Hi @Emmanuel Convers
Thank you for providing additional details. Your patience and understanding are truly appreciated.
Here's a clear, step-by-step analysis and resolution plan based on the latest documentation, community feedback, and system updates.
Confirm Runtime Environment Compatibility
- Oracle Connector v2.0 requires TLS 1.3, which is only supported on Windows Server 2022 or Windows 11.
- Ensure your SHIR VMs are running a TLS 1.3–capable OS.
- If TLS 1.3 is not listed under protocols, you may need to enable it via registry edits. [How to get...rosoft Q&A]
Optimize CLOB Handling
- CLOB fields are known to cause performance degradation due to their size and streaming behavior.
- Oracle Connector v2.0 introduces stricter parsing and mapping logic, which may affect how CLOBs are processed. [Azure Data...rosoft Q&A]
- Avoid SELECT statements with semicolons and ensure proper parameterization in queries.
Use Staging Techniques
- Instead of direct copy, use a staging table or intermediate format (e.g., CSV or Avro) before converting to Parquet.
- This can reduce transformation overhead and improve throughput.
Use Custom Query with SUBSTR
- If data size permits, use Oracle SQL DBMS_LOB.SUBSTR function to retrieve chunks of CLOB as strings.
- Example SQL query in ADF Copy Activity source:
SELECT EXAMPLE_ID, DBMS_LOB.SUBSTR(EXAMPLE_CLOB, 4000, 1) AS EXAMPLE_CLOB_PART FROM EXAMPLE_TABLE
Note: This will read the first 4000 characters. You can split the CLOB into multiple parts by running multiple copy activities with different offsets.
Tune Integration Runtime Settings
- Increase batch size and parallel copy threads in the Copy Activity settings.
- Use compression where possible to reduce payload size.
Latest Updates & Known Issues
- Release Notes show multiple bug fixes and enhancements to Parquet and ORC formats, but no direct fix for CLOB performance yet. [Release Notes (10)]
- The issue is acknowledged by Microsoft staff and is under review, but no ETA for a fix is available. [Azure Data...rosoft Q&A]
For mitigation workaround:
- Use staged export (Data Pump or external table).
- Use DBMS_LOB.SUBSTR to split large CLOBs into manageable strings if feasible.
- Continue using v1.0 for CLOB-heavy tables until October 31, 2025, if feasible.
- For v2.0:
- Use staging and transformation strategies.
Thanks,
Vrishabh