AWS Certified Data Engineer – Associate (DEA-C01)
The data pipeline exam. It covers ingesting, transforming, storing, operating and governing data on AWS.
Verified against official sources on 2026-11-17.
AWS Certified Data Engineer – Associate (DEA-C01) checks that you can build data pipelines on AWS: ingest and transform data, choose and model data stores, run and monitor pipelines, check data quality, and secure and govern the data. Banking data pipelines (transactions, reconciliation, regulatory reporting) are one of your priorities, so this exam fits well.
At a glance
| Code | DEA-C01 |
| Level | Associate |
| Intended audience | 2–3 years in data engineering, plus 1–2 years hands-on with AWS |
| Questions | 65 (50 scored + 15 unscored) |
| Time | 130 minutes |
| Passing score | 720 (scaled 100–1000) |
| Price | USD 150 |
| Question types | Multiple choice, multiple response |
| Valid for | 3 years. Renewed by passing AIP-C01, or with the paid Skill Builder maintain option |
| Out of scope | ML training and inference, language-specific syntax, drawing business conclusions from data |
Official domains and weightings
| Domain | Weight |
|---|---|
| 1. Data Ingestion and Transformation | 34% |
| 2. Data Store Management | 26% |
| 3. Data Operations and Support | 22% |
| 4. Data Security and Governance | 18% |
Key AWS services tested
- Ingestion: Kinesis Data Streams, Amazon Data Firehose, Amazon MSK, DMS (including change data capture), AWS Transfer Family, DataSync, AppFlow, EventBridge, S3 event notifications
- Transformation and orchestration: AWS Glue (ETL jobs, crawlers, Data Catalog, job bookmarks, Glue Data Quality), Amazon EMR, Lambda, Step Functions, Amazon MWAA, Glue workflows
- Storage and modelling: S3 (data lake zones, partitioning, lifecycle), Redshift (and Redshift Serverless, Spectrum, distribution and sort keys), DynamoDB, RDS and Aurora, Apache Iceberg tables, file formats (Parquet, ORC, Avro, CSV, JSON)
- Analytics: Athena (partitions, partition projection, CTAS), QuickSight, OpenSearch Service
- Security and governance: Lake Formation (column, row and tag-based access), IAM, KMS, Macie, CloudTrail, Secrets Manager, Glue Data Catalog permissions, data sharing
What you already know vs what you must learn
| You already know (from .NET, SQL Server and banking) | You must learn |
|---|---|
| T-SQL, joins, indexes, query plans | Columnar formats and partitioning: why Parquet partitioned by date beats CSV for Athena and Redshift Spectrum |
| SSIS, SQL Agent jobs, batch file processing | Glue ETL (PySpark basics), job bookmarks, crawlers, and orchestration with Step Functions or MWAA |
| SQL Server CDC and replication | DMS full load plus CDC, and Kinesis and Firehose for streaming |
| Row-level security in SQL Server | Lake Formation permissions on S3 data, and LF-tags |
| Reconciliation and end-of-day batch windows | Event-driven pipelines (S3 events to EventBridge to Step Functions), handling late-arriving data, idempotent reprocessing |
Preparation order and study plan (about 50–60 hours over 5–7 weeks)
| Week | Focus | Hours |
|---|---|---|
| 1 | Exam guide. Module 07 data lake design (zones, formats, partitioning) | 8–10 |
| 2 | Glue (crawlers, ETL, bookmarks, Data Quality), Athena. Build the transaction pipeline project | 10–12 |
| 3 | Streaming: Kinesis, Firehose, MSK basics. DMS CDC from SQL Server | 8–10 |
| 4 | Redshift (Serverless, distribution and sort keys, Spectrum), DynamoDB patterns (Module 06) | 8 |
| 5 | Domain 4: Lake Formation, KMS, Macie, PII masking in the pipeline. Domain 3: monitoring | 8 |
| 6 | Free Exam Prep course, the 20-question set, this page's questions, a mock exam | 6–8 |
Mapped platform modules
- Module 07 — Data engineering & ETL: the core of Domains 1–3
- Module 06 — Databases: choosing a data store, Redshift, DynamoDB, DMS
- Module 08 — Messaging: Kinesis, EventBridge, Step Functions
- Module 05 — Security: KMS, Macie, data classification (Domain 4)
