ExamPark
Course menuMLA-C02

Exams / MLA-C02 / Study guide / Transforming data with Glue, EMR and Data Wrangler

Transforming data with Glue, EMR and Data Wrangler: MLA-C02 study notes and practice questions

What to know about Transforming data with Glue, EMR and Data Wrangler for the MLA-C02 exam, from the Data Preparation for ML and AI domain (28% of the exam guide), then 2 free questions to test it. Part of the MLA-C02 study guide.

Last verified against current AWS documentation ·How we write our questions

What to know for the exam

Transformation questions describe raw data and the people preparing it, and ask which service reshapes it with least effort.

Choosing a tool

Tool Best for
SageMaker Data Wrangler (in SageMaker Canvas) Visual preparation of ML data with built-in transforms, data quality and insights reports, and export to pipelines or Feature Store
AWS Glue DataBrew Analysts who don’t code: profile data, build reusable recipes, run them as scheduled jobs
AWS Glue ETL Serverless Spark or Python jobs for large, repeatable transformations; Glue Studio for visual authoring
Amazon EMR / EMR Serverless Very large Spark, Hive or Presto workloads, or custom frameworks
SageMaker Processing Running your own preprocessing or evaluation scripts (scikit-learn, Spark, custom containers) as managed jobs, often as a pipeline step
Athena SQL reshaping of data in S3, such as CTAS to Parquet

Glue features to know

  • Job bookmarks track what earlier runs processed, so daily jobs read only new data.
  • Push-down predicates read only the partitions you name.
  • DynamicFrames handle messy schemas; resolveChoice fixes mixed types.
  • Flex execution lowers cost for jobs that can wait.
  • Glue Data Quality checks rules inside the job.

Common transformations for ML

  • Join sources, deduplicate, fix types and handle missing values.
  • Convert to columnar formats and partition by date.
  • Split into training, validation and test sets before fitting scalers or encoders.
  • Write curated outputs to S3 or Feature Store.

Choosing by the question’s clues

  • “No code”, “analysts”, “reusable recipe”: DataBrew.
  • “Data scientists preparing features visually and exporting to a pipeline”: Data Wrangler.
  • “Petabytes, existing Spark code”: EMR or Glue.
  • “Run our preprocessing script as a step”: SageMaker Processing.

Common traps

  • Reprocessing the full history every night when bookmarks or predicates limit the job to new data.
  • Spark code for analysts who want a visual tool.
  • Running preprocessing on the training instance when a Processing step separates and reuses it.

Try the 2 practice questions

Practice Transforming data with Glue, EMR and Data WranglerTake the free practice exam

Question 1

Harder · choose 2 ·Link to this question (question 1)

A regional bank loads daily transaction files into Amazon S3, partitioned by business date. A nightly AWS Glue Spark job reads the raw transactions, builds the training table, and writes it to the curated bucket.

The job currently reprocesses all files since launch, which takes most of the night and grows each month. The team must process only the files and partitions that are new since the last successful run, without changing the source files or the bank's daily upload process.

Which TWO changes meet these requirements? (Choose two.)

Choose two answers.

  1. A. Set the job to overwrite the entire curated table on every run, so the output is always rebuilt from the full history.
  2. B. Enable job bookmarks on the Glue job so that Glue tracks which input data it has already processed in earlier runs.
  3. C. Add more DPUs or workers to the job so that the full history finishes faster each night.
  4. D. Use a push-down predicate that limits the read to the business-date partitions that are newer than the last successful run.
  5. E. Run an AWS Glue Data Quality ruleset on the raw transactions before the job starts so that bad rows are removed from the input.
Show the answer and explanation

Answers: B, D

  • A. Wrong: A full rebuild is the cause of the slowdown, not the fix. The job would still read and write all the history each night.
  • B. Right: Correct. Job bookmarks record processed S3 objects between runs, so each run reads only new input. This needs no change to the source files or the upload process. Learn more: Tracking processed data with job bookmarks (opens in a new tab)
  • C. Wrong: More workers shorten the run but still reprocess everything, so the job keeps growing with each month of data. The requirement asks for incremental processing.
  • D. Right: Correct. A push-down predicate lets Glue read only the partitions that match a filter, so the job skips older dates. Combined with bookmarks, the job reads only new partitions and files. Learn more: Managing partitions for ETL output in AWS Glue (opens in a new tab)
  • E. Wrong: Data quality rules check or filter rows, but they do not limit which input the job reads. The job would still process all files on every run.

Explanation

Job bookmarks and push-down predicates both limit a Glue job to new input: bookmarks track what earlier runs processed, and predicates filter the partitions read. Neither changes the source data or the daily upload. Overwriting the table or adding workers still reprocesses the whole history.

Study

Question 2

Easier ·Link to this question (question 2)

A logistics broker receives carrier invoices as CSV files from about 300 carriers. The files have inconsistent date formats, duplicated carrier names and missing currency codes, and the analytics team needs them cleaned before modeling.

The analysts do not write code, but they need to profile each dataset, apply cleaning steps visually and save the steps so they can rerun them each week.

Which AWS service best meets these requirements?

Choose one answer.

  1. A. AWS Glue Spark ETL jobs written in Python, with each cleaning step coded by a data engineer and scheduled to run weekly with an AWS Glue trigger.
  2. B. Amazon EMR with a Hive cluster that analysts access through SQL notebooks they run themselves.
  3. C. Amazon Athena with CREATE TABLE AS SELECT statements that the analysts write by hand for each cleaning step.
  4. D. AWS Glue DataBrew, which profiles the datasets and lets analysts build recipes of visual cleaning steps that run as scheduled jobs.
Show the answer and explanation

Answer: D

  • A. Wrong: Glue is capable, but the steps would need code written by data engineers, which the analysts cannot maintain. The requirement is for analysts to work visually without code.
  • B. Wrong: Hive and SQL still require writing code, and the broker would manage a cluster for a task a managed visual service handles. The option does not meet the no-code requirement.
  • C. Wrong: Athena runs SQL, so the analysts would need to write statements, and profiling and visual steps are not part of the workflow. This option misses the no-code requirement.
  • D. Right: Correct. DataBrew provides a visual interface with profiling and many built-in transformations, and recipes can run as scheduled jobs. The analysts can rerun the saved steps each week without writing code. Learn more: What is AWS Glue DataBrew? (opens in a new tab)

Explanation

AWS Glue DataBrew lets analysts profile data and build reusable visual recipes that run as scheduled jobs, with no code. Spark jobs, EMR and Athena SQL all require code that the analysts do not write.

Study

Want more on Transforming data with Glue, EMR and Data Wrangler? The 5 full-length practice sets have 8 more Transforming data with Glue, EMR and Data Wrangler questions, 325 in all, none of them in the free exam. One payment of $14.99, no subscription. See the practice sets

Study Transforming data with Glue, EMR and Data Wrangler

More Data Preparation for ML and AI topics