- Use a Self-Hosted Integration Runtime or a securely configured Azure Integration Runtime to connect to Oracle on-prem.
- Copy Activity suffices for initial loading and straightforward upserts. Combine it with a Data Flow (or a separate pipeline step) for deletes if needed.
- Implement a watermark strategy to handle deltas (UPSERT/DELETE), leveraging the
LAST_UPDATE
field or equivalent. - Rely on ADF’s built-in error handling and retries via activity settings and pipeline branching.
- Scale Cosmos DB RUs appropriately during bulk loads to avoid throttling and tune parallelism.
This setup should cover your pure ETL scenario for Oracle-to-Cosmos DB synchronization without requiring an overly complex design. Let me know if you need detailed steps on any of these.