How to choose and configure a backend¶

Semolina supports multiple data warehouse backends:

Snowflake – via semolina[snowflake]
Databricks – via semolina[databricks]

The query API is identical for both – only the connection configuration changes.

Register a connection pool¶

Two patterns are available for creating and registering a connection pool.

TOML config (recommended)¶

from semolina import register, pool_from_config

pool, dialect = pool_from_config()  # reads .semolina.toml
register("default", pool, dialect=dialect)

pool_from_config() reads .semolina.toml from the current directory, creates an adbc-poolhouse connection pool, and returns it with the matching dialect. See How to connect to Snowflake or How to connect to Databricks for the TOML fields.

Manual pool construction¶

Use manual construction when credentials come from a vault, secrets manager, or need programmatic configuration.

Snowflake

from adbc_poolhouse import SnowflakeConfig, create_pool
from semolina import register

config = SnowflakeConfig(
    account="xy12345.us-east-1",
    user="myuser",
    password="mypassword",
    database="analytics",
    warehouse="compute_wh",
)
pool = create_pool(config)
register("default", pool, dialect="snowflake")

Databricks

from adbc_poolhouse import DatabricksConfig, create_pool
from semolina import register

config = DatabricksConfig(
    host="workspace.cloud.databricks.com",
    http_path="/sql/1.0/warehouses/abc123",
    token="dapi...",
)
pool = create_pool(config)
register("default", pool, dialect="databricks")

Query with a registered pool¶

Once a pool is registered, the query API works the same regardless of backend:

from semolina import SemanticView, Metric, Dimension


class Sales(SemanticView, view="sales"):
    revenue = Metric()
    country = Dimension()


cursor = (
    Sales.query()
    .metrics(Sales.revenue)
    .dimensions(Sales.country)
    .execute()
)

for row in cursor.fetchall_rows():
    print(row.country, row.revenue)

Test locally without a warehouse¶

MockPool accepts fixture data and returns it through the same cursor interface, so you can test query logic without any warehouse connection:

from semolina import MockPool, register

pool = MockPool()
pool.load(
    "sales",
    [
        {"revenue": 1000, "country": "US"},
        {"revenue": 2000, "country": "CA"},
    ],
)
register("default", pool, dialect="mock")

See How to test application code with MockPool for testing patterns with MockPool.