search

astronomer / dag-factory

Dynamically generate Apache Airflow DAGs from YAML configuration files
Apache License 2.0
1.21k stars 183 forks source link
airflow apache-airflow dags python

readme

dag-factory

Github Actions Coverage PyPi Code Style Downloads

Welcome to dag-factory! dag-factory is a library for Apache Airflow® to construct DAGs declaratively via configuration files.

The minimum requirements for dag-factory are:

For a gentle introduction, please take a look at our Quickstart Guide. For more examples, please see the examples folder.

Quickstart

The following example demonstrates how to create a simple DAG using dag-factory. We will be generating a DAG with three tasks, where task_2 and task_3 depend on task_1. These tasks will be leveraging the BashOperator to execute simple bash commands.

screenshot

  1. To install dag-factory, run the following pip command in your Apache Airflow® environment:

    pip install dag-factory

  2. Create a YAML configuration file called config_file.yml and save it within your dags folder:

    example_dag1:
default_args:
owner: 'example_owner'
retries: 1
start_date: '2024-01-01'
schedule_interval: '0 3 * * *'
catchup: False
description: 'this is an example dag!'
tasks:
task_1:
  operator: airflow.operators.bash_operator.BashOperator
  bash_command: 'echo 1'
task_2:
  operator: airflow.operators.bash_operator.BashOperator
  bash_command: 'echo 2'
  dependencies: [task_1]
task_3:
  operator: airflow.operators.bash_operator.BashOperator
  bash_command: 'echo 3'
  dependencies: [task_1]

    We are setting the execution order of the tasks by specifying the dependencies key.

  3. In the same folder, create a python file called generate_dags.py. This file is responsible for generating the DAGs from the configuration file and is a one-time setup. You won't need to modify this file unless you want to add more configuration files or change the configuration file name.

from airflow import DAG  ## by default, this is needed for the dagbag to parse this file
import dagfactory
from pathlib import Path

config_file = Path.cwd() / "dags/config_file.yml"
dag_factory = dagfactory.DagFactory(config_file)

dag_factory.clean_dags(globals())
dag_factory.generate_dags(globals())

After a few moments, the DAG will be generated and ready to run in Airflow. Unpause the DAG in the Apache Airflow® UI and watch the tasks execute!

screenshot

Please look at the examples folder for more examples.

Features

Multiple Configuration Files

If you want to split your DAG configuration into multiple files, you can do so by leveraging a suffix in the configuration file name.

# 'airflow' word is required for the dagbag to parse this file
from dagfactory import load_yaml_dags

load_yaml_dags(globals_dict=globals(), suffix=['dag.yaml'])

Dynamically Mapped Tasks

If you want to create a dynamic number of tasks, you can use the mapped_tasks key in the configuration file. The mapped_tasks key is a list of dictionaries, where each dictionary represents a task.

...
  tasks:
    request:
      operator: airflow.operators.python.PythonOperator
      python_callable_name: example_task_mapping
      python_callable_file: /usr/local/airflow/dags/expand_tasks.py # this file should contain the python callable
    process:
      operator: airflow.operators.python_operator.PythonOperator
      python_callable_name: expand_task
      python_callable_file: /usr/local/airflow/dags/expand_tasks.py
      partial:
        op_kwargs:
          test_id: "test"
      expand:
        op_args:
          request.output
      dependencies: [request]

mapped_tasks_example.png

Datasets

dag-factory supports scheduling DAGs via Apache Airflow Datasets.

To leverage, you need to specify the Dataset in the outlets key in the configuration file. The outlets key is a list of strings that represent the dataset locations. In the schedule key of the consumer dag, you can set the Dataset you would like to schedule against. The key is a list of strings that represent the dataset locations. The consumer dag will run when all the datasets are available.

producer_dag:
  default_args:
    owner: "example_owner"
    retries: 1
    start_date: '2024-01-01'
  description: "Example DAG producer simple datasets"
  schedule_interval: "0 5 * * *"
  tasks:
    task_1:
      operator: airflow.operators.bash_operator.BashOperator
      bash_command: "echo 1"
      outlets: [ 's3://bucket_example/raw/dataset1.json' ]
    task_2:
      bash_command: "echo 2"
      dependencies: [ task_1 ]
      outlets: [ 's3://bucket_example/raw/dataset2.json' ]
consumer_dag:
  default_args:
    owner: "example_owner"
    retries: 1
    start_date: '2024-01-01'
  description: "Example DAG consumer simple datasets"
  schedule: [ 's3://bucket_example/raw/dataset1.json', 's3://bucket_example/raw/dataset2.json' ]
  tasks:
    task_1:
      operator: airflow.operators.bash_operator.BashOperator
      bash_command: "echo 'consumer datasets'"

datasets_example.png

Custom Operators

dag-factory supports using custom operators. To leverage, set the path to the custom operator within the operator key in the configuration file. You can add any additional parameters that the custom operator requires.

...
  tasks:
    begin:
      operator: airflow.operators.dummy_operator.DummyOperator
    make_bread_1:
      operator: customized.operators.breakfast_operators.MakeBreadOperator
      bread_type: 'Sourdough'

custom_operators.png

Callbacks

dag-factory also supports using "callbacks" at the DAG, Task, and TaskGroup level. These callbacks can be defined in a few different ways. The first points directly to a Python function that has been defined in the include/callbacks.py file.

example_dag1:
  on_failure_callback: include.callbacks.example_callback1
...

Here, the on_success_callback points to first a file, and then to a function name within that file. Notice that this callback is defined using default_args, meaning this callback will be applied to all tasks.

example_dag1:
  ...
  default_args:
    on_success_callback_file: /usr/local/airflow/include/callbacks.py
    on_success_callback_name: example_callback1

dag-factory users can also leverage provider-built tools when configuring callbacks. In this example, the send_slack_notification function from the Slack provider is used to dispatch a message when a DAG failure occurs. This function is passed to the callback key under on_failure_callback. This pattern allows for callback definitions to take parameters (such as text, channel, and username, as shown here).

Note that this functionality is currently only supported for on_failure_callback's defined at the DAG-level, or in default_args. Support for other callback types and Task/TaskGroup-level definitions are coming soon.

example_dag1:
  on_failure_callback:
    callback: airflow.providers.slack.notifications.slack.send_slack_notification
    slack_conn_id: example_slack_id
    text: |
      :red_circle: Task Failed.
      This task has failed and needs to be addressed.
      Please remediate this issue ASAP.
    channel: analytics-alerts
    username: Airflow
...

Notes

HttpSensor (since 1.0.0)

The package airflow.providers.http.sensors.http is available for Airflow 2.0+

The following example shows response_check logic in a python file:

task_2:
  operator: airflow.providers.http.sensors.http.HttpSensor
  http_conn_id: 'test-http'
  method: 'GET'
  response_check_name: check_sensor
  response_check_file: /path/to/example1/http_conn.py
  dependencies: [task_1]

The response_check logic can also be provided as a lambda:

task_2:
  operator: airflow.providers.http.sensors.http.HttpSensor
  http_conn_id: 'test-http'
  method: 'GET'
  response_check_lambda: 'lambda response: "ok" in reponse.text'
  dependencies: [task_1]

Benefits

Contributing

Contributions are welcome! Just submit a Pull Request or Github Issue.