Based on the DEA-C01 exam guideReport an error
Today is real-exam mode. You solve integrated scenarios that mix the 4 domains and get a feel for the exam pace. The point is less to get the answers right than to sharpen the "requirement keyword → service mapping" reflex. First, turn the time into numbers.
| Pass | Allocated time | Per question | Checkpoint |
|---|---|---|---|
| First pass (all 65 questions once each) | About 90 minutes | About 83 seconds | Around question 22 at 30 minutes, around question 43 at 60 minutes |
| Second review (flagged questions) | About 30 minutes | Depends on the number of flagged questions | If you flagged 15 questions, 2 minutes each |
| Third check | About 10 minutes | n/a | Number of selections in multiple response, blank answers |
Multiplying the domain weights by the 50 scored questions gives a rough question count. Since the weights are "the ratio among scored questions", treat them only as approximations.
| Domain | Weight | Approximate share of the 50 scored questions |
|---|---|---|
| 1. Ingestion and Transformation | 34% | 17 questions |
| 2. Data Store Management | 26% | 13 questions |
| 3. Operations and Support | 22% | 11 questions |
| 4. Security and Governance | 18% | 9 questions |
Break a long question down in the following order.
💡 Related theory: Most wrong answers are "technically possible but violate the requirements in operational burden, cost, or real-time capability." The heart of the Associate exam is choosing "most suitable", not "can do".
Before a qualifier in the question sentence is a criterion for choosing the answer, it is a tool for crossing out options. For each qualifier, the shape of the option that drops out first is predictable.
| Qualifier | Options to cross out first | Side that survives |
|---|---|---|
| Minimal operational overhead | Installing directly on EC2, always-on clusters, hand-written consumers and schedulers | Firehose, Glue, Athena, Lambda, Step Functions |
| Minimal cost (intermittent use) | Provisioned resources running all day | Pay-per-use serverless, S3 tiering |
| Minimal cost (constant, heavy use) | Configurations where scanned-data or request-count charges grow without bound | Provisioned capacity, format and partition optimization |
| Real-time (within seconds) | Batch schedules, load paths with large buffers | Stream + Flink/Lambda |
| Near-real-time (minutes allowed) | Hand-built stream processing | Firehose, micro-batching |
| Minimal code changes | Rewriting for a different engine | Compatible services (MSK, EMR, MWAA) |
| Keep existing data / no downtime | Dump and restore, one-time copy | DMS full load + CDC |
| Least privilege | Wildcard policies, long-term access keys | Roles, narrowly scoped policies, Lake Formation |
| Scenario signal | Typical answer combination |
|---|---|
| Real-time clickstream → zero-management loading → SQL analysis | Kinesis Firehose → S3 → Athena |
| Hourly incremental ETL → load into DW | Glue (bookmarks) → Redshift |
| Real-time replication of DB changes | DMS (CDC) → S3/Redshift |
| Coordinating heterogeneous stages + retries | Step Functions |
| Sensitive data discovery + column access control | Macie + Lake Formation |
Even the same pattern changes its combination when one more condition is attached. Integrated questions ask about this "swap of one slot".
| Base combination | Added condition | Slot that changes |
|---|---|---|
| Firehose → S3 → Athena | Other apps also read the same events in real time | Put Kinesis Data Streams in front, and Firehose becomes one of its consumers |
| Firehose → S3 → Athena | Keyword search and second-level dashboards are needed | Make the destination OpenSearch Service |
| Glue (bookmarks) → Redshift | Source rows are updated and deleted | DMS CDC instead of bookmarks |
| Glue (bookmarks) → Redshift | Processing tens of TB with existing Hive scripts | EMR instead of Glue |
| DMS (CDC) → S3 | Also doing real-time aggregation in 5-minute windows | DMS → Kinesis Data Streams → Managed Service for Apache Flink |
| Chaining Glue jobs with Step Functions | Hundreds of Airflow DAGs already exist | MWAA |
| Querying S3 with Athena | Hundreds of people view the same dashboard all day | Redshift (Spectrum for cold data) |
| Lake Formation column permissions | You don't even know where the PII is | Put Macie in front |
| SSE-S3 | Audit key usage and block decryption per role | SSE-KMS customer managed key |
Below are three passages that do not overlap with the practice questions, broken down in four steps. Don't just read them; cover each step and answer it yourself first.
Passage A) A retailer collects sales events generated at store POS terminals. An
inventory alert app and a pricing analytics app must each read the events
independently, and if there is a bug in the processing logic, the past
3 days must be reprocessed. What is the most suitable ingestion service?
1) Question -> one ingestion service
2) Keywords -> two consumers reading independently / reprocess 3 days
3) Eliminate -> SQS (one message goes to one consumer, no replay), Firehose (no reprocessing)
4) Choose -> Kinesis Data Streams + extend retention to 3 days or more
(the default 24 hours does not satisfy "3 days")Passage B) An analytics team converts the previous day's orders (CSV) in S3 to Parquet
every morning and loads them into Redshift. If the conversion fails, the
load must be skipped and the owner notified. They don't want to operate
servers. What is the configuration with the least operational overhead?
