Review data-source changes
Cassis compares each schema refresh with the previous snapshot and shows which tables, columns, joins, metrics, or virtual tables need attention.
Open the Data source feed under Review. Changes arrive after a warehouse resync or an updated DDL upload.
- Added
- Model a new table or column only when it belongs in the business context. Until then, it stays in the Schema view.
- Renamed or retyped
- Review the proposed reference updates. Cassis preserves descriptions, synonyms, and units where possible.
- Removed
- Check every affected join, metric, and virtual table before accepting the removal.
Approve or dismiss each item. Dismissed items can be restored later. The queue also reads headlessly, from a terminal or by an agent, so CI can see what is waiting: see Source-change commands and Source changes over MCP. Approving and dismissing stay in the web app.
Follow the project’s editing path. Apply the fix in Cassis for an app-managed project. For a git-managed project, make the equivalent repository change so the next import does not overwrite the app edit.
What a DDL upload covers
A DDL file speaks only for the schemas it contains. A partial export, such as Snowflake’s per-schema GET_DDL, adds or updates those schemas and leaves the rest of the source alone. The upload dialog asks which one you have: Update only the schemas in the file, or Replace the complete source schema, where a schema missing from the file counts as dropped. From a pipeline, cassis schema push makes the same choice with --complete.
USE SCHEMA and USE DATABASE statements are honored, so an export whose CREATE TABLE statements carry no schema name still lands in the right schema. Table and column names match case-insensitively, so re-exporting the same schema in another case produces no changes instead of phantom renames and removals.
Cassis rejects a file it cannot trust rather than import half a schema. A truncated or corrupted export fails when it is parsed, and the error says to re-export. When an upload covers only some schemas, one that would remove most of the tables tracked in those schemas is refused with the question “is the file complete?”, instead of filling the feed with removals.
To refresh the source, see Connect a warehouse or upload DDL.