COF-C03 Question 60
Single answerSnowpark OptimizedA data science team runs a Snowpark Python pipeline that performs feature engineering and in-database model inference on hundreds of millions of rows. The workload is CPU- and memory-intensive, and it runs much slower than expected on a standard warehouse even after increasing the warehouse size. The architect wants to improve performance without moving data out of Snowflake or redesigning the code into standard SQL. Which option is the best recommendation?
- A
Use a Snowpark-optimized warehouse for the Snowpark Python workload because it provides more memory per node for resource-intensive Snowpark processing.
- B
Convert the Snowpark Python pipeline into external Spark jobs because Snowflake warehouses are intended only for SQL workloads.
- C
Replace the warehouse with a serverless task because serverless compute is designed to accelerate long-running Snowpark Python jobs.
- D
Enable Search Optimization Service on the input tables because it is the primary performance feature for CPU- and memory-bound Snowpark execution.
Show answer and explanation
Correct answer: A
Explanation
Snowpark-optimized warehouses are intended for resource-intensive Snowpark workloads, especially where additional memory per node helps performance. This makes them a strong fit for Python-based feature engineering, model scoring, and similar operations that run inside Snowflake but are not easily rewritten as standard SQL. In contrast, Search Optimization Service targets specific query-pruning patterns rather than general Snowpark compute bottlenecks, and serverless tasks address orchestration rather than warehouse specialization. Snowflake documentation describes Snowpark-optimized warehouses as a specialized warehouse type for memory-intensive Snowpark processing and recommends them when standard warehouses are not sufficient for such workloads.
- A. Correct.
Correct. Snowpark-optimized warehouses are designed for workloads such as Snowpark Python that need more memory and compute efficiency for resource-intensive processing, including data engineering, machine learning, and similar in-database workloads. In this scenario, the team wants to keep processing inside Snowflake and avoid redesigning the code into SQL, so a Snowpark-optimized warehouse is the most appropriate recommendation.
- B. Incorrect.
Incorrect. Snowflake supports Snowpark workloads directly, and Snowpark-optimized warehouses exist specifically to improve performance for memory-intensive Snowpark processing. Moving the workload to external Spark contradicts the requirement to keep data in Snowflake and is not the best recommendation when Snowflake provides a native option.
- C. Incorrect.
Incorrect. Serverless tasks are used to run scheduled task workloads without managing warehouse infrastructure, but they are not the main mechanism for accelerating heavy Snowpark Python jobs. They do not replace the need to choose the right warehouse type for resource-intensive execution.
- D. Incorrect.
Incorrect. Search Optimization Service can improve the performance of selective point-lookups and certain query access patterns, but it does not address the core issue here: a CPU- and memory-intensive Snowpark Python pipeline. It is not the primary tuning feature for this type of workload.