Amazon AWS Certified Solutions Architect - Associate SAA-C03 Practice Test 63

Question 620

A production PostgreSQL database must remain online while rows are replicated continuously into a new AWS-managed database after an initial load. Which migration capability supports this?

Use AWS Database Migration Service full load plus change data capture, with supported engine and replication settings.
Run DMS in full-load-only mode during a live migration and assume committed source writes after the snapshot are automatically replayed.
Take a one-time database snapshot after the initial migration and assume it stays in sync with continuing PostgreSQL writes.
Enable the target database Multi-AZ failover and treat it as change-data capture from the separate source database.

Explanation

Correct Answer

DMS CDC processes ongoing source changes after the baseline copy, reducing migration downtime.

Incorrect Answers

Full load alone does not replicate ongoing changes; DMS requires compatible CDC configuration to keep the destination synchronized.

A static snapshot does not deliver new commits. DMS with supported ongoing CDC replication can apply changes after initial load.

Multi-AZ protects availability of a target deployment; it does not replicate transactions from the independent live source without a migration/CDC process.

Question 621

A 120 GiB archive must be uploaded to Amazon S3 over a connection that drops periodically. Which transfer strategy handles large data and retransmits only failed pieces?

Upload the entire 120-GiB object as a single nonmultipart PutObject and expect failed attempts to resume from the last successful byte.
Use a single PutObject upload, raising the SDK socket timeout and retrying the complete 120-GiB payload whenever the connection drops.
Use S3 multipart upload with parallel appropriately sized parts and per-part retry/completion.
Split the 120-GiB object into separate final S3 keys but never issue CompleteMultipartUpload for one unified requested archive.

Explanation

Correct Answer

Multipart uploads allow failed parts to be retried without restarting the entire object and support large objects.

Incorrect Answers

Single-request object PUTs do not provide independent durable part checkpoints or part-specific retries on flaky connections.

A single S3 PUT cannot upload a 120-GiB object and full-payload retries are not the per-part resume mechanism provided by multipart upload.

Independent objects are not the same destination key: successful multipart part uploads must be completed with their ETags to produce one archive.

Question 622

A security analytics service consumes Kinesis events. A failed release may require reprocessing 36 hours of events without doubling downstream charges. Select THREE design elements.

Persist consumer checkpoints or sequence positions so the service can restart from the required point.
Use a Kinesis stream with 24-hour retention for a replay window of 36 hours and assume switching to TRIM_HORIZON restores expired records.
Disable checkpoints and always start at latest when investigating last-day errors.
Set Kinesis stream record retention to at least the business replay window with headroom.
Keep the default one-day retention but assume a 36-hour replay is always possible.
Make downstream effects idempotent or detect duplicate events during replay.

Explanation

Correct Answer

Offsets provide a reproducible starting position instead of accidentally skipping or rereading arbitrary portions.

Expired records cannot be reconstructed by moving an iterator; replay must fit within retention.

Reprocessing records can repeat side effects; stable event IDs or equivalent deduplication prevent duplicated billing or counts.

Incorrect Answers

Kinesis discards records older than configured retention; iterators cannot resurrect data already trimmed from the stream.

A latest-only iterator skips precisely the historical events needed for recovery.

Records outside the retention window may already have been trimmed.

Question 623

A pipeline must run one Glue job, wait for completion, validate a row count, then invoke a publish Lambda only on successful validation. Which orchestration service fits?

Start the Glue job and the publish Lambda in parallel without a success dependency, then trust the Lambda to read a complete output.
Use EventBridge to trigger Glue and publish concurrently without checking for Glue completion or validation success.
Use an S3 lifecycle transition after a fixed delay as the sole signal that the Glue job succeeded and row counts passed.
Use AWS Step Functions with Glue task integration, validation states and conditional next steps.

Explanation

Correct Answer

A state machine encodes data-job dependencies, retries and failure paths rather than relying on timing guesses.

