databricks-bundle-decorators 0.10.4

Decorator-based framework for defining Databricks jobs and tasks as Python code.

databricks-bundle-decorators

Decorator-based framework for defining Databricks jobs and tasks as Python code. Define pipelines using @task, @job, and job_cluster() — they compile into Databricks Asset Bundle resources.

Why databricks-bundle-decorators?

Writing Databricks jobs in raw YAML is tedious and disconnects task logic from orchestration configuration. databricks-bundle-decorators lets you express both in Python:

• Airflow TaskFlow-inspired pattern — define @task functions inside a @job body; dependencies are captured automatically from call arguments.
• IoManager pattern — large data (DataFrames, datasets) flows between tasks through external storage automatically.
• Explicit task values — small scalars (str, int, float, bool) can be passed between tasks via set_task_value / get_task_value, like Airflow XComs.
• Pure Python — write your jobs and tasks as decorated functions, run databricks bundle deploy, and the framework generates all Databricks Job configurations for you.

Installation

uv add databricks-bundle-decorators

With cloud-specific extras for the built-in PolarsParquetIoManager:

uv add databricks-bundle-decorators[azure]  # or [aws], [gcp], [polars]

Quickstart

uv init my-pipeline && cd my-pipeline
uv add databricks-bundle-decorators[azure]
uv run dbxdec init

This scaffolds a complete pipeline project. Define your jobs in src/<package>/pipelines/:

import polars as pl

from databricks_bundle_decorators import job, job_cluster, params, task
from databricks_bundle_decorators.io_managers import PolarsParquetIoManager

io = PolarsParquetIoManager(
    base_path="abfss://[e-mail]/staging",
)

cluster = job_cluster(
    name="small",
    spark_version="16.4.x-scala2.12",
    node_type_id="Standard_E8ds_v4",
    num_workers=1,
)

@job(
    params={"url": "https://api.github.com/events"},
    cluster=cluster,
)
def my_pipeline():
    @task(io_manager=io)
    def extract() -> pl.DataFrame:
        import requests
        return pl.DataFrame(requests.get(params["url"]).json())

    @task
    def transform(df: pl.DataFrame):
        print(df.head(10))

    data = extract()
    transform(data)

Deploy:

databricks bundle deploy --target dev

Documentation

Full documentation is available at boccileonardo.github.io/databricks-bundle-decorators:

• Getting Started — scaffolding, first pipeline, deploy
• How It Works — task dependencies, IoManager, task values
• Docker Deployment — pre-built container images
• API Reference — @task, @job, IoManager, and more

Development

git clone https://github.com/<org>/databricks-bundle-decorators.git
cd databricks-bundle-decorators
uv sync
uv run pytest tests/ -v

Releasing

Automated (recommended)

Run the release automation action, pick patch/minor/major. The workflow bumps the version in pyproject.toml, commits, tags, builds, creates a GitHub Release, and publishes to PyPI.

Manual

uv version --bump patch  # or minor, major
git commit -am "release: v$(uv version)" && git push
# Create a GitHub Release with the new tag → publish.yaml pushes to PyPI