Hello @shan-databricks  !

One additional point, I would also validate the expected load with the SQL Server DBA because even if Lakeflow manages the parallelism internally the source SQL Server still needs to handle those concurrent reads. For 100 tables, I would start with one pipeline/gateway, monitor extraction duration and SQL Server load, then only split into multiple pipelines/gateways if there is a clear operational need such as different refresh SLAs, very large tables, or source side throttling. Don't forget that for tables with primary keys CT is generally preferred over CDC to reduce source overhead.