Incorrect Answers

A successful start request is not successful job completion; publishing must wait for ETL completion and validation of produced data.

EventBridge can route events, but triggering both stages without a completion-and-validation dependency allows publication of incomplete output.

Lifecycle age-based transitions neither inspect Glue task status nor evaluate the required conditional data-quality result.

Question 624

Athena, Glue and Redshift Spectrum consumers must share names and schemas for tables over a partitioned S3 data lake. Which metadata facility fits?

Use the AWS Glue Data Catalog as a governed shared source of database, table, schema and partition metadata.
Create independent hard-coded table schemas in each Athena and Redshift Spectrum client, ignoring shared catalog locations and partition metadata.
Have each engine independently infer CSV column types on every query and assume the resulting incompatible schemas are centrally governed.
Grant every query user S3 ListBucket and assume object listing alone defines schemas and partitions for all three services.

Explanation

Correct Answer

Supported query services can reference Glue catalog definitions rather than maintain conflicting per-engine schemas.

Incorrect Answers

Divergent client schemas are brittle; a governed Glue Data Catalog enables shared schema and partition discovery.

Ad hoc client-specific inference duplicates schemas rather than sharing consistent Glue Data Catalog metadata across Athena, Glue and Spectrum.

Bucket listing reveals object keys, not the consistent table schema/partition metadata that the engines need.

Question 625

PutRecords calls begin to throttle after the camera fleet doubles traffic, while the application has an unchanged one-shard provisioned stream. What caused the failure?

One Kinesis shard is guaranteed to accept any record rate as long as the cumulative payload remains under 1 MiB per second.
Increase the stream's record-retention duration without increasing provisioned shard count and assume PutRecords write throughput automatically scales as a result.
Add another GetRecords consumer to the same one-shard stream and expect additional read processes to increase producer PutRecords capacity.
The stream exceeded the per-shard write throughput or record-rate ceiling.

Explanation

Correct Answer

Inspect WriteProvisionedThroughputExceeded and partition-key distribution; retry backoff alone cannot create shard capacity.

Incorrect Answers

A provisioned shard has both bytes-per-second and records-per-second write quotas; either can trigger throttling.

Record retention controls how long previously ingested data is available for replay; it does not increase the per-shard PutRecords byte/s or record/s write limits.

Consumers only retrieve records and cannot raise producer-side write throughput. For PutRecords in provisioned mode, per-shard ingress and hot partition-key distribution remain the relevant bottlenecks.

Question 626

Five standard GetRecords consumers contend for one shard’s shared read throughput; each becomes slower as consumers are added. Why?

Registering four ordinary polling consumers using the same shard iterator automatically creates dedicated enhanced fan-out channels.
Shared GetRecords readers collectively use one shard read quota rather than receiving independent enhanced fan-out lanes.
Increase shard retention from one day to seven days and expect each ordinary GetRecords consumer to receive a separate dedicated read lane.
Assign five consumer applications different IAM role names while keeping all of them on GetRecords polling, assuming role identity partitions the shard read quota.

Explanation

Correct Answer

Enhanced fan-out avoids per-consumer contention for the documented dedicated bandwidth, subject to consumer limits.

Incorrect Answers

Shared GetRecords polling remains subject to the common read limit; fan-out requires registering and using the EFO consumer API.

Extended retention affects replay history, not the shared polling throughput ceiling. Dedicated per-consumer throughput requires registering and consuming through enhanced fan-out.

IAM roles control API permissions but do not create separate shard read bandwidth. Ordinary polling consumers continue sharing the common read throughput irrespective of role naming.

Question 627

A team configures a Firehose S3 stream with a 5-minute buffer and expects each incoming event to create an immediate separate S3 object. Why are objects delayed and combined?

