Per-Database & Schema Config¶

You can set warehouse and role defaults per database or schema directory. This lets different parts of your project run with different compute or permissions without touching ~/.clair/environments.yml.

__database_config__.py¶

Place this file at the database directory level:

my_project/
└── reports/
    ├── __database_config__.py    ← applies to all Trouves under reports/
    └── orders/
        └── summary.py

# reports/__database_config__.py
from clair import DatabaseDefaults

defaults = DatabaseDefaults(
    warehouse="reporting_wh",   # larger warehouse for BI queries
    role="reporter",
)

__schema_config__.py¶

Place this file at the schema directory level to override the database config for that schema:

my_project/
└── reports/
    ├── __database_config__.py
    └── orders/
        ├── __schema_config__.py    ← applies to all Trouves under reports/orders/
        └── summary.py

# reports/orders/__schema_config__.py
from clair import SchemaDefaults

defaults = SchemaDefaults(
    warehouse="orders_wh",  # override for this schema only
)

Resolution order¶

For each Trouve, the effective warehouse and role are resolved in this order (later values win):

1. Environment defaults (warehouse and role from ~/.clair/environments.yml)
2. __database_config__.py in the Trouve's database directory
3. __schema_config__.py in the Trouve's schema directory

A SchemaDefaults value overrides a DatabaseDefaults value for the same field.

Field reference¶

Both DatabaseDefaults and SchemaDefaults support the same fields:

Fields not set (left as None) fall through to the next level in the resolution order.

Example: mixed warehouses¶

my_project/
├── source/
│   └── __database_config__.py    # defaults = DatabaseDefaults(warehouse="ingest_wh", role="loader")
├── refined/
│   └── __database_config__.py    # defaults = DatabaseDefaults(warehouse="transform_wh", role="transformer")
└── reports/
    └── __database_config__.py    # defaults = DatabaseDefaults(warehouse="reporting_wh", role="reporter")

Each database runs on the appropriate compute without any per-command flags.