- Module 11 — Monitoring: pipeline monitoring and alerts (Domain 3)
Practical skills to build
- Build the Module 07 banking pipeline: SFTP drop, raw S3 zone, a Glue job that validates and quarantines rows, partitioned Parquet output, Athena queries, and a daily reconciliation report
- Add Glue Data Quality rules (for example, valid IBAN format, positive amounts, no duplicate transaction IDs)
- Stream fictional card transactions through Kinesis to Firehose to S3 as Parquet
- Restrict analysts from seeing PII columns with Lake Formation, and prove it with Athena
- Compare scan cost for the same query on CSV and on partitioned Parquet
Exam-prep checklist
- I can choose between Kinesis Data Streams, Firehose, MSK, SQS and DMS for a given ingestion scenario
- I know Glue job bookmarks, crawlers, the Data Catalog and Glue Data Quality
- I can explain partitioning, partition projection, compression and file sizing for Athena performance and cost
- I know Redshift distribution styles, sort keys, Spectrum and data sharing at a high level
- I can design S3 lifecycle rules for raw, curated and archive data
- I can secure a data lake with Lake Formation (column, row and tag-based access) and KMS
- I know how to orchestrate with Step Functions compared with MWAA, and how to retry and alert on failures
- I can mask or tokenise PII inside a pipeline
Official links
Practice questions
These 5 questions are original, banking-flavoured scenarios. They aren't real exam questions.
A correspondent bank sends a daily CSV settlement file by SFTP. The receiving bank wants the files to land directly in S3, wants no servers to patch, and needs the arrival of each file to trigger a validation pipeline. What should the data engineer use?
- A. An EC2 instance running OpenSSH that copies files to S3 with a cron job
- B. AWS Transfer Family (SFTP) with an S3 bucket as storage, and S3 event notifications sent to EventBridge to start a Step Functions workflow
- C. AWS DataSync from the partner's server
- D. Amazon Data Firehose with an SFTP source
Correct answer: B
Explanation: Transfer Family provides a fully managed SFTP endpoint that writes straight to S3, with per-user access controlled by IAM roles. S3 events, routed through EventBridge, start the pipeline with no polling. You can add PGP decryption with Transfer Family managed workflows or in the pipeline's first step.
Why the others are wrong:
- A: It works, but you have to patch, harden, scale and monitor a server yourself. That goes against the "no servers to patch" requirement.
- C: DataSync copies data between storage systems you control using an agent. It doesn't provide an SFTP endpoint for an external partner to push files to.
- D: Firehose ingests streaming records. It has no SFTP source.
Analysts query 3 years of transaction history in S3 with Athena. The data is uncompressed CSV, and almost every query filters on transaction_date and selects 5 of 60 columns. Queries are slow and each one scans terabytes. What gives the biggest improvement?
- A. Compress the CSV files with gzip
- B. Convert the data to Parquet, partitioned by date, for example with a Glue ETL job or Athena CTAS, and register the partitions (or use partition projection)
- C. Ask AWS to increase the Athena service quota
- D. Load the data into Amazon RDS for PostgreSQL
Correct answer: B
Explanation: Athena charges by data scanned. Parquet is columnar, so a query reading 5 of 60 columns reads only those columns. It's also compressed and holds statistics that let Athena skip data. Partitioning by date means queries filtered on date read only the matching partitions. Together these often cut scanned data, and cost, by orders of magnitude.
Why the others are wrong:
- A: gzip reduces size, but every column is still read. Gzipped files also can't be split, which limits parallelism.
- C: Quotas limit concurrency. They don't change how much data a query has to scan.
