My preference is option 1 

• Delta Sharing is the most efficient and secure integration between Databricks and external systems.

• No JDBC bottlenecks (no long-running queries, no network saturation).

• Data shared as Delta format, which is natively optimized for Databricks.

• Lower operational overhead — Databricks reads the Delta Shares directly.

• Good for large volumes (Finance, SCM, HCM typically generate big datasets).

• Strong governance and lineage support.

Also I don't like to use JDBC, I avoid using it unless there are no other options

• Not scalable for large Oracle Fusion workloads.

• JDBC pulls are:

  • slow

  • stateful

  • prone to timeouts

  • difficult to parallelize

  • expensive for large history loads

• High latency for production-grade pipelines.

• You must manage incremental logic manually (ROWIDs, timestamps, etc.).