Starts at 1:22:18 in this video — SQL Warehouse full course — streaming tables (1:22:18)
Match method to the problem
- 2 MB CSV on a laptop → Workspace UI upload
- New files landing forever in S3/ADLS/GCS → Auto Loader (often via streaming table /
read_files) - Live table access for another org, no CSV emails → Delta Sharing
- Programmatic load from an app → API / partner connector
- Browse/install a published data product → Marketplace
Auto Loader
Auto Loader incrementally ingests new files in cloud object storage and tracks what was already processed. Formats include JSON, CSV, Parquet. In Databricks SQL you often wrap this as a streaming table using read_files. Analysts may not write the Python API on the exam, but they must recognize when Auto Loader is the right story.
Delta Sharing
Open protocol to share live UC tables with other Databricks accounts or open recipients without copying a weekly CSV. Providers manage the share; recipients get a foreign/shared catalog. Access remains governed.
S3 / cloud paths
Production cloud ingest uses a UC external location (storage credential + path), not a random unmanaged URL that bypasses grants.