AWS DEA-C01 exam pathFree assessment

AWS DEA-C01 Practice Tests That Find Your Weak Spots

5 practice tests, a 65-question timed simulation, and a free assessment.

AWS Certified Data Engineer - Associate

Find your AWS DEA-C01 gaps in 15 minutes.

Learner actions will appear after the current account is confirmed.

Official exam guide updated Dec 12, 2025
Sample report preview

Preview the output before you study

Take a free assessment, see the decisions costing you points, or confirm where to validate breadth next instead of browsing a generic study list.

Sample readiness result690 / 1000

Know where you stand before another full practice test.

Sample top gapData ingestion and transformation decisions

Targeted repair: Fix a focused gap around batch boundaries, transformation engines, and idempotent pipeline output.

ReportTop gapsPass PlanTargeted repairsUnlocksPractice + Timed Exam
01Answer exam-style questions
02Work your Pass Plan
03Complete the final simulation
Try it

Answer one AWS DEA-C01 question

If you miss it, CramHQ identifies the decision you misunderstood, helps you fix it, then checks whether you can apply it in a different scenario.

A scheduled Spark step on an Amazon EMR on EC2 cluster began failing after a dependency bundle was updated and daily input increased. Amazon CloudWatch shows brief CPU spikes and several pending YARN containers immediately before the step exits. Diagnostic output is retained in Amazon S3. The team is considering a larger cluster because the job must finish reliably. What should the data engineer do first?

  1. Increase spark.sql.shuffle.partitions and rerun; if task durations become uniform, retain the setting and add core nodes only if YARN containers continue queuing.
  2. Increase the Amazon EMR managed scaling maximum and rerun unchanged; treat successful completion as evidence that node limits caused the failure, then retain the higher maximum.
  3. Roll back the dependency bundle and rerun; if the step completes, retain the previous dependencies and increase core nodes only when later runs queue YARN containers.
  4. Review S3 diagnostics for the failed step and Spark driver/executors; align the terminating exception with prolonged low YARNMemoryAvailablePercentage and elevated ContainerPendingRatio before changing node counts.
Show the explanation

The retained diagnostics can identify application or dependency exceptions, while prolonged YARN metric pressure indicates insufficient resources. Brief activity immediately before termination does not establish that additional nodes would address the failure.

Try it now

Free AWS DEA-C01 practice questions

Eight practice questions from the CramHQ AWS DEA-C01 bank, with answers and explanations. No signup needed. When you are ready, start the free assessment. CramHQ ranks your weak spots across the whole exam by the points they cost, and your Pass Plan starts with the costliest.

Question 1 of 8Data Ingestion and Transformation

An Amazon S3 prefix receives append-only CSV objects from one producer. The records already conform to the downstream schema, and repeated identifiers are valid. Consumers need the same records in Parquet, while recurring runs should skip previously processed objects. An AWS Glue workflow currently reads the prefix through two catalog tables and joins the results by identifier, causing record loss and longer runtimes. Which replacement should the data engineer choose?

  1. Use one AWS Glue ETL job that reads new CSV records, enables job bookmarks, joins them with the existing Parquet dataset by identifier, and rewrites the dataset.
  2. Use one AWS Glue ETL job that reads the prefix twice, enables job bookmarks, joins both inputs by identifier, and writes Parquet.
  3. Use one AWS Glue ETL job that reads the CSV prefix once, disables job bookmarks, removes duplicate identifiers, and writes Parquet.
  4. Use one AWS Glue ETL job that reads the CSV prefix once, enables job bookmarks, and writes the records directly as Parquet.
Show answer and explanation

Answer: D

The required change affects only representation. A single read-and-write path preserves valid repeated identifiers, while AWS Glue job bookmarks track previously processed S3 objects for subsequent runs.

“Consumers need the same records in Parquet”

What to notice

This limits the task to preserving source records while changing file representation, so added joins, reconciliation, or deduplication need explicit support from the stem.

How to decide

Scope the operation first: separate format conversion from record matching or cleansing, then compare choices against record preservation.

If you picked A

Why it is tempting: It mentions job bookmarks and Parquet, so it appears to address incremental runs and the required output format.

