DEA-C01 Study Guide: Follow the Data Through the Pipeline, One Decision at Each Stage
The AWS Certified Data Engineer – Associate exam (DEA-C01) reads like a series of pipeline reviews. Data arrives somewhere, is changed, stored, queried, monitored and protected, and at each stage the question asks which AWS service or setting fits. The best way to study it is to follow that path. This guide walks the pipeline stage by stage, with the decision each stage asks for and the mistakes that cost points.
Study guide · AWS Certified Data Engineer – Associate (DEA-C01)

The exam
65 questions (50 scored, 15 unscored) in 130 minutes; passing score 720 of 1,000. Domains: Data Ingestion and Transformation (34%), Data Store Management (26%), Data Operations and Support (22%), Data Security and Governance (18%).
In the 395 DEA-C01 questions we wrote for this site, four services dominate: Amazon S3, AWS Glue, Amazon Redshift and Amazon Athena, in that order. Together they appear in more than half of the questions (57%). Know those four deeply before you spread out to EMR, Kinesis, Lake Formation and Step Functions.
Stage 1: getting data in
The decision is batch or streaming, and how much code you want to own.
- Files dropped on a schedule → S3, often with an event (S3 Event Notifications or EventBridge) to start processing.
- Databases → AWS DMS for migration and change data capture; zero-ETL integrations into Redshift where available.
- Continuous events → Kinesis Data Streams or Amazon MSK when several consumers need the stream; Amazon Data Firehose when the job is only "deliver to S3 or Redshift". (Our Kinesis versus MSK comparison covers the clues.)
- SaaS sources → Amazon AppFlow.
Stage 2: transforming it
| Engine | Choose it when |
|---|---|
| AWS Glue (Spark jobs) | Serverless ETL at scale with the Data Catalog; job bookmarks to process only new data. |
| Amazon EMR | You need control over the cluster or open-source tools beyond Spark (Hive, Presto, HBase), or very large steady workloads. |
| AWS Lambda | Small, quick transformations per file or event (it stops after 15 minutes). |
| Amazon Redshift (SQL) | The data is already in the warehouse: transform it there (ELT) with SQL and stored procedures. |
| AWS Glue DataBrew | Analysts must clean data visually, without writing code. |
Orchestration is its own decision: Step Functions for AWS-native workflows with branching and retries, Amazon MWAA when the team already writes Airflow DAGs, Glue workflows for pipelines made only of Glue jobs and crawlers, EventBridge for event-driven or scheduled starts.
Stage 3: storing it
S3 is the data lake; the decisions are about layout. Store analytical data in a columnar format (Parquet or ORC), compressed, partitioned by the column you filter on most (usually date), and in files large enough to be efficient (every file adds listing and request overhead). Thousands of tiny files are a classic cause of slow Glue jobs and Athena queries.
Redshift is the warehouse: provisioned clusters or Redshift Serverless, COPY from S3 for loading, Redshift Spectrum to query S3 data without loading it, materialised views for repeated aggregations. DynamoDB appears for key-value access at any scale, with its partition key design.
Going deeper: why Parquet and partitions cut Athena's bill
Athena charges by the amount of data each query scans. Partitioning lets Athena skip whole S3 prefixes that the WHERE clause rules out (with partition projection it does not even need to look them up in the catalog). Parquet then reduces what is read inside the files that remain: it stores each column separately, so a query that needs three of forty columns reads only those three, and it keeps minimum and maximum values for each block of rows, so blocks that cannot match a filter are skipped without being read. Compression shrinks what is left. The same query over the same data can scan a small fraction of the bytes it scans over uncompressed CSV, which is why "convert to Parquet and partition by date" is so often the correct answer to both "too slow" and "too expensive".
Going deeper: Redshift distribution styles and skew
Redshift spreads each table's rows across the slices of its compute nodes. KEY distribution puts rows with the same key value on the same slice, so joins on that key need no data movement; but if a few key values hold most rows, those slices do most of the work (skew). EVEN spreads rows round-robin, ALL copies a small table to every node, and AUTO lets Redshift choose and adjust. When a question describes one slow node or a join that shuffles data, the fix is usually the distribution key; when it describes range filters on dates, it is usually the sort key.
Stage 4: running and fixing it
This domain is about what happens after deployment: monitoring with CloudWatch metrics and alarms, logging, data quality checks (AWS Glue Data Quality), re-running failed steps, and tuning. Typical stems: a job suddenly takes twice as long (look for small files, skew, or a missing partition filter), a crawler creates the wrong schema, a scheduled job runs twice.
Beginner tip: upload the same small dataset to S3 twice, once as CSV and once as Parquet, create an Athena table over each, and run the same query on both. Athena shows "data scanned" for every query. Seeing the difference yourself makes a whole family of storage and cost questions easy.
Stage 5: protecting it
Encryption at rest (SSE-S3, SSE-KMS) and in transit; IAM for who can call which service; AWS Lake Formation for fine-grained permissions on databases, tables, columns and rows in the data lake, including tag-based access; Amazon Macie to find personal data in S3; CloudTrail for the audit trail. A common trap: answering a column-level access requirement with S3 bucket policies. Bucket policies work on objects, not on columns; that is Lake Formation's job.
A study order
- S3 layout, Glue (Data Catalog, crawlers, jobs, bookmarks) and Athena: one week.
- Redshift: loading, Spectrum, distribution and sort keys, workload management: one week.
- Streaming and orchestration: Kinesis, Firehose, MSK, Step Functions, MWAA, EventBridge.
- Operations, then security and governance, then mixed practice sets.
Practise
Practise in the question bank: every question explains each option, right and wrong. Start with the domain you just studied, then take a timed mock exam.
- DEA-C01 exam page · All 495 DEA-C01 questions · Timed mock exam · 28-day study plan
- By domain: Data Ingestion and Transformation · Data Store Management · Data Operations and Support · Data Security and Governance
Exam facts as published in the AWS exam guide for DEA-C01. Service features such as zero-ETL integrations change often; check the current documentation.