Skip to main content
The features described in this article are in public preview and may change.
After you register a Snowflake external data source, reference it in transformation SQL with ext_snowflake, then write results to a destination in Cognite Data Fusion (CDF).

Write the SQL

Reference the source by externalId:
For a data model destination, map columns to view properties and include externalId:
Prefer a data model destination. Writing the result to CDF RAW adds the staging copy this path is meant to skip.

Incremental loads with is_new

By default, each run reads the full table. Incremental loads need a TIMESTAMP or epoch-millisecond LONG column that your pipeline updates when a row changes. Snowflake does not create or update that column. You must populate it on every insert and update, then pass it to is_new(). CDF stores the high-water mark; the first run processes all rows.
Passing the table alias as the second argument — is_new('cursor', a) — is rejected for Snowflake. That alias form is for data modeling sources (cdf_nodes, cdf_edges, cdf_data_models). Always pass a column for Snowflake.
Deleted rows are not returned, and CDF does not remove the matching instances. For general is_new patterns, see SQL patterns and best practices.

Create and run the transformation

Create a transformation with a data model destination (example):
Bind session nonces and run the transformation as you do for other Transformations jobs. See Create a transformation.

Next steps

Last modified on November 17, 2026