You are tasked with loading data from a series of CSV files stored in an Amazon S3 bucket into Snowflake. The CSV files contain a header row, but some files have slight variations in the number and or

Question

Options

• ACreate a separate external table for each CSV file with a different column structure.
• BDefine a single external table with a VARIANT column and use Snowflake's CSV parsing
• CCreate a single target table with all possible columns from all CSV files, using 'SKIP_HEADER = 1'
• DCreate a VIEW on top of the external table to ensure that column names are consistent across all
• EPre-process the CSV files to standardize the column order and names before loading them into

• A
  4% (1)
• B
  19% (5)
• C
  65% (17)
• D
  4% (1)
• E
  8% (2)

Explanation

Option C is correct because creating a single target table that includes all possible columns across every file variant, combined with SKIP_HEADER = 1, lets Snowflake's COPY INTO command skip the header row and map data using column-name matching (MATCH_BY_COLUMN_NAME = CASE_INSENSITIVE) - handling different column orders gracefully while extra columns are ignored via ERROR_ON_COLUMN_COUNT_MISMATCH = FALSE.

Why the distractors fail:

• A - Creating a separate external table per file is operationally unscalable and defeats the goal of unified loading.
• B - Storing CSV data in a VARIANT column introduces unnecessary complexity; Snowflake is already purpose-built for structured CSV ingestion without needing semi-structured workarounds.
• D - A view reshapes how you query data but does nothing to fix the underlying ingestion problem of mismatched columns during the load itself.
• E - Pre-processing adds an extra pipeline stage and operational burden; the correct answer solves the problem natively within Snowflake.

Memory tip: Think of option C as the "union table" strategy - like a SQL UNION ALL that needs a superset of all columns to accommodate every participant. If the table has every possible column and you skip the header row, Snowflake can match what it finds and NULL-fill what it doesn't, keeping the load simple and the pipeline single-stage.

No community discussion yet for this question.