Why it is wrong: For append-only source objects that already conform to the target schema, convert the source records directly and preserve record identity unless the stem requires reconciliation.

If you picked B

Why it is tempting: It still uses AWS Glue, enables job bookmarks, and writes Parquet, so the service and output format sound aligned.

Why it is wrong: If there is one append-only source and no enrichment requirement, read that source once and convert its records; do not manufacture a multi-source integration step.

If you picked C

Why it is tempting: Removing duplicate identifiers can sound like a data-quality improvement, and the option still converts CSV to Parquet.

Why it is wrong: A format conversion job should preserve valid source records unless the scenario explicitly asks for deduplication or cleansing.

Question 2 of 8Data Ingestion and Transformation

An eight-shard provisioned Amazon Kinesis Data Streams stream ingests 4 MiB/s with evenly distributed keys. Two independent KCL applications each process every record, and their combined reads consume about half of each shard's available read throughput. Six more applications must start within minutes and also process every record; producer traffic will not change. Resharding requires a lengthy validation cycle. Which change should the data engineer make?

  1. Register one enhanced fan-out consumer per KCL application and configure each application with its own consumer ARN; retain the eight shards.
  2. Assign each KCL application a unique application name and lease table while retaining shared-throughput reads and the eight shards.
  3. Register one enhanced fan-out consumer and configure all eight KCL applications with its consumer ARN; retain the eight shards.
  4. Increase the provisioned stream to 32 shards and keep all eight KCL applications on shared-throughput reads; retain the partition keys.
Show answer and explanation

Answer: A

Each registered enhanced fan-out consumer receives dedicated throughput of up to 2 MiB/s per shard. Eight registrations support the additional full-stream readers without resharding or changing producers.

“Six more applications must start within minutes and also process every record”

What to notice

This identifies the workload as consumer fan-out: more independent applications need the full stream, so the answer must address reader throughput per application.

How to decide

Scope the change to the read side first, then compare options that give each full-stream application independent read capacity.

If you picked B

Why it is tempting: Unique application names and lease tables are real KCL isolation mechanisms, so they look like a way to separate the applications.

Why it is wrong: Separate KCL application names and lease tables isolate processing state; they do not give each application dedicated read throughput from a shard.

If you picked C

Why it is tempting: The answer names enhanced fan-out and keeps the current shard count, which matches part of the desired direction.

Why it is wrong: Dedicated enhanced fan-out throughput is scoped to each registered consumer, so independent applications need separate registered consumers when each must read the full stream.

If you picked D

Why it is tempting: More shards can increase aggregate shared read capacity, so scaling shards looks like a capacity fix.

Why it is wrong: Use shard scaling for stream capacity or partition distribution problems; use per-consumer fan-out when the bottleneck is many independent applications reading every record.

Question 3 of 8Data Store Management

An Amazon Redshift RA3 table stores 12 TB of append-only events. A reconciliation job repeatedly filters on one unique BIGINT event_id; query metrics show that these statements read nearly all table blocks. The table is sorted by event_time and distributed by tenant_id, which already collocates its large joins. The table uses ENCODE AUTO. Which table redesign will most directly improve the reconciliation job while preserving join performance?

  1. Set event_id as both the single-column SORTKEY and DISTKEY; retain ENCODE AUTO.
  2. Retain the current keys; set event_id to ENCODE AZ64 and payload to ENCODE ZSTD.
  3. Declare event_id as the PRIMARY KEY; retain the current SORTKEY, DISTKEY, and ENCODE AUTO.
  4. Set event_id as the single-column SORTKEY; retain DISTKEY tenant_id and ENCODE AUTO.
Show answer and explanation

Answer: D

A sort key orders rows so zone maps can exclude blocks that cannot contain the requested event_id. Retaining tenant_id distribution preserves the existing join colocation.

“repeatedly filters on one unique BIGINT event_id; query metrics show that these statements read nearly all table blocks”

What to notice

This cue points to a physical layout problem for scan pruning on the filtered column, not a general storage-size problem.

How to decide

Scope the fix to the access pattern first: identify which table-layout feature can reduce block reads for the repeated filter.

