dbt-airflow

A Python package that helps Data and Analytics engineers render dbt projects in Apache Airflow DAGs such that models, seeds, snapshots and tests are represented by individual Airflow Task.

dbt is a command-line tool that enables data teams build, maintain and test data models in a scalable fashion. The biggest challenge though is how to embed dbt in modern data workflows and infrastructure. dbt CLI is indeed a powerful tool, but if used as is, it will create silos in the way an organisation manages its data. Every contributor is able to run dbt commands from their local machine (or even a host machine), but how do you know if a model run by another contributor has failed, or succeeded? How can you enable shared visibility over data models, within the team?

One way to host dbt projects and orchestrate dbt tasks is via Apache Airflow. In its simplest form, an Airflow DAG that will build and test data models will consist of two tasks, one that executes dbt run command followed by an Airflow task that executes dbt test.

But what happens when model builds or tests fail? Should we re-run the whole dbt project (that could involve hundreds of different models and/or tests) just to run a single model we've just fixed? This doesn't seem to be a good practice since re-running the whole project will be time-consuming and expensive.

A potential solution to this problem is to create individual Airflow tasks for every model, seed, snapshot and test within the dbt project. If we were about to do this work manually, we would have to put huge effort that would also be prone to errors. Additionally, it would beat the purpose of dbt, that among other features, it also automates model dependency management.

dbt-airflow is a package that builds a layer in-between Apache Airflow and dbt, and enables teams to automatically render their dbt projects in a granular level such that they have full control to individual dbt resource types. Every dbt model, seed, snapshot or test will have its own Airflow Task so that you can perform any action at a task-level.

Here's how the popular Jaffle Shop dbt project will be rendered on Apache Airflow via dbt-airflow:

Features

• Render a dbt project as a TaskGroup consisting of Airflow Tasks that correspond to dbt models, seeds, snapshots and tests
• Every model, seed and snapshot resource that has at least a single test, will also have a corresponding test task as a downstream task
• Add tasks before or after the whole dbt project
• Introduce extra tasks within the dbt project tasks and specify any downstream or upstream dependencies
• Create sub-TaskGroups of dbt Airflow tasks based on your project's folder structure

How does it work

The library essentially builds on top of the metadata generated by dbt-core and are stored in the target/manifest.json file in your dbt project directory. This means that you first need to compile (or run any other dbt command that creates the manifest file) before creating your Airflow DAG. This means the dbt-airflow package expects that you have already compiled your dbt project so that an up to date manifest file can then be used to render the individual tasks.

Installation

The package is available on PyPI and can be installed through pip:

pip install dbt-airflow

dbt needs to connect to your target environment (database, warehouse etc.) and in order to do so, it makes use of different adapters, each dedicated to a different technology (such as Postgres or BigQuery). Therefore, before running dbt-airflow you also need to ensure that the required adapter(s) are installed in your environment.

For the full list of available adapters please refer to the official dbt documentation.

Usage

Building an Airflow DAG using dbt-airflow

from datetime import datetime, timedelta
from pathlib import Path

from airflow import DAG
from airflow.operators.python import PythonOperator
from airflow.operators.empty import EmptyOperator

from dbt_airflow.core.config import DbtAirflowConfig, DbtProjectConfig, DbtProfileConfig
from dbt_airflow.core.task_group import DbtTaskGroup
from dbt_airflow.core.task import ExtraTask
from dbt_airflow.operators.execution import ExecutionOperator

with DAG(
    dag_id='test_dag',
    start_date=datetime(2021, 1, 1),
    catchup=False,
    tags=['example'],
    default_args={
        'owner': 'airflow',
        'retries': 1,
        'retry_delay': timedelta(minutes=2),
    },
) as dag:
    extra_tasks = [
        ExtraTask(
            task_id='test_task',
            operator=PythonOperator,
            operator_args={
                'python_callable': lambda: print('Hello world'),
            },
            upstream_task_ids={
                'model.example_dbt_project.int_customers_per_store',
                'model.example_dbt_project.int_revenue_by_date',
            },
        ),
        ExtraTask(
            task_id='another_test_task',
            operator=PythonOperator,
            operator_args={
                'python_callable': lambda: print('Hello world 2!'),
            },
            upstream_task_ids={
                'test.example_dbt_project.int_customers_per_store',
            },
            downstream_task_ids={
                'snapshot.example_dbt_project.int_customers_per_store_snapshot',
            },
        ),
        ExtraTask(
            task_id='test_task_3',
            operator=PythonOperator,
            operator_args={
                'python_callable': lambda: print('Hello world 3!'),
            },
            downstream_task_ids={
                'snapshot.example_dbt_project.int_customers_per_store_snapshot',
            },
            upstream_task_ids={
                'model.example_dbt_project.int_revenue_by_date',
            },
        ),
    ]

    t1 = EmptyOperator(task_id='dummy_1')
    t2 = EmptyOperator(task_id='dummy_2')

    tg = DbtTaskGroup(
        group_id='dbt-company',
        dbt_project_config=DbtProjectConfig(
            project_path=Path('/opt/airflow/example_dbt_project/'),
            manifest_path=Path('/opt/airflow/example_dbt_project/target/manifest.json'),
        ),
        dbt_profile_config=DbtProfileConfig(
            profiles_path=Path('/opt/airflow/example_dbt_project/profiles'),
            target='dev',
        ),
        dbt_airflow_config=DbtAirflowConfig(
            extra_tasks=extra_tasks,
            execution_operator=ExecutionOperator.BASH,
            test_tasks_operator_kwargs={'retries': 0},
        ),
    )

    t1 >> tg >> t2