How do dbt models relate to Dagster assets?

dbt models are assets: they produce data and can have dependencies. Because of these similarities, Dagster can translate each of your dbt models into a Dagster Software-defined Asset (SDA).

How can Dagster do this? Each component of a Dagster asset has an equivalent counterpart in a dbt model:

• The asset key for a dbt model is (by default) the name of the model
• The upstream dependencies of a dbt model are defined with ref or source calls within the model's definition
• The computation required to compute the asset from its upstream dependencies is the SQL within the model's definition

These similarities make it natural to interact with dbt models as Dagster assets. Using dbt with Dagster, you can create an asset graph like the following:

From code like this:

from pathlib import Path

from dagster_dbt import DbtCliResource, dbt_assets, get_asset_key_for_model
from dagster_fivetran import build_fivetran_assets

from dagster import AssetExecutionContext, asset

fivetran_assets = build_fivetran_assets(
    connector_id="postgres",
    destination_tables=["users", "orders"],
)


@dbt_assets(manifest=Path("manifest.json"))
def dbt_project_assets(context: AssetExecutionContext, dbt: DbtCliResource):
    yield from dbt.cli(["build"], context=context).stream()


@asset(
    compute_kind="tensorflow",
    deps=[get_asset_key_for_model([dbt_project_assets], "daily_order_summary")],
)
def predicted_orders():
    ...

Let's break down what's happening in this example:

• Using build_fivetran_assets, we load two tables (users, orders) from a Fivetran Postgres connector as Dagster assets
• Using @dbt_assets, Dagster reads from a dbt project's manifest.json and creates Dagster assets from the dbt models it finds
• Lastly, we create a Dagster @asset named predicted_orders that has an upstream dependency on a dbt asset named daily_order_summary