dagster-turbine 0.5.13

Dagster integration for Turbine data quality checks

dagster-turbine

Dagster integration for Turbine. Surfaces every Check Definition the Turbine server knows about as a native Dagster AssetCheckSpec, executes each one against the running Turbine HTTP API, and publishes the same authenticated TurbineClient as a Dagster resource for your own assets.

Two entry points ship in the package:

• TurbineChecksComponent — declarative Component; configure once in YAML, get asset checks for free.
• TurbineServerResource — injectable resource for custom assets/checks that want to call Turbine directly.

Install

uv add dagster-turbine

Python 3.12 or newer. Pulls only dagster and turbine-client — the Turbine engine never enters your Dagster environment.

Minimal YAML (declarative)

# defs/quality/defs.yaml
type: dagster_turbine.TurbineChecksComponent
attributes:
  base_url: http://turbine:8000
  auth: { type: bearer, token: ${TURBINE_TOKEN} }

Reload your code location and every contract on the server appears as a multi_asset_check per table, one AssetCheckSpec per registered Turbine check.

Contract Selection

Narrow the discovery to a subset of contracts. Omit contracts (or set it to null) to cover every contract — the zero-config default.

type: dagster_turbine.TurbineChecksComponent
attributes:
  base_url: http://turbine:8000
  contracts: [orders, customers]
  auth: { type: bearer, token: ${TURBINE_TOKEN} }

Run options

type: dagster_turbine.TurbineChecksComponent
attributes:
  base_url: http://turbine:8000
  run:
    incremental: false   # rows newer than the last watermark; mutually exclusive with partition windows
    flag_rows: true      # persist failing-row PKs to the flag matrix

Partitioned assets automatically forward their partition window as the Check Window's since / until. Combining a partitioned asset with incremental: true is rejected at execute time — pick the data interval or the watermark.

Blocking (severity-driven)

Each generated AssetCheckSpec.blocking defaults to whether the underlying Check Definition's severity is error (the gating default). Advisory checks (severity: warn, authored in the Quality Spec) become non-blocking automatically.

Override per dimension when the topology demands it:

type: dagster_turbine.TurbineChecksComponent
attributes:
  base_url: http://turbine:8000
  blocking_dimensions: [validity]   # any check in this dimension blocks regardless of severity

Auth

Three modes, picked via the auth discriminator:

auth: { type: none }                                # in-cluster, unauthenticated
auth: { type: bearer, token: ${TURBINE_TOKEN} }     # static bearer token
auth:
  type: azure_ad                                    # OAuth client-credentials behind PAX
  tenant_id: ${AZ_TENANT_ID}
  client_id: ${AZ_CLIENT_ID}
  client_secret: ${AZ_CLIENT_SECRET}
  scope: api://turbine/.default

TurbineServerResource

Inject the same configured TurbineClient into your own assets:

import dagster as dg
from dagster_turbine import TurbineServerResource, BearerAuthConfig

@dg.asset
def turbine_contracts(turbine: TurbineServerResource):
    return turbine.get_client().list_contracts()

defs = dg.Definitions(
    assets=[turbine_contracts],
    resources={
        "turbine": TurbineServerResource(
            base_url="http://turbine:8000",
            auth=BearerAuthConfig(token="..."),
        ),
    },
)

Compatibility

dagster-turbine >= 0.5.12 requires turbine-data >= 0.5.12 on the server.