1) Materialize code-managed reference data

assets/python/customer_regions.py is a Python asset. Bruin runs its materialize() function in the declared python:3.13 image, validates its returned rows against the asset schema, and writes a create+replace ClickHouse table.

The file starts with a Python docstring that Bruin reads as the asset definition:

"""@bruin

name: customer_regions
connection: clickhouse-default

materialization:
  type: table
  strategy: create+replace
image: python:3.13

parameters:
  enforce_schema: true

columns:
  - name: country
    type: String
    primary_key: true
    checks:
      - name: not_null
      - name: unique
  - name: sales_region
    type: LowCardinality(String)
    checks:
      - name: not_null
  - name: support_tier
    type: LowCardinality(String)
    checks:
      - name: accepted_values
        value: [strategic, standard]

@bruin"""

connection chooses the ClickHouse destination and image chooses the Python runtime. The declared schema is still the contract: enforce_schema: true validates what the Python function returns, country is unique, and support_tier is limited to the two known values.

The executable part is ordinary Python:

def materialize():
    return [
        {
            "country": "United Kingdom",
            "sales_region": "EMEA",
            "support_tier": "strategic",
        },
        {
            "country": "United States",
            "sales_region": "North America",
            "support_tier": "strategic",
        },
        {
            "country": "Austria",
            "sales_region": "EMEA",
            "support_tier": "standard",
        },
    ]

Bruin calls materialize() and writes the returned dictionaries into customer_regions. The dictionary keys must match the declared columns. create+replace then replaces the complete table, which suits this tiny code-managed lookup.

The function returns three country-to-region mappings. The asset also defines an accepted-values check for support_tier, a primary key on country, enforce_schema: true, and the same ownership and metadata fields seen in the SQL assets. country_revenue.sql then treats customer_regions as an ordinary upstream table.

This is the pattern to use when a small reference dataset is better expressed as reviewed code than as a CSV. See the Python asset reference and the Python Materialization tutorial for a fuller example.