If you picked A

Why it is tempting: The event_id sort key part is useful because the workload repeatedly filters on event_id.

Why it is wrong: Use the sort key for block skipping, but keep the distribution key that already colocates large joins unless join placement also needs to change.

If you picked B

Why it is tempting: Better compression can reduce storage and some I/O, especially on large tables.

Why it is wrong: Choose the physical design lever that matches the bottleneck: sort order helps Redshift skip blocks for filter predicates; compression does not order rows by the filtered column.

If you picked C

Why it is tempting: In some database systems, primary keys are closely associated with indexed access paths.

Why it is wrong: In Amazon Redshift, primary key constraints are informational; they do not create physical indexes, enforce uniqueness, or reorganize table blocks.

Show all 8 practice questions
Question 4 of 8Data Store Management

A team bulk-loaded documents and precomputed embeddings into Amazon OpenSearch Service. Validation confirms the arrays match the embedding model's dimension and output. Because the index used dynamic mapping, the embedding field is mapped as float. A k-NN request fails before scoring, and the team must restore search quickly without changing established relevance behavior. Which action should the data engineer take?

  1. Create a replacement OpenSearch Service index with a dimension-matched knn_vector mapping; regenerate every embedding through an ingest pipeline, then switch the alias.
  2. Create a replacement OpenSearch Service index with index.knn enabled and a dimension-matched knn_vector mapping; use Reindex to copy records unchanged, then switch the alias.
  3. Close the existing index, enable index.knn, change the embedding field from float to knn_vector with Put Mapping, reopen it, and retain the stored vectors.
  4. Create a replacement index with index.knn enabled but retain the float mapping; reindex the existing vectors, increase ef_search, then switch the alias.
Show answer and explanation

Answer: B

The stored values are already valid. OpenSearch requires a knn_vector mapping for k-NN search, and an existing field's type cannot be changed. Reindexing into the correctly mapped replacement preserves the vectors and relevance behavior.

“the embedding field is mapped as float”

What to notice

This shows that the stored values were treated as ordinary numeric values, so separate value validity from the schema needed for vector search.

How to decide

Scope the repair to the index schema and field type before considering embedding generation or search tuning.

If you picked A

Why it is tempting: Creating a correctly mapped replacement index is part of the right recovery path, so adding an ingest pipeline can sound thorough.

Why it is wrong: If embeddings are already valid, fix the OpenSearch vector field mapping and copy the values unchanged; do not regenerate embeddings just to repair indexing.

If you picked C

Why it is tempting: Closing and reopening an index can make schema maintenance feel possible without moving data.

Why it is wrong: Do not rely on Put Mapping to convert an existing float field into knn_vector; create a replacement index with the correct vector mapping.

If you picked D

Why it is tempting: The answer mentions a replacement index, reindexing, and a k-NN search parameter, which are all related to vector search operations.

Why it is wrong: OpenSearch k-NN search still requires the embedding field itself to be mapped as knn_vector; index settings and ef_search do not make a float field searchable as a vector.

Question 5 of 8Data Operations and Support

An AWS Glue PySpark job processes sales transactions. Its output must keep every transaction and add the item's total revenue and position within each customer account and business date. Items with equal totals must receive the same position, and the following distinct total must advance by one. Which transformation should the data engineer use?

  1. Calculate revenue with a SUM window by account, date, and item; then use rank in an account-date window ordered by that revenue descending.
  2. Group by account, date, and item and sum revenue; then use dense_rank in an account-date window ordered by that revenue descending.
  3. Calculate revenue with a SUM window by account, date, and item; then use dense_rank in an account-date window ordered by that revenue descending.
  4. Calculate revenue with a SUM window by account, date, and item; then use row_number in an account-date window ordered by that revenue descending.
Show answer and explanation

Answer: C

The SUM window adds each item's total without changing transaction granularity. dense_rank assigns equal totals the same position and increments by one for each distinct total.

“Its output must keep every transaction”

What to notice

Keeping every transaction means the transformation must preserve input-row granularity while adding calculated values.

How to decide

Scope the output grain first. Eliminate choices that replace transaction rows with grouped summary rows before comparing the remaining calculation behavior.

