Skip to main content

Using Delta Lake with Dagster

This tutorial focuses on how to store and load Dagster asset definitions in a Delta Lake.

By the end of the tutorial, you will:

  • Configure a Delta Lake I/O manager
  • Create a table in Delta Lake using a Dagster asset
  • Make a Delta Lake table available in Dagster
  • Load Delta tables in downstream assets

While this guide focuses on storing and loading Pandas DataFrames in Delta Lakes, Dagster also supports using PyArrow Tables and Polars DataFrames. Learn more about setting up and using the Delta Lake I/O manager with PyArrow Tables and Polars DataFrames in the Delta Lake reference.

Prerequisites​

To complete this tutorial, you'll need to install the dagster-deltalake and dagster-deltalake-pandas libraries:

pip install dagster-deltalake dagster-deltalake-pandas

Step 1: Configure the Delta Lake I/O manager​

The Delta Lake I/O manager requires some configuration to set up your Delta Lake. You must provide a root path where your Delta tables will be created. Additionally, you can specify a schema where the Delta Lake I/O manager will create tables.


from dagster_deltalake import LocalConfig
from dagster_deltalake_pandas import DeltaLakePandasIOManager

from dagster import Definitions

defs = Definitions(
    assets=[iris_dataset],
    resources={
        "io_manager": DeltaLakePandasIOManager(
            root_uri="path/to/deltalake",  # required
            storage_options=LocalConfig(),  # required
            schema="iris",  # optional, defaults to "public"
        )
    },
)

With this configuration, if you materialized an asset called iris_dataset, the Delta Lake I/O manager would store the data within a folder iris/iris_dataset under the provided root directory path/to/deltalake.

Finally, in the Definitions object, we assign the DeltaLakePandasIOManager to the io_manager key. io_manager is a reserved key to set the default I/O manager for your assets.

Step 2: Create Delta Lake tables​

The Delta Lake I/O manager can create and update tables for your Dagster-defined assets, but you can also make existing Delta Lake tables available to Dagster.

Store a Dagster asset as a table in Delta Lake

To store data in Delta Lake using the Delta Lake I/O manager, the definitions of your assets don't need to change. You can tell Dagster to use the Delta Lake I/O manager, like in Step 1, and Dagster will handle storing and loading your assets in Delta Lake.

import pandas as pd

from dagster import asset


@asset
def iris_dataset() -> pd.DataFrame:
    return pd.read_csv(
        "https://docs.dagster.io/assets/iris.csv",
        names=[
            "sepal_length_cm",
            "sepal_width_cm",
            "petal_length_cm",
            "petal_width_cm",
            "species",
        ],
    )

In this example, we first define an asset. Here, we fetch the Iris dataset as a Pandas DataFrame and rename the columns. The type signature of the function tells the I/O manager what data type it is working with, so it's important to include the return type pd.DataFrame.

When Dagster materializes the iris_dataset asset using the configuration from Step 1, the Delta Lake I/O manager will create the table iris/iris_dataset if it doesn't exist and replace the contents of the table with the value returned from the iris_dataset asset.

Step 3: Load Delta Lake tables in downstream assets​

Once you've created an asset that represents a table in your Delta Lake, you will likely want to create additional assets that work with the data. Dagster and the Delta Lake I/O manager allow you to load the data stored in Delta tables into downstream assets.


import pandas as pd

from dagster import asset

# this example uses the iris_dataset asset from Step 2


@asset
def iris_cleaned(iris_dataset: pd.DataFrame) -> pd.DataFrame:
    return iris_dataset.dropna().drop_duplicates()

In this example, we want to provide the iris_dataset asset to the iris_cleaned asset. Refer to the Store a Dagster asset as a table in Delta Lake example in step 2 for a look at the iris_dataset asset.

In iris_cleaned, the iris_dataset parameter tells Dagster that the value for the iris_dataset asset should be provided as input to iris_cleaned. If this feels too magical for you, refer to the docs for explicitly specifying dependencies.

When materializing these assets, Dagster will use the DeltaLakePandasIOManager to fetch the iris/iris_dataset as a Pandas DataFrame and pass the DataFrame as the iris_dataset parameter to iris_cleaned. When iris_cleaned returns a Pandas DataFrame, Dagster will use the DeltaLakePandasIOManager to store the DataFrame as the iris/iris_cleaned table in your Delta Lake.

Completed code example​

When finished, your code should look like the following:

import pandas as pd
from dagster_deltalake import LocalConfig
from dagster_deltalake_pandas import DeltaLakePandasIOManager

from dagster import AssetSpec, Definitions, asset

iris_harvest_data = AssetSpec(key="iris_harvest_data")


@asset
def iris_dataset() -> pd.DataFrame:
    return pd.read_csv(
        "https://docs.dagster.io/assets/iris.csv",
        names=[
            "sepal_length_cm",
            "sepal_width_cm",
            "petal_length_cm",
            "petal_width_cm",
            "species",
        ],
    )


@asset
def iris_cleaned(iris_dataset: pd.DataFrame) -> pd.DataFrame:
    return iris_dataset.dropna().drop_duplicates()


defs = Definitions(
    assets=[iris_dataset, iris_harvest_data, iris_cleaned],
    resources={
        "io_manager": DeltaLakePandasIOManager(
            root_uri="path/to/deltalake",
            storage_options=LocalConfig(),
            schema="IRIS",
        )
    },
)

For more Delta Lake features, refer to the Delta Lake reference.

For more information on asset definitions, see the Assets documentation.

For more information on I/O managers, refer to the I/O manager documentation.