Supabase Pipelines is a managed change data capture (CDC) product that uses Postgres logical replication to deliver published data to analytical destinations in near real time.
Key benefits
- Initial sync: Copy existing rows from published tables.
- Ongoing replication: Capture and deliver subsequent INSERT, UPDATE, DELETE, and TRUNCATE operations selected by the publication.
- Managed operation: Monitor pipeline status, lag, table state, and errors in the Dashboard.
- Workload isolation: Keep analytical queries away from the primary database.
Destinations
BigQuery is currently available. ClickHouse, DuckLake, and Snowflake are in Early Access. Request access to these destinations.
Setup
Create a Postgres publication for the tables to replicate. In Database > Replication, add a Pipelines destination, configure its settings, and monitor the pipeline from the Dashboard.
Requirements
Requirements depend on the destination. BigQuery and ClickHouse ReplacingMergeTree require source tables to have primary keys and require the publication to include those columns. ClickHouse updates require REPLICA IDENTITY FULL. ClickHouse deletes require primary-key or full identity. DuckLake updates and deletes require a primary-key identity, replica-identity index, or full identity. With a primary-key identity or replica-identity index, include every identity column in the publication. Snowflake updates require REPLICA IDENTITY FULL. Snowflake deletes require a published row identity.
Pipelines is valuable for:
- Near real-time analytics data movement
- Analytics separation from production
- Managed replication to supported destination systems
Limitations
Schema change support is destination-specific and limited. Destination-specific constraints apply.
BigQuery and DuckLake keep current-state tables synchronized. ClickHouse supports current-state ReplacingMergeTree tables or append-only MergeTree CDC history. Snowflake stores append-only CDC history. Source TRUNCATE operations and table resets erase accumulated destination data. Schema changes can alter retained history in append-only ClickHouse and Snowflake tables.