If you picked A

Why it is tempting: The SUM window by account, date, and item correctly keeps every transaction and adds the item total, so the first half looks right.

Why it is wrong: A row-preserving aggregate is necessary here, but the complete transformation must also assign tied totals the same position and advance the next distinct total by one.

If you picked B

Why it is tempting: groupBy feels appropriate because the output needs one total for each account, date, and item.

Why it is wrong: When every original transaction must remain in the result, calculate the total with a window over the grouping columns instead of grouping the rows into one result per item.

If you picked D

Why it is tempting: The SUM window correctly preserves every transaction, and row_number can look like a generic way to create a position column.

Why it is wrong: A row-preserving aggregate must be paired with a ranking function whose tie behavior matches the requirement; row_number assigns a different number to each row.

Question 6 of 8Data Operations and Support

A team runs Spark jobs continuously on a long-lived Amazon EMR cluster. Demand is stable, but engineers want to reduce cluster administration. Finance requires the core nodes to consume a specific existing EC2 On-Demand Capacity Reservation and task nodes to maintain a separately defined Spot capacity target. An engineer proposes moving the workload to EMR Serverless. Which design should the team choose?

  1. Move to Amazon EMR Serverless; configure pre-initialized capacity for the baseline and maximum application capacity for additional workers.
  2. Retain Amazon EMR on EC2 with instance fleets; set the core On-Demand target with use-capacity-reservations-first and its resource group ARN, and set the task Spot target.
  3. Retain Amazon EMR on EC2 with instance fleets; set both capacity targets, but configure the core fleet to use open capacity reservations first.
  4. Retain Amazon EMR on EC2 with uniform instance groups; use On-Demand core nodes and Spot task nodes, relying on automatic reservation matching.
Show answer and explanation

Answer: B

Instance fleets expose independent On-Demand and Spot target capacities. The usage strategy and CapacityReservationResourceGroupArn direct the core fleet toward reservations in the specified resource group.

“Finance requires the core nodes to consume a specific existing EC2 On-Demand Capacity Reservation and task nodes to maintain a separately defined Spot capacity target.”

What to notice

This is the hard constraint: the answer must support both targeted EC2 reservation consumption for core nodes and a separate Spot target for task nodes.

How to decide

Treat this as a capacity-control cue. Compare each option by whether it exposes the required EC2 reservation targeting and the separate Spot task capacity target before considering administration reduction.

If you picked A

Why it is tempting: EMR Serverless does reduce cluster management and can define baseline and maximum worker capacity.

Why it is wrong: Use serverless analytics only when its capacity controls satisfy the hard workload constraints; it does not provide EMR on EC2 instance-fleet purchasing targets or targeted EC2 Capacity Reservation selection.

If you picked C

Why it is tempting: The option stays on EMR on EC2 with instance fleets and mentions using capacity reservations first, so it matches much of the scenario language.

Why it is wrong: A requirement to consume a specific existing Capacity Reservation requires targeted reservation configuration, including the appropriate reservation resource group, not only open reservation preference.

If you picked D

Why it is tempting: Uniform instance groups can separate core and task purchasing choices, which resembles the required On-Demand and Spot split.

Why it is wrong: Use the EMR capacity construct that exposes the required reservation targeting controls; broad On-Demand and Spot separation is not enough when the reservation must be specific.

Question 7 of 8Data Security and Governance

An AWS Glue job runs in private subnets and sends a small batch each hour to Amazon Kinesis Data Streams by calling PutRecords. The subnet route tables already direct outbound traffic through approved NAT gateways, and the job role has the required stream permissions. The organization does not require AWS PrivateLink for this flow. Which connectivity and authorization design meets the requirements with the lowest additional ongoing cost?

  1. Create a Kinesis Data Streams interface VPC endpoint with private DNS and an endpoint policy permitting PutRecords, then remove the job role's stream permissions.
  2. Use the standard Regional Kinesis Data Streams endpoint directly from the private subnets, relying on the job role's IAM policy without NAT or an endpoint.
  3. Create a Kinesis Data Streams interface VPC endpoint with private DNS and a restrictive endpoint policy, while retaining the job role's IAM policy.
  4. Use the standard Regional Kinesis Data Streams endpoint through the existing NAT gateways, and retain authorization in the job role's IAM policy.
Show answer and explanation

Answer: D

The existing NAT route provides connectivity to the Regional endpoint, and IAM already authorizes PutRecords. For this low-volume flow, avoiding an unnecessary interface endpoint minimizes additional recurring charges.

