, as well as diagnose problems when modules are not loaded properly.

Often you want to use your own python code in your Airflow deployment, for example common code, libraries, you might want to generate DAGs using shared python code and have several DAG python files.

You can do it in one of those ways:

  • add your modules to one of the folders that Airflow automatically adds to PYTHONPATH

  • add extra folders where you keep your code to PYTHONPATH

  • package your code into a Python package and install it together with Airflow.

The next chapter has a general description of how Python loads packages and modules, and dives deeper into the specifics of each of the three possibilities above.

How package/modules loading in Python works

The list of directories from which Python tries to load the module is given by the variable sys.path. Python really tries to intelligently determine the contents of this variable, including depending on the operating system and how Python is installed and which Python version is used.

You can check the contents of this variable for the current Python environment by running an interactive terminal as in the example below:

>>> import sys
>>> from pprint import pprint
>>> pprint(sys.path)
['',
 '/home/arch/.pyenv/versions/3.7.4/lib/python37.zip',
 '/home/arch/.pyenv/versions/3.7.4/lib/python3.7',
 '/home/arch/.pyenv/versions/3.7.4/lib/python3.7/lib-dynload',
 '/home/arch/venvs/airflow/lib/python3.7/site-packages']

sys.path is initialized during program startup. The first precedence is given to the current directory, i.e, path[0] is the directory containing the current script that was used to invoke or an empty string in case it was an interactive shell. Second precedence is given to the PYTHONPATH if provided, followed by installation-dependent default paths which is managed by site module.

sys.path can also be modified during a Python session by simply using append (for example, sys.path.append("/path/to/custom/package")). Python will start searching for packages in the newer paths once they’re added. Airflow makes use of this feature as described in the section Adding directories to the PYTHONPATH.

In the variable sys.path there is a directory site-packages which contains the installed external packages, which means you can install packages with pip or anaconda and you can use them in Airflow. In the next section, you will learn how to create your own simple installable package and how to specify additional directories to be added to sys.path using the environment variable PYTHONPATH.

Also make sure to Add init file to your folders.

Typical structure of packages

This is an example structure that you might have in your dags folder:

<DIRECTORY ON PYTHONPATH>
| .airflowignore  -- only needed in ``dags`` folder, see below
| -- my_company
              | __init__.py
              | common_package
              |              |  __init__.py
              |              | common_module.py
              |              | subpackage
              |                         | __init__.py
              |                         | subpackaged_util_module.py
              |
              | my_custom_dags
                              | __init__.py
                              | my_dag1.py
                              | my_dag2.py
                              | base_dag.py

In the case above, these are the ways you could import the python files:

from my_company.common_package.common_module import SomeClass
from my_company.common_package.subpackage.subpackaged_util_module import AnotherClass
from my_company.my_custom_dags.base_dag import BaseDag

You can see the .airflowignore file at the root of your folder. This is a file that you can put in your dags folder to tell Airflow which files from the folder should be ignored when the Airflow scheduler looks for DAGs. It should contain either regular expressions (the default) or glob expressions for the paths that should be ignored. You do not need to have that file in any other folder in PYTHONPATH (and also you can only keep shared code in the other folders, not the actual DAGs).

In the example above the dags are only in my_custom_dags folder, the common_package should not be scanned by scheduler when searching for DAGS, so we should ignore common_package folder. You also want to ignore the base_dag.py if you keep a base DAG there that my_dag1.py and my_dag2.py derives from. Your .airflowignore should look then like this:

my_company/common_package/.*
my_company/my_custom_dags/base_dag\.py

If DAG_IGNORE_FILE_SYNTAX is set to glob, the equivalent .airflowignore file would be:

my_company/common_package/
my_company/my_custom_dags/base_dag.py

Built-in PYTHONPATH entries in Airflow

Airflow, when running dynamically adds three directories to the sys.path:

  • The dags folder: It is configured with option dags_folder in section [core].

  • The config folder: It is configured by setting AIRFLOW_HOME variable ({AIRFLOW_HOME}/config) by default.

  • The plugins Folder: It is configured with option plugins_folder in section [core].

