Managing semantic metric views#

Cosmos can render adapter-native semantic metric views (Databricks Unity Catalog metric views and Snowflake semantic views) as their own semantic_metric_view resource type, instead of a regular model.

This is opt-in. Enable it with the enable_semantic_metric_view_reclassification: setting (AIRFLOW__COSMOS__ENABLE_SEMANTIC_METRIC_VIEW_RECLASSIFICATION=True).

Prerequisites#

How it works#

dbt has no distinct resource type for these objects. They are model nodes with config.materialized: metric_view (Databricks) or semantic_view (Snowflake). Once enabled, Cosmos detects that materialization at parse time and reclassifies the node as semantic_metric_view, which changes:

  • Task identity: a _semantic_metric_run task name suffix and a DbtSemanticMetricView* operator instead of DbtRun*. The underlying dbt command is still dbt run.

  • TestBehavior.BUILD: collapses into DbtBuild, same as any other buildable resource.

  • Selectors: resource_type:model no longer matches it. Use resource_type:semantic_metric_view instead. This selector only works when Cosmos itself performs node selection (e.g. LoadMode.DBT_MANIFEST). Under LoadMode.DBT_LS, select/exclude are passed directly to dbt ls, which has no semantic_metric_view resource type, so use config.materialized:metric_view or config.materialized:semantic_view instead. The same applies to ExecutionMode.WATCHER, whose producer passes select/exclude to dbt build. Cosmos raises an error if they contain resource_type:semantic_metric_view.

Testing#

Cosmos renders test tasks for semantic metric views like for any other model. Generic tests work on Databricks metric views. On Snowflake, tests that aggregate over a column, such as accepted_values, fail on semantic views. Use singular tests written with SEMANTIC_VIEW(...) instead.

Example#

{{ config(materialized='metric_view') }}

select * from {{ ref('customers') }}

from cosmos import DbtTaskGroup, LoadMode, RenderConfig

# LoadMode.DBT_MANIFEST (or another mode where Cosmos applies selection itself)
semantic_metric_views_only = DbtTaskGroup(
    render_config=RenderConfig(
        load_method=LoadMode.DBT_MANIFEST, select=["resource_type:semantic_metric_view"]
    ),
)

# LoadMode.DBT_LS: select/exclude are passed straight to `dbt ls`, which has no `semantic_metric_view`
# resource type, so filter on the underlying materialization instead
semantic_metric_views_only_dbt_ls = DbtTaskGroup(
    render_config=RenderConfig(
        load_method=LoadMode.DBT_LS, select=["config.materialized:metric_view"]
    ),
)

See Selecting & Excluding for the full selector reference.