“already direct outbound traffic through approved NAT gateways”

What to notice

An existing approved egress path can satisfy connectivity to public Regional service endpoints, so a new private endpoint needs a stated requirement or cost justification.

How to decide

Separate connectivity from authorization, then verify whether the provided network path already satisfies the connectivity requirement.

If you picked A

Why it is tempting: The endpoint policy mentions PutRecords, so it can look like the endpoint itself authorizes the request.

Why it is wrong: A VPC endpoint policy restricts which requests may use the endpoint, but the caller still needs identity-based IAM permission for the stream action.

If you picked B

Why it is tempting: The job role already has the required permissions, so the authorization part of the flow is satisfied.

Why it is wrong: IAM authorizes the API call; a private-subnet workload still needs a reachable network path such as NAT or an interface endpoint to call a public Regional service endpoint.

If you picked C

Why it is tempting: The design preserves IAM authorization and adds a private endpoint, so it looks more controlled than using the existing NAT path.

Why it is wrong: Use an interface endpoint when private service access is required or otherwise justified; do not add it for lowest additional cost when approved NAT already provides the needed reachability.

Question 8 of 8Data Security and Governance

A contractor's statement of work covers implementing pipelines and queries using the notebooks, data, and compute in one existing Amazon SageMaker Unified Studio project. An audit finds that the contractor was assigned as an owner of the division's domain unit. The contractor will not create projects or manage other users' access. Which action should the administrator take?

  1. Remove the contractor as domain unit owner, then add the contractor to the existing project as a contributor.
  2. Create a nested domain unit, move the existing project into it, make the contractor its domain unit owner, and add the contractor as a contributor.
  3. Retain the contractor as domain unit owner, and also add the contractor to the existing project as a contributor.
  4. Remove the contractor as domain unit owner, then add the contractor to the existing project as an owner.
Show answer and explanation

Answer: A

A contributor can use the project's shared data and resources. Removing domain unit ownership also removes delegated authority over division-level authorization policies that the contractor does not need.

“one existing Amazon SageMaker Unified Studio project”

What to notice

This limits the required access boundary to an existing project rather than a broader organizational unit.

How to decide

Scope the role to the existing project first, then verify whether any wording requires boundary administration.

If you picked B

Why it is tempting: Moving the project and adding contributor membership appears to preserve the needed project access.

Why it is wrong: Do not create or delegate a domain unit when project contributor membership alone satisfies work inside one existing project.

If you picked C

Why it is tempting: Adding contributor access fixes the immediate project access need, so the retained owner role can seem harmless.

Why it is wrong: After assigning the needed project role, remove broader domain unit authority unless the scenario requires domain unit administration.

If you picked D

Why it is tempting: Project owner seems scoped to the project, so it can look safer than domain unit owner.

Why it is wrong: Use the least project-level role that supports the stated work. A project owner manages the project and its members; a contributor performs the project work described here.

CramHQ links wrong answers to the mistake behind them. Start with the free assessment, and CramHQ keeps finding your gaps as you practice and adds a repair for each to your Pass Plan.

Start the free AWS DEA-C01 assessment

Why study with CramHQ

Strengthen your AWS data-engineering decisions.

Start with the free AWS Data Engineer Associate assessment. CramHQ uses your result to rank the next step, explain the decisions that matter, and check you again.

Learner actions will appear after the current account is confirmed.

  1. A clear next step, chosen for you

    The Pass Plan ranks gaps by the points they cost and opens the top repair.

  2. Understand why an answer works

    The reasoning behind the right answer and where the alternatives break.

  3. Prove it on a retest

    A fresh question on the same decision verifies the gap is closed.

  4. Track it on your readiness map

    Strengthened areas change colour; pick the next priority from what is left.

  • Flashcards saved from your repairs
  • Optional hints when you are stuck
  • Timed simulations at exam pace
Exam syllabus

Exam-weighted coverage.

