StreamVid maintains terabytes of raw CSV logs in S3 and needs automated profiling and repeatable transformation (parse timestamps, drop low-cardinality columns, convert to Parquet) before training. Which DataBrew workflow best meets these requirements and what must be ensured for scheduled runs?

The correct option accurately describes the capabilities of AWS Glue DataBrew for data preparation. DataBrew is designed for visual data profiling, cleaning, and normalization, supporting various transformations including parsing timestamps, dropping columns, and converting formats like CSV to Parquet. It allows scheduling jobs for repeatable transformations. Crucially, the DataBrew job execution role needs appropriate S3 permissions to read input and write output, and network connectivity (VPC configuration) must be correct. The incorrect options propose less suitable or inaccurate approaches: Launching an AWS Glue ETL job instead of DataBrew is incorrect because DataBrew can schedule recurring jobs and write Parquet, making it a more direct solution for this use case. Using IAM user credentials stored in the job is an anti-pattern for security. SageMaker Data Wrangler is a viable tool for data preparation, but the statement that DataBrew cannot produce Parquet outputs or scheduled runs is false. Creating a Lambda function for terabytes of data is inefficient and complex for profiling and transformation compared to DataBrew, which is purpose-built for this scale and task.