Apache Airflow utilities for configuration of DAGs and DAG environments
This library allows for YAML-driven configuration of Airflow, including DAGs, Operators, and declaratively defined DAGs (à la dag-factory).
It is built with Pydantic, Hydra, and OmegaConf.
Consider the following basic DAG:
from airflow import DAG
from airflow.operators.bash import BashOperator
from datetime import datetime, timedelta
with DAG(
dag_id="test-dag",
default_args={
"depends_on_past": False,
"email": ["my.email@myemail.com"],
"email_on_failure": False,
"email_on_retry": False,
"retries": 0,
},
description="test that dag is working properly",
schedule=timedelta(minutes=1),
start_date=datetime(2024, 1, 1),
catchup=False,
tags=["utility", "test"],
):
BashOperator(
task_id="test-task",
bash_command="echo 'test'",
)We can already see many options that we might want to drive centrally via config, perhaps based on some notion of environment (e.g. dev, prod, etc).
"email": ["my.email@myemail.com"]"email_on_failure": False"email_on_retry": False"retries": 0schedule=timedelta(minutes=1)tags=["utility", "test"]
If we want to change these in our DAG, we need to modify code. Now imagine we have hundreds of DAGs, this can quickly get out of hand, especially since Airflow DAGs are Python code, and we might easily inject a syntax error or a trailing comma or other common problem.
Now consider the alternative, config-driven approach:
config/dev.yaml
# @package _global_
_target_: airflow_config.Configuration
default_args:
_target_: airflow_config.TaskArgs
owner: test
email: [myemail@myemail.com]
email_on_failure: false
email_on_retry: false
retries: 0
depends_on_past: false
default_dag_args:
_target_: airflow_config.DagArgs
schedule: "01:00"
start_date: "2024-01-01"
catchup: false
tags: ["utility", "test"]from airflow.operators.bash import BashOperator
from airflow_config import DAG, load_config
config = load_config(config_name="dev")
with DAG(
dag_id="test-dag",
description="test that dag is working properly",
schedule=timedelta(minutes=1),
config=config
):
BashOperator(
task_id="test-task",
bash_command="echo 'test'",
)This has a number of benefits:
- Make changes without code changes, with static type validation
- Make changes across any number of DAGs without having to copy-paste
- Organize collections of DAGs into groups, e.g. via enviroment like
dev,prod, etc
- Configure DAGs from a central config file or...
- from multiple env-specific config files (e.g.
dev,uat,prod) - Specialize DAGs by
dag_idfrom a single file (e.g. set each DAG'sschedulefrom a single shared file) - Generate entire DAGs declaratively, like astronomer/dag-factory
- Configure other extensions like:
- airflow-priority
- airflow-balancer
- airflow-supervisor
- or write your own pydantic-based model and get yaml-based configuration for free
class Configuration(BaseModel):
# default task args
# https://airflow.apache.org/docs/apache-airflow/stable/_api/airflow/models/baseoperator/index.html#airflow.models.baseoperator.BaseOperator
default_task_args: TaskArgs
# default dag args
# https://airflow.apache.org/docs/apache-airflow/stable/_api/airflow/models/dag/index.html#airflow.models.dag.DAG
default_dag_args: DagArgs
# string (dag id) to Dag mapping
dags: Optional[Dict[str, Dag]]
# string (dag id) to Task mapping
tasks: Optional[Dict[str, Task]]
# used for extensions to inject arbitrary configuration.
# See e.g.: https://github.com/airflow-laminar/airflow-supervisor?tab=readme-ov-file#example-dag-airflow-config
extensions: Optional[Dict[str, BaseModel]]Here is an example configuration defined via yaml:
# @package _global_
_target_: airflow_config.Configuration
default_task_args:
_target_: airflow_config.TaskArgs
owner: blerg
email: []
email_on_failure: false
email_on_retry: false
retries: 0
depends_on_past: false
default_dag_args:
_target_: airflow_config.DagArgs
start_date: ["2025-01-01", "America/New_York"]
catchup: false
max_active_runs: 1
dags:
reboot:
tags: ["reboot", "utility"]
description: "Reboot machines"
schedule: "0 0 * * *"
max_active_tasks: 1
clean-logs:
tags: ["celery", "utility"]
description: "Clean worker logs"
schedule: "0 4 * * *"This software is licensed under the Apache 2.0 license. See the LICENSE file for details.