1) Question -> the configuration with the least operational overhead
2) Keywords -> daily batch / order (convert -> load) / notify on failure / no servers
3) Eliminate -> EC2 cron (server operations, doesn't know whether the earlier step succeeded),
MWAA (cost of an always-on environment, no mention of existing Airflow assets)
4) Choose -> EventBridge schedule -> Step Functions runs the Glue job -> COPY in order,
SNS notification in CatchPassage C) Several research teams query a medical data lake with Athena. Patient
identifier columns must be visible only to approved teams, and it must be
possible to audit who decrypted data with which key. Which combination
meets the requirements? (Choose two)
1) Question -> a combination, two items
2) Keywords -> column-level access control / decryption auditing
3) Eliminate -> S3 bucket policy (cannot hide columns), SSE-S3 (key usage cannot be audited),
GuardDuty (threat detection)
4) Choose -> Lake Formation column permissions + SSE-KMS customer managed key
(key usage recorded by CloudTrail)An integrated scenario looks long, but each sentence asks about only one slot. Write the slot beside each sentence and it turns into four short problems.
| Sentence in the passage | Slot | Candidates it narrows to |
|---|---|---|
| "Thousands of sensor events arrive per second" | Ingestion (Domain 1) | Kinesis Data Streams, Firehose, MSK |
| "Store it in S3 as Parquet and query it by date" | Storage (Domain 2) | Date partitions, Glue Data Catalog, Athena |
| "An alert is needed if loading stops for more than an hour" | Operations (Domain 3) | CloudWatch alarm, SNS |
| "The equipment location column must be visible only to the operations team" | Security (Domain 4) | Lake Formation column permissions |
| Type | What it looks like | Target time | How to solve |
|---|---|---|---|
| Simple mapping | One or two sentences, "Which service is most suitable?" | 30-60 seconds | Decide by a single keyword |
| Look-alike service comparison | Two or more services of the same category among the options | 1-2 minutes | Separate them with a single distinguishing axis (reprocessing or not, who operates it, etc.) |
| Failure diagnosis | "These symptoms appear. What is the cause/action?" | 2 minutes | Trace backward in the order symptom → cause → action. Cross out first the options that cite a control the service doesn't have |
| Design combination | Long passage, options that are combinations of several services | 2-3 minutes | Assign sentences to slots and compare only the other slots |
| Multiple response | "Choose two/three" | 2-3 minutes | Count the required number and judge each option true or false one by one |
If you score today's 8 questions and end with "I'll read the missed ones again", the same mistake will repeat in the exam room. You have to classify the reasons to get a prescription.
| Reason for the miss | Diagnostic question | Prescription |
|---|---|---|
| Knowledge gap | Even after the explanation, is it the first time you've heard of that feature? | Go back to that week's main note |
| Missed qualifier | Was the answer you chose functionally possible? | Keep to the order of reading the question sentence first |
| Confused look-alike services | Were the correct answer and your answer a pair in the same category? | Say the distinguishing axis in one word from Day 4's service pair tables |
| Chose an over-engineered design | Did you choose the more powerful service? | Be suspicious of "Flink for simple loading" and "Redshift for occasional queries" |
| Time pressure | Would you have gotten it right if time had remained? | Set checkpoint times and practice flagging and moving on |
Tomorrow we put together the "requirement → service" translation table and the commonly missed pitfalls to look at right before the exam.
Click a choice to reveal the answer and explanation.
Question 1
You must receive a mobile app's clickstream in real time, load it into S3 as Parquet without managing additional servers, and let analysts query it with ad hoc SQL. Which combination is the most suitable?
Question 2
You must continuously replicate changes from an on-premises PostgreSQL database to an AWS analytics environment in near real time. After the initial full load, ongoing CDC is required. Which service is the most suitable?
Question 3
You need to run multiple Lambda functions and Glue jobs in sequence, with per-step branching, automatic retries, and visual execution tracking. You want to minimize operational overhead by running serverless. Which service is the most suitable?
Question 4
Several departments in a data lake use the same table, but only the HR team may view the salary column and it must be masked for other teams. You want to control permissions centrally at the column level. Which service is the most suitable?
Question 5
Redshift dashboard queries frequently access only the last 30 days of data, while cold data older than one year is queried only occasionally. You want to reduce cost while still being able to query the cold data with SQL when needed. Which approach is the most suitable?
Question 6
A Glue ETL job occasionally fails and is rerun, but the same data ends up loaded into the target more than once. Which design principle should be prioritized to make retries safe?
Question 7
You are building a serverless ETL that processes only the new logs arriving hourly in the S3 raw zone and converts them into the curated zone (Parquet). Files that have already been processed must not be read again. Which combination is the most suitable?
Question 8
You must aggregate sales totals per store in real time over 5-minute windows from a stream of order events. You want to handle the window aggregation using SQL. Which service is the most suitable?