A data engineering team is tasked with loading a large dataset (5TB) into Snowflake from an external S3 bucket. The data loading process is experiencing significant performance bottlenecks. Which of t

Question

Options

• AIncrease the size of the virtual warehouse to a larger size (e.g., from SMALL to LARGE) before
• BUse multiple virtual warehouses concurrently to load different subsets of the data from S3.
• CUse a larger virtual warehouse indefinitely to handle any potential performance peaks, even after
• DDisable auto-suspend on the virtual warehouse to prevent it from idling during the data load.
• EPartition the data in S3 into smaller files and ensure the virtual warehouse is appropriately sized

• A
  69% (42)
• B
  5% (3)
• C
  10% (6)
• D
  16% (10)

Explanation

Temporarily scaling up the virtual warehouse (A) provides more compute nodes to process files concurrently, directly reducing load time for a single bulk operation. Partitioning the 5TB dataset into smaller files in S3 (E) is equally critical because Snowflake's COPY INTO command parallelizes at the file level - each warehouse node picks up individual files, so one massive file means one thread; hundreds of smaller files (ideally 100–250MB each) means full parallel utilization across all nodes.

Why the distractors fail:

• B (multiple warehouses): Snowflake doesn't natively coordinate multiple warehouses for a single load; orchestrating this manually adds complexity without meaningful gain over one properly sized warehouse.
• C (permanently large warehouse): Keeping an oversized warehouse running indefinitely wastes credits - scale up for the load, then scale back down.
• D (disable auto-suspend): Auto-suspend only governs idle shutdown; the warehouse stays active during an active load regardless, so disabling it has zero performance impact.

Memory tip: Think "small files, big warehouse - temporarily." Snowflake parallelism is file-driven, so shredding data into small chunks is half the battle; the warehouse size determines how many chunks get processed at once. Scale up for the job, then scale back down to save cost.

No community discussion yet for this question.