A company ingests application logs into Amazon S3 through Amazon Kinesis Data Firehose. The logs arrive as newline-delimited JSON, and analysts query them with Amazon Athena. Query performance is poor because the JSON files are small and uncompressed. The engineer must improve Athena query performance while keeping the raw JSON available for a downstream legacy system. Which change should the engineer make?
Converting to Parquet with a Firehose record format conversion gives Athena a columnar, compressed format that scans far less data, directly improving query performance. Delivering the raw JSON to a separate prefix through a second stream preserves the legacy feed, satisfying both requirements without duplicating logic in the consumers.
Why this answer
Athena performance improves when it scans less data, and converting the stream to Parquet with a record format conversion makes the stored objects columnar and compressed. A second delivery stream that leaves the raw JSON in place satisfies the legacy consumer, so the engineer gets both performance and compatibility.
Exam trap
The trap here is assuming buffer tuning or query-side filters can substitute for changing the storage format, when only a columnar, compressed layout reduces the bytes Athena actually reads.