1Z0-184-25 Question 174
Single answerYour data science team has generated a large set of numeric embeddings stored in CSV files for an advanced recommendation engine. Each vector has 128 floating-point components. You want to load these vectors into a new table in an Oracle 23c Database running on Oracle Cloud Infrastructure (OCI). The table schema includes a VECTOR(128) column to store embeddings. Which SQL*Loader approach ensures accurate numeric data loading while achieving high throughput and minimizing downtime for your production application?
- A
Use a conventional path load with ROWS=1 in the control file to ensure row-by-row processing, preserving accuracy of float values.
- B
Leverage direct path loading with PARALLEL=TRUE and numeric external data types that match the VECTOR(128) column, then rebuild indexes after the load.
- C
Convert the CSV files into a single JSON file and use SQL*Loader in stream mode for faster parsing of multi-row data.
- D
Use the DBMS_CLOUD package to directly query the CSV files over Object Storage, bypassing SQL*Loader entirely.
Show answer and explanation
Correct answer: B
Explanation
For large-scale vector data, direct path loads in SQL*Loader can significantly improve performance by writing directly above the High Water Mark (HWM). Specifying numeric external types (such as FLOAT EXTERNAL) that match the VECTOR(128) column prevents precision loss. Enabling parallel load (PARALLEL=TRUE) in the control file allows multiple concurrent streams, further reducing total load time. Oracle� documentation recommends rebuilding indexes after a direct path load to avoid load contention (see Oracle Database Utilities Guide for best practices).
- A. Incorrect.
Option 1: While row-by-row processing can be useful in certain scenarios, it significantly reduces load performance and is typically not necessary for preserving numeric precision. Conventional path loads are slower, especially with large datasets.
- B. Correct.
Option 2: This is the correct answer. Direct path loading with PARALLEL=TRUE and specifying numeric external data types matching the VECTOR(128) column improves throughput and accuracy. Rebuilding indexes after the load also avoids contention during the bulk load process.
- C. Incorrect.
Option 3: SQLLoader does not inherently parse JSON for vector data. Converting everything to JSON would add extra overhead and complexity, and there is no built-in stream mode in SQLLoader specifically aimed at JSON vector data loading.
- D. Incorrect.
Option 4: The DBMS_CLOUD package is a viable alternative for some loading scenarios, but this question specifically focuses on SQLLoader usage. Additionally, DBMS_CLOUD may not address advanced tuning needs (such as parallel direct path loading) for large-volume vector data as effectively as SQLLoader does.