Configuration¶
canary_framework.config is one class: Settings, which is both a
pydantic-settings
BaseSettings and a unit.
Provisional API
Extensions may change in a minor release until they are declared stable — see Versioning & Compatibility.
Usage¶
from canary_framework import Canary, dep, start
from canary_framework.config import Settings
from pydantic_settings import SettingsConfigDict
class AppConfig(Settings):
model_config = SettingsConfigDict(env_file=".env", env_prefix="APP_")
dsn: str = "sqlite://"
pool_size: int = 5
class Database(Canary):
config = dep(AppConfig)
@start
async def connect(self) -> None:
self.pool = await open_pool(self.config.dsn, size=self.config.pool_size)
Reading the environment and .env, and validating types, is all pydantic-settings — written
exactly as its own docs describe. Settings adds one thing: it is a unit, so whoever needs it
writes dep(AppConfig) and the whole graph shares one instance.
Invalid configuration raises its validation error in the init phase, when this
configuration unit is constructed — before any unit has run a @start.
No-argument construction¶
The framework constructs every unit with no arguments, settings included: every field must be obtainable from a default or from the environment.
Different settings in a test¶
from canary_framework.test import running
async with running(Service, {AppConfig: AppConfig(dsn="sqlite://:memory:")}) as service:
...
Dependencies and hooks of its own¶
Settings is an ordinary unit, so it can declare dependencies and hooks. A dep() inside the
model is not taken for a field:
class AppConfig(Settings):
secrets = dep(SecretStore)
dsn: str = "sqlite://"
password: str = ""
@init
def load_password(self) -> None:
self.password = self.secrets.read("db-password")
An attribute assigned in a hook must be a declared field — that is pydantic's rule for models.
Fields cannot be named init / start / stop; those are the unit's methods.