Registry, database, and collection¶

The connection hierarchy mirrors MongoDB:

Registry (one AsyncMongoClient)
└── Database (one native AsyncDatabase view)
    └── Collection (one native AsyncCollection view)

from mongoz import Collection, Database, Registry

registry = Registry("mongodb://localhost:27017")
database: Database = registry.get_database("app")
collection: Collection = database.get_collection("users")

await collection.driver.create_index("email", unique=True)

registry.driver, database.driver, and collection.driver return typed native PyMongo Async objects. They are escape hatches, not ownership transfers: Registry.close() remains the only client cleanup owner.

Multiple databases¶

A document's Meta.database selects its default database. Derive an isolated manager for another database with using():

eu_users = User.objects.using("tenant_eu")
us_users = User.objects.using("tenant_us")

await eu_users.create(name="European user")
await us_users.create(name="US user")

Because managers use clone-on-write state, neither derivation changes User.objects or the other manager. The application still owns tenant authorization; never choose a database directly from an unvalidated request value.

Inspection¶

await registry.get_databases() and await database.get_collections() expose server state through wrappers. await registry.address reports the selected server address when PyMongo has one. These operations can perform network I/O and therefore belong inside the active lifecycle.