Note

The DAGS folder in Airflow 2 should not be shared with the webserver. While you can do it, unlike in Airflow 1.10, Airflow has no expectations that the DAGS folder is present in the webserver. In fact it’s a bit of security risk to share the dags folder with the webserver, because it means that people who write DAGS can write code that the webserver will be able to execute (ideally the webserver should never run code which can be modified by users who write DAGs). Therefore if you need to share some code with the webserver, it is highly recommended that you share it via config or plugins folder or via installed Airflow packages (see below). Those folders are usually managed and accessible by different users (Admins/DevOps) than DAG folders (those are usually data-scientists), so they are considered as safe because they are part of configuration of the Airflow installation and controlled by the people managing the installation.

Best practices for module loading

There are a few gotchas you should be careful about when you import your code.

Use unique top package name

It is recommended that you always put your dags/common files in a subpackage which is unique to your deployment (my_company in the example below). It is far too easy to use generic names for the folders that will clash with other packages already present in the system. For example if you create airflow/operators subfolder it will not be accessible because Airflow already has a package named airflow.operators and it will look there when importing from airflow.operators.

Don’t use relative imports

Never use relative imports (starting with .) that were added in Python 3.

This is tempting to do something like that it in my_dag1.py:

from .base_dag import BaseDag  # NEVER DO THAT!!!!

You should import such shared dag using full path (starting from the directory which is added to PYTHONPATH):

from my_company.my_custom_dags.base_dag import BaseDag  # This is cool

The relative imports are counter-intuitive, and depending on how you start your python code, they can behave differently. In Airflow the same DAG file might be parsed in different contexts (by schedulers, by workers or during tests) and in those cases, relative imports might behave differently. Always use full python package paths when you import anything in Airflow DAGs, this will save you a lot of troubles. You can read more about relative import caveats in this Stack Overflow thread.

Add __init__.py in package folders

When you create folders you should add __init__.py file as empty files in your folders. While in Python 3 there is a concept of implicit namespaces where you do not have to add those files to folder, Airflow expects that the files are added to all packages you added.

Inspecting your PYTHONPATH loading configuration

You can also see the exact paths using the airflow info command, and use them similar to directories specified with the environment variable PYTHONPATH. An example of the contents of the sys.path variable specified by this command may be as follows:

Python PATH: [/home/rootcss/venvs/airflow/bin:/usr/lib/python38.zip:/usr/lib/python3.8:/usr/lib/python3.8/lib-dynload:/home/rootcss/venvs/airflow/lib/python3.8/site-packages:/home/rootcss/airflow/dags:/home/rootcss/airflow/config:/home/rootcss/airflow/plugins]

Below is the sample output of the airflow info command:

Apache Airflow: 2.0.0b3

System info
OS              | Linux
architecture    | x86_64
uname           | uname_result(system='Linux', node='85cd7ab7018e', release='4.19.76-linuxkit', version='#1 SMP Tue May 26 11:42:35 UTC 2020', machine='x86_64', processor='')
locale          | ('en_US', 'UTF-8')
python_version  | 3.8.6 (default, Nov 25 2020, 02:47:44)  [GCC 8.3.0]
python_location | /usr/local/bin/python

Tools info
git             | git version 2.20.1
ssh             | OpenSSH_7.9p1 Debian-10+deb10u2, OpenSSL 1.1.1d  10 Sep 2019
kubectl         | NOT AVAILABLE
gcloud          | NOT AVAILABLE
cloud_sql_proxy | NOT AVAILABLE
mysql           | mysql  Ver 8.0.22 for Linux on x86_64 (MySQL Community Server - GPL)
sqlite3         | 3.27.2 2019-02-25 16:06:06 bd49a8271d650fa89e446b42e513b595a717b9212c91dd384aab871fc1d0alt1
psql            | psql (PostgreSQL) 11.9 (Debian 11.9-0+deb10u1)

Paths info
airflow_home    | /root/airflow
system_path     | /usr/local/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
python_path     | /usr/local/bin:/opt/airflow:/files/plugins:/usr/local/lib/python38.zip:/usr/local/lib/python3.8:/usr/
                | local/lib/python3.8/lib-dynload:/usr/local/lib/python3.8/site-packages:/files/dags:/root/airflow/conf
                | ig:/root/airflow/plugins
airflow_on_path | True