Firehose must commit each telemetry record as an individual S3 object immediately, regardless of buffer size and buffer interval configuration.
Set a custom S3 destination prefix for each telemetry record and assume unique key prefixes disable Firehose destination buffering.
Firehose buffers records based on size/time settings before writing batched objects to S3.
Increase the S3 destination buffer size while expecting every individual record to appear as an independent object sooner than the existing configured buffer interval.

Explanation

Correct Answer

It is a managed batched delivery path, not per-event synchronous S3 PutObject behavior.

Incorrect Answers

Amazon Data Firehose buffers records and delivers batched objects according to its configured hints; individual-record immediate S3 writes are not guaranteed.

An S3 prefix organizes delivered objects and may support dynamic partitioning; it does not guarantee immediate one-record-per-object delivery. Firehose still uses destination buffering behavior.

Larger buffers generally aggregate more records before delivery and can extend object arrival time at lower volume. Buffer thresholds are delivery hints, not a per-record synchronous guarantee.

Question 628

A Glue CSV-to-Parquet job writes files, but Athena still scans the original CSV path. What pipeline link was missed?

The transformed Parquet dataset is automatically selected by every existing Athena table simply because both datasets are in the same bucket.
The query/catalog table still points to the source CSV location rather than the transformed Parquet output.
Update only the Parquet object filenames to end in .parquet but leave the Athena table location and input serialization pointing to the raw CSV files.
Run a Glue crawler against the existing raw CSV source prefix again and expect it to change the destination of existing Athena queries to the separate new Parquet path.

Explanation

Correct Answer

Update the Glue Data Catalog metadata and validate queried partitions; writing Parquet alone does not redirect readers.

Incorrect Answers

Queries follow catalog location and partition metadata; an existing table can still read old CSV paths unless its data location is updated.

File extensions do not redirect an existing Athena table or change its catalog serialization. Its table location and schema/partition metadata must reference the transformed dataset.

A crawler on the old source may refresh CSV metadata but does not cause queries to a different curated location. Catalog table definition and partitions must be deliberately updated.

Question 629

A streaming Glue job is restarted after failure without a durable checkpoint directory and duplicates prior transformed outputs. What was missing?

Durable checkpoint/offset state and idempotent sink handling for replay after a restart.
A restarted streaming ETL job always starts at the latest stream offset, so missing unprocessed records is the correct way to guarantee no duplicates.
Increase a Kinesis retention period and assume that doubles every shard’s producer write capacity in both stream capacity modes.
Use one fixed partition key for all producers and assume on-demand mode guarantees no hot-partition throughput restrictions.

Explanation

Correct Answer

Streaming input delivery is not proof of exactly-once downstream side effects; recovery must track processed positions.

Incorrect Answers

Reliable recovery requires persisted source offsets and appropriately idempotent downstream effects; latest-only restarts can skip unprocessed events.

Retention affects the amount of data retained for replay but does not remove per-mode ingestion and service quota limits.

Even with suitable on-demand capacity, a single skewed key can concentrate traffic; a sound design must examine partition-key distribution and mode limits.

Question 630

An Athena query over logs has a restrictive WHERE date clause but still scans every month of S3 data. The table is unpartitioned. Why?

A partition predicate can prune data even when the Glue table has no registered partitions, no projected partitions and no partition folders.
Add a restrictive HAVING condition to the Athena query but leave the S3 source unpartitioned, assuming post-aggregation filters eliminate file scans.
Without usable partition metadata/pruning, the predicate may not eliminate whole S3 file groups from the scan.
Create a Glue crawler that infers the same unpartitioned table schema and assume refreshed column types alone permit date partition pruning.

Explanation

Correct Answer

Define partitions and appropriate Parquet layout; SQL filtering after full reads does not guarantee a low byte-scan charge.

Incorrect Answers

Athena needs effective partition metadata or projection for whole-partition pruning; ordinary row filters do not automatically avoid all file reads.

HAVING filters aggregation results, not the physical S3 object groups read to compute them. Partition layout and metadata are needed to prune months before scanning.

Schema discