Compacting small Parquet files into larger ones is a transformation, not a metadata operation. The AWS-native pattern is a Glue ETL Spark job that reads the source prefix, repartitions or coalesces to a target file size (typically 128–512 MB), and writes the compacted Parquet output back to a separate compacted prefix (or to a separate compacted table). The Glue Workflow can run nightly, after the producer's writes complete, and a Glue Data Quality rule can validate the compacted output before the SageMaker training job consumes it.

Crawlers do not compact data — they only update Catalog metadata. S3 lifecycle policies only transition objects between storage classes or expire them; they do not merge files. Athena CTAS can produce a compacted output but its semantics for in-place rewriting are awkward and risk colliding with the producer's append pattern; the canonical answer is the dedicated Glue ETL job. Apache Iceberg tables provide built-in compaction primitives and are an emerging alternative, but Glue ETL is the simplest exam-default pattern. The MLA-C01 stem signal 'compact many small Parquet files into fewer larger ones' is the Glue ETL Spark signature.

Plain-Language Explanation:

Compacting small files is like a junk-drawer cleanup — you collect 1,000 loose paperclips, batch them into bundles of 100, and put the bundles back. The cleanup is real work (Glue ETL Spark job). It is not a label change (Crawler) or a moving company that sorts by storage temperature (lifecycle policy).

Option Analysis:1523

A：Wrong because Crawlers do not move or compact data; they update Catalog metadata.
B：Wrong because S3 lifecycle policies only transition or delete objects; they do not merge files.
C：Wrong because Athena CTAS into the same source prefix conflicts with the producer pattern; Glue ETL into a separate compacted prefix is the canonical answer.
D：Correct because a Glue ETL Spark job is the AWS-native pattern for compacting small Parquet files into larger ones.