- D: RDS isn't designed for analytics over terabytes of history, and it adds cost and operational work.
A nightly AWS Glue ETL job reads every file under s3://raw/transactions/ and writes curated Parquet. As history grows, each run reprocesses everything, taking longer and creating duplicates in the output. How should the engineer make the job process only new data?
- A. Turn on S3 Versioning on the raw bucket
- B. Turn on Glue job bookmarks for the job
- C. Run the Glue crawler more often
- D. Register the bucket with Lake Formation
Correct answer: B
Explanation: Job bookmarks store state between runs, so the job processes only data it hasn't seen before. Your script has to use the transformation_ctx parameter for this to work. For extra safety, make writes idempotent, for example by overwriting the target partition or by using Iceberg MERGE, so that a rerun after a failure doesn't create duplicates.
Why the others are wrong:
- A: Versioning keeps old object versions. It doesn't track what a job has processed.
- C: Crawlers update the Data Catalog schema and partitions. They don't change what the job reads.
- D: Lake Formation governs permissions, not incremental processing.
The curated transactions table in the data lake contains customer_name, iban and amount. The fraud team may see all columns. The finance analytics team must never see customer_name or iban, but must query the rest through Athena. What's the most maintainable approach?
- A. Write an S3 bucket policy that denies finance access to those columns
- B. Grant permissions with AWS Lake Formation: column-level permissions, or a data filter or LF-tags, that exclude the PII columns for the finance role
- C. Create a separate copy of the table without PII for finance, and keep both in sync
- D. Give the finance role an IAM policy that allows only
athena:StartQueryExecution
Correct answer: B
Explanation: Lake Formation adds table, column, row and cell-level permissions on top of the Glue Data Catalog. Athena, Redshift Spectrum, EMR and Glue enforce these permissions. Tag-based access control (LF-tags such as classification=pii) scales well as new tables are added. Whether finance needs that data at all is a GDPR data-minimisation question for your privacy team.
Why the others are wrong:
- A: S3 permissions work at the object level. A Parquet file holds all columns, so a bucket policy can't filter columns.
- C: Copies double the storage, can drift out of sync, and multiply PII handling. That's a governance risk.
- D: Permission to run queries doesn't restrict which columns the query can read.
Card authorisation events arrive in a Kinesis data stream at about 2,000 records per second. The fraud analytics team wants them in S3 as Parquet, partitioned by hour, within a few minutes of arrival, with as little custom code as possible. What should the engineer build?
- A. A Lambda function that writes each record to S3 as its own object
- B. An Amazon Data Firehose stream reading from the Kinesis stream, with record format conversion to Parquet (using a Glue Data Catalog table schema) and a time-based S3 prefix
- C. A Glue crawler pointed at the Kinesis stream
- D. An SQS queue in front of S3
Correct answer: B
Explanation: Firehose is a managed delivery service. It buffers records by size or time, can convert JSON to Parquet or ORC using a schema from the Glue Data Catalog, and writes to S3 with time-based or dynamic partitioning prefixes. You don't manage any servers or batching code.
Why the others are wrong:
- A: One object per record creates millions of tiny files, which are slow and expensive to query. You'd also have to write the batching and format conversion yourself.
- C: Crawlers catalog data at rest. They don't move streaming data.
- D: SQS is a message queue with no delivery to S3 or format conversion.
Independent learning resource — not affiliated with, endorsed or sponsored by Amazon Web Services. AWS and related marks are trademarks of Amazon.com, Inc. or its affiliates. All banking data in examples is fictional. Sample architectures are for learning and are not automatically compliant with GDPR, PCI DSS or any banking regulation.
AWS Certified Security – Specialty (SCS-C03)
The most important exam for banking. It covers detection, incident response, infrastructure security, IAM, data protection and governance.
AWS Certified CloudOps Engineer – Associate (SOA-C03)
The operations exam, formerly SysOps Administrator. It covers monitoring, reliability, automation, security, networking and troubleshooting on AWS.