Skip to content

DuckLake

A DuckLake is a catalog database with Parquet files in storage. Depositori reads the catalog, read-only: tables, views and their SQL, columns and comments.

  • The Owner role in Depositori.
  • From whoever runs the lake: the catalog (a .ducklake file, or a Postgres database as postgres:dbname=… host=…), the data path (for example az://container/folder) and, for data on Azure Storage, the storage account name.
  • A source account for the catalog: a stored database login, or on Azure a Microsoft account that is a user of the Postgres server.
  • For data on Azure Storage: the role Storage Blob Data Reader on the container, for the Microsoft account that reads it.
  1. Open Connections, choose Add connection, then DuckLake.
  2. Under Reads with, keep your own Microsoft account, or choose Add a login for a database login.
  3. Fill in the Catalog and a Name.
  4. Under More settings, fill in the Data path and the Storage account. The other fields there are only needed when your lake differs from the default.
  5. Test, then Save connection.
Connect DuckLake, filled in for AcmeConnect DuckLake, filled in for Acme

On the connection’s card, choose Add as a model. Pick the schemas you want and choose Import. The tables become a model.

The card of the DuckLake connectionThe card of the DuckLake connection

Taki may query is off. Read Taki, the AI analyst before you turn it on.

  • A catalog is never a web address. It is a .ducklake file, or a database: postgres:…, mysql:… or sqlite:….
  • Taki queries the lake’s tables only, never files or web addresses.

Depositori in the cloud is stricter:

  • The catalog is in PostgreSQL (or a .ducklake file). A MySQL catalog is refused.
  • The data is on Azure Storage, in the storage account you name on the connection: az://… or abfss://…. Data behind a web address (http:// or https://) is refused. Your Microsoft account’s access goes only to that storage account, never to another host.