MLA-C01 Practice Question 72: Glue Job Bookmarks
AWS Certified Machine Learning Engineer – Associate (MLA-C01) · Test 8 · question 2 of 10
A nightly AWS Glue ETL job turns raw JSON clickstream files into Parquet features for a recommendation model. Each day upstream systems add new files under s3://raw/clicks/ and never modify old ones. The job reads the whole prefix every night, so its runtime and cost grow daily, and events from earlier days are written to the output again, creating duplicate training rows.
What should the engineer change so that each run processes only files it has not processed before?
Show answer and explanation
Answer:
AWS Glue job bookmarks record state about data a job has already processed. On a scheduled rerun, the job processes only the new data and does not reprocess old data. Because the upstream systems only ever add new files, bookmarks turn the nightly job into an incremental one. That fixes both the growing runtime and the duplicate training rows (option D).
- A is wrong because a crawler updates *schema and partition* metadata in the Data Catalog. The catalog table still covers every file, so the job would read everything again. - B is wrong because moving data to an archive tier to hide it from a job is fragile and destructive: reading archived objects fails until they are restored, and the raw history would no longer be ready for rebuilding features. - C is wrong because more workers make the full scan faster but still reprocess every file and still write duplicate rows, at a higher cost per run.
Reference: Tracking processed data using job bookmarks