A. Incorrect.
EMR adds compute cost and extra pipeline steps.

B. Incorrect.
Aurora and Glue introduce unnecessary services and cost.

C. Incorrect.
Glue copying files adds job cost and does not address the best Redshift loading pattern.

D. Correct.
A manifest file lets Redshift load files from multiple S3 locations with a single COPY operation, improving parallel ingestion without new compute services.

A. Correct.
QuickSight needs access to the S3 data location that Athena reads, typically through the QuickSight service role and bucket policy permissions.

B. Correct.
If the S3 objects are encrypted with a customer managed KMS key, QuickSight also needs KMS decrypt permissions through the key policy and IAM permissions.

C. Incorrect.
S3 buckets are not normally shared for Athena and QuickSight access by AWS RAM.

D. Incorrect.
Redshift is not required for QuickSight to query Athena datasets.

E. Incorrect.
Public access would be insecure and is not required for cross-account QuickSight access.

A. Incorrect.
Firehose data format conversion expects JSON input with a schema; CSV usually needs transformation first.

B. Correct.
A lightweight Lambda transform can convert CSV to JSON, and Firehose can then use schema-based format conversion to deliver Parquet files.

C. Incorrect.
EMR adds cluster management and batch delay.

D. Incorrect.
Custom Parquet producer logic increases development effort and gives up Firehose's managed delivery features.

A. Incorrect.
EKS adds cluster operations overhead for small daily files.

B. Incorrect.
A provisioned EMR cluster is excessive for files under 100 MB.

C. Incorrect.
Spark is useful for distributed processing, but it adds unnecessary overhead for small files.

D. Correct.
Glue Python shell jobs are cost-effective for small-scale Python transformations and can use libraries such as pandas.

A. Correct.
Lambda can start Athena queries programmatically without managing servers.

B. Correct.
Step Functions can wait, poll query status, branch on success or failure, and coordinate the next query without paying for idle compute.

C. Incorrect.
A Glue job that sleeps while waiting wastes ETL job runtime.

D. Incorrect.
MWAA can orchestrate workflows, but it is more operationally heavy and costly for a small sequence of Athena queries.

E. Incorrect.
EC2 polling requires server management and pays for idle time.

A. Correct.
S3 Standard supports immediate frequent access for new records, and Standard-IA reduces storage cost for infrequently accessed records that still need quick retrieval.

B. Incorrect.
Standard-IA has retrieval charges and a minimum storage duration, so it is not ideal for frequently accessed new records.

C. Incorrect.
Glacier Deep Archive is for rarely accessed archive data and does not provide immediate access.

D. Incorrect.
Keeping all records in Standard misses the cost optimization opportunity after access drops.

A. Incorrect.
Regular expressions are not sufficient for probabilistic matching across records without a common key.

B. Incorrect.
Filtering selects rows that match explicit criteria; it does not learn duplicate or related record patterns.

C. Incorrect.
The scenario says no shared unique identifier exists.

D. Correct.
FindMatches uses machine learning to identify matching records and can be used in Glue-based ETL workflows.

A. Correct.
A Glue crawler can discover schema and partition changes and update the Data Catalog automatically on a schedule.

B. Incorrect.
Manual partition maintenance creates ongoing operational overhead.

C. Incorrect.
S3 Inventory reports object metadata; it does not update Glue table partitions by itself.

D. Incorrect.
Removing partitions would reduce query efficiency and does not solve catalog synchronization cleanly.

A. Correct.
The target table is accounts. WHEN MATCHED THEN DELETE removes target rows where the join condition matches source rows.

B. Incorrect.
This command deletes matching target rows; it does not keep only matching rows.

C. Incorrect.
The source table is not dropped or deleted by the MERGE statement.

D. Incorrect.
MERGE statements for supported table formats can include delete actions.

A. Incorrect.
IAM roles do not directly replace credentials for a typical on-premises MySQL database.

B. Incorrect.
KMS key rotation is not the same as database credential rotation, and DynamoDB would require custom rotation logic.

C. Correct.
Secrets Manager stores credentials securely, supports automatic rotation workflows, and lets applications retrieve the current secret at runtime.

D. Incorrect.
S3 Lifecycle manages object storage transitions and expiration; it does not rotate database credentials.