Coverage follows the published AWS DEA-C01 skill areas measured as of Dec 12, 2025, so your practice stays aligned.

Data Ingestion and Transformation

34%

18 concepts

Data Store Management

26%

17 concepts

Data Operations and Support

22%

17 concepts

Data Security and Governance

18%

20 concepts

Official AWS exam changes

AWS exam update history

Changes recorded from official AWS exam sources.

Published Dec 12, 2025Major updateAWS added data-engineering skills for LLM processing, open table formats, vector indexes, catalogs, and governance.Integrated by CramHQ

Compared with Exam guide version 1.0

Added

  • LLM-based data processing
  • Open table formats and vector index concepts
  • Business data catalogs, vectorization, and governance skills

Changed

  • Knowledge and skills were consolidated under each task.
  • Several existing skills and the in-scope service list were updated.

Removed

  • AWS Cloud9, AWS CodeCommit, and AWS Schema Conversion Tool were removed from the in-scope list.
Updated

CramHQ completed the DEA-C01 update for AWS exam guide version 1.1, including the added data-processing and governance scope.

View official source
Practice and validate

Build breadth before the timed simulation.

Use separate exam-style practice quizzes to validate breadth, review explanations and your answers, use hints when stuck, and build confidence before the timed simulation, with an optional timer.

5 Practice Quizzes

01Practice quiz

AWS DEA-C01 Practice Quiz 1

Questions
65
Timer
Optional
02Practice quiz

AWS DEA-C01 Practice Quiz 2

Questions
65
Timer
Optional
03Practice quiz

AWS DEA-C01 Practice Quiz 3

Questions
65
Timer
Optional
04Practice quiz

AWS DEA-C01 Practice Quiz 4

Questions
65
Timer
Optional
05Practice quiz

AWS DEA-C01 Practice Quiz 5

Questions
65
Timer
Optional
Timed readiness check

Check readiness under timed conditions.

Use this after practice when you want an exam-style pacing, stamina, and breadth check.

FinalReadiness simulation
CramHQ target720 / 1000

AWS DEA-C01 Final Readiness Simulation

Questions
65
Time
130 min
Mode
Timed

Start with the free assessment; full Pass Plan access unlocks the timed exam simulation.

Question formats

Practice the formats this exam path supports.

CramHQ shows the interaction patterns available for AWS DEA-C01 from the active exam path inventory.

Single answer

Choose the best answer from exam-style options.

Multi-answer

Select every correct option when more than one applies.

Exam integrity

Original practice, aligned to public objectives.

CramHQ helps you practice the decisions behind AWS DEA-C01 without using copied or confidential exam content.

Practice contentOriginal questions

Original practice questions aligned to public exam objectives.

Coverage sourcePublic objectives

Coverage is checked against the published exam objectives.

Provider relationshipIndependent prep
AWS DEA-C01 exam guide

CramHQ is independent and is not affiliated with AWS.

Quick answers

AWS DEA-C01 practice test questions, answered.

Do I need to sign up?

No. You can start the assessment and review your report before creating an account. Sign up when you want to save the report and continue the Pass Plan.

Is the report free?

Yes. The assessment, readiness report, and your first targeted repair are free.

Are these copied exam questions or dumps?

No. CramHQ uses original practice items designed to test the decisions and concepts behind the exam.

Is CramHQ affiliated with AWS?

No. CramHQ is independent and is not endorsed by AWS.

What do I get after the assessment?

Save your report to keep the Pass Plan, fix your first gap free, and unlock the full Pass Plan when you want every repair, practice quizzes, and timed simulation.

I have already studied a lot. Is this still worth it?

That is exactly who it is built for. CramHQ keeps tracking the last few gaps between you and a comfortable pass, the ones a raw score hides, and puts the one costing you the most points at the top of your Pass Plan.

How long do I get access?

12 months of full access from purchase, with room to come back after a break.

What is your refund policy?

Paid exam paths come with a 7-day money-back guarantee. Conditions apply; see the Refund Policy. You can also try the assessment and your first repair free before you buy. Refund Policy

More exam paths

Choose the next certification path that fits your current goal.