AWS Certified Data Engineer - Associate (DEA-C01)

Classify a source by volume, arrival pattern, change behavior, schema stability and recovery needs before selecting a service. Contrast scheduled files or database extracts with streams. Record what constitutes a unique business event, how late arrivals are handled and whether a re-read can be safely deduplicated. Pick a suitable place for raw, immutable input so the transformation can be replayed.

Reviewable practice outputWrite two source contracts—one batch and one near-real-time—with expected cadence, schema, idempotency key, failure handling and evidence that a replay will not silently lose or double-count records.

Separate parsing, validation, enrichment and output writing so a failed step can be isolated. For a Glue job, examine what a bookmark tracks and what it does not: bookmark state applies to supported sources, while the output location still needs deliberate replay handling. Define a stable transformation context and decide whether a correction requires a new target, a controlled overwrite or a merge.

Reviewable practice outputDraw an incremental job with its source, bookmark state, transformations and destination. Describe what happens on a second run, a failed run and a historical backfill.

Express job dependencies and retry behavior explicitly. Compare a time-based schedule, a source event and a workflow or state machine against the arrival contract. Avoid starting downstream publication before the upstream validation has passed. Record a dead-letter, retry or manual-review path for failed events and choose notifications that include enough context to investigate safely.

Reviewable practice outputPrepare a workflow diagram with trigger, stages, retry limit, failure branch, owner and a test that proves a failed stage cannot publish unvalidated output.

Compare object storage, operational databases and analytical stores from the reader's access pattern rather than the source format alone. Consider transaction needs, update rate, query concurrency, latency, file format and partitioning. Review when an open table format or vector index is relevant instead of assuming every dataset needs one. A catalog makes data discoverable but does not clean bad records or guarantee a useful table design. Document where schema changes are reviewed before they reach consumers.

Reviewable practice outputBuild a store decision table for raw events, daily analytical reports and a low-latency lookup. Include one rejected option and a query or operational reason for rejecting it.

Specify column meanings, nullable fields and evolution rules before a crawler or consumer discovers a table. Distinguish table metadata from the underlying S3 objects. Choose retention, storage classes and deletion controls from the data's value and policy requirements. Check how a partition or small-file strategy affects reads and costs instead of assuming more partitions are always better.

Reviewable practice outputCreate a sample dataset contract with schema version, catalog owner, partition choice, retention policy and an example of a change that needs a compatibility review.

Define measurable checks for completeness, uniqueness, referential consistency and freshness at the appropriate stage. Decide whether a bad record is quarantined, repaired or causes the batch to fail. Distinguish data-quality results from infrastructure-health metrics. Capture enough sample identifiers to reproduce a defect without exposing sensitive personal data in logs or alerts.

Reviewable practice outputWrite three quality rules for a fictional order dataset, with a threshold, failed-record route, owner and evidence that the published table excludes unacceptable data.

Name the human or workload principal and distinguish IAM permissions, S3 resource access, KMS key authorization and Lake Formation or catalog permissions where they apply. Least privilege should match the dataset and action, not grant broad lake access. Check encryption, network reachability, audit trails and privacy boundaries separately; fixing one layer does not automatically repair the others.

Reviewable practice outputDraw a permission trace for a reporting role reading one governed table. Explain which principal, catalog object, S3 location and key are involved and what another role must still be unable to read.

Monitor source lag, job failures, validation results, query latency and spend as separate signals. Start optimization with the observed bottleneck: data skew, excess small files, read pattern, processing capacity or avoidable scans. Rehearse replay and restore procedures with sample data. After each practice set, explain the service trade-off and use the current official exam guide to revisit weak domains.

Reviewable practice outputProduce a short data-product runbook: owner, source contract, freshness target, quality gate, access boundary, replay procedure, cost signal and a domain-tagged mistake log.