feat: Add Oracle 26ai as default vector database

[ssraghavan-oci]

Summary

Replace Milvus with Oracle AI Database 26ai as the default vector store for the NVIDIA RAG Blueprint. This provides CPU-optimized IVF indexes with optional hybrid search support.

Changes

• Add OracleVDB class implementing VDBRagIngest interface
• Add oracle_queries.py with DDL and SQL query helpers
• Register Oracle as default in vdb/__init__.py (Milvus remains as fallback)
• Add oracledb dependency to pyproject.toml
• Add Oracle environment variables to deploy/compose/.env
• Add oracle-26ai Docker service profile to vectordb.yaml
• Add comprehensive migration guide in docs/migration-oracle-26ai.md

Key Features

• CPU-based IVF indexes - Optimized for deployments without GPU
• Hybrid search - Combined vector similarity with Oracle Text full-text search
• LangChain OracleVS integration - For retrieval operations
• Connection pooling - Via oracledb thin client (no Oracle Client installation needed)
• Environment-based configuration - For credentials and index settings

Configuration

export APP_VECTORSTORE_NAME=oracle
export ORACLE_USER=rag_user
export ORACLE_PASSWORD=your_password
export ORACLE_DSN=localhost:1521/FREEPDB1

Testing

1. Install dependencies: pip install nvidia_rag[oracle]
2. Set Oracle credentials
3. Deploy: docker compose -f deploy/compose/docker-compose-rag-server.yaml up -d --build
4. Health check: curl http://localhost:8081/v1/health?check_dependencies=true

[smasurekar]

Kindly add "oracle" profile for minio container as well. As it is used internally for Image/Table/charts/etc. citations

[smasurekar]

Kindly update the uv.lock file based on the library updates

[smasurekar]

Kindly add:

    "oracledb>=2.0.0",
    "langchain-community>=0.4",

in common dependencies as well. So they could be used installed in ingestor-server and rag-server containers

[smasurekar]

The code structure looks great and aligns well with the requirements! 💯

I did run into an issue during ingestion that I wanted to bring to your attention:

Error in src/nvidia_rag/utils/vdb/oracle/oracle_vdb.py:617:

ERROR:nvidia_rag.ingestor_server.main:Ingestion failed due to error: the JSON object must be str, bytes or bytearray, not list
Traceback (most recent call last):
  File "/workspace/.venv/lib/python3.13/site-packages/nvidia_rag/ingestor_server/main.py", line 480, in __run_background_ingest_task
    metadata_schema = vdb_op.get_metadata_schema(collection_name)
  File "/workspace/.venv/lib/python3.13/site-packages/nvidia_rag/utils/vdb/oracle/oracle_vdb.py", line 617, in get_metadata_schema
    return json.loads(row[0])
           ~~~~~~~~~~^^^^^^^^
  File "/usr/lib/python3.13/json/__init__.py", line 345, in loads
    raise TypeError(f'the JSON object must be str, bytes or bytearray, '
                    f'not {s.__class__.__name__}')
TypeError: the JSON object must be str, bytes or bytearray, not list

For testing:

• Kindly add Oracle test sequence in tests/integration/test_sequences.yaml (similar to the Elasticsearch one at lines 37-42)
• Integration tests can be run with: pip install -r tests/integration/requirements.txt and python -m tests.integration.main --sequence oracle

Let me know if you need any additional details. Thanks for your work on this integration!