Config info
executor             | LocalExecutor
task_logging_handler | airflow.utils.log.file_task_handler.FileTaskHandler
sql_alchemy_conn     | postgresql+psycopg2://postgres:airflow@postgres/airflow
dags_folder          | /files/dags
plugins_folder       | /root/airflow/plugins
base_log_folder      | /root/airflow/logs

Providers info
apache-airflow-providers-amazon           | 1.0.0b2
apache-airflow-providers-apache-cassandra | 1.0.0b2
apache-airflow-providers-apache-druid     | 1.0.0b2
apache-airflow-providers-apache-hdfs      | 1.0.0b2
apache-airflow-providers-apache-hive      | 1.0.0b2

Adding directories to the PYTHONPATH

You can specify additional directories to be added to sys.path using the environment variable PYTHONPATH. Start the python shell by providing the path to root of your project using the following command:

PYTHONPATH=/home/arch/projects/airflow_operators python

The sys.path variable will look like below:

>>> import sys
>>> from pprint import pprint
>>> pprint(sys.path)
['',
 '/home/arch/projects/airflow_operators'
 '/home/arch/.pyenv/versions/3.7.4/lib/python37.zip',
 '/home/arch/.pyenv/versions/3.7.4/lib/python3.7',
 '/home/arch/.pyenv/versions/3.7.4/lib/python3.7/lib-dynload',
 '/home/arch/venvs/airflow/lib/python3.7/site-packages']

As we can see that our provided directory is now added to the path, let’s try to import the package now:

>>> import airflow_operators
Hello from airflow_operators
>>>

We can also use PYTHONPATH variable with the airflow commands. For example, if we run the following airflow command:

PYTHONPATH=/home/arch/projects/airflow_operators airflow info

We’ll see the Python PATH updated with our mentioned PYTHONPATH value as shown below:

Python PATH: [/home/arch/venv/bin:/home/arch/projects/airflow_operators:/usr/lib/python38.zip:/usr/lib/python3.8:/usr/lib/python3.8/lib-dynload:/home/arch/venv/lib/python3.8/site-packages:/home/arch/airflow/dags:/home/arch/airflow/config:/home/arch/airflow/plugins]

Creating a package in Python

This is most organized way of adding your custom code. Thanks to using packages, you might organize your versioning approach, control which versions of the shared code are installed and deploy the code to all your instances and containers in controlled way - all by system admins/DevOps rather than by the DAG writers. It is usually suitable when you have a separate team that manages this shared code, but if you know your python ways you can also distribute your code this way in smaller deployments. You can also install your Plugins and Provider packages as python packages, so learning how to build your package is handy.

Here is how to create your package:

  1. Before starting, install the following packages:

setuptools: setuptools is a package development process library designed for creating and distributing Python packages.

wheel: The wheel package provides a bdist_wheel command for setuptools. It creates .whl file which is directly installable through the pip install command. We can then upload the same file to PyPI.

pip install --upgrade pip setuptools wheel
  1. Create the package directory - in our case, we will call it airflow_operators.

mkdir airflow_operators
  1. Create the file __init__.py inside the package and add following code:

print("Hello from airflow_operators")

When we import this package, it should print the above message.

  1. Create setup.py:

import setuptools

setuptools.setup(
    name="airflow_operators",
    packages=setuptools.find_packages(),
)
  1. Build the wheel:

python setup.py bdist_wheel

This will create a few directories in the project and the overall structure will look like following:

.
├── airflow_operators
│   ├── __init__.py
├── airflow_operators.egg-info
│   ├── PKG-INFO
│   ├── SOURCES.txt
│   ├── dependency_links.txt
│   └── top_level.txt
├── build
│   └── bdist.macosx-10.15-x86_64
├── dist
│   └── airflow_operators-0.0.0-py3-none-any.whl
└── setup.py
  1. Install the .whl file using pip:

pip install dist/airflow_operators-0.0.0-py3-none-any.whl
  1. The package is now ready to use!

>>> import airflow_operators
Hello from airflow_operators
>>>

The package can be removed using pip command:

pip uninstall airflow_operators

For more details on how to create to create and publish python packages, see Packaging Python Projects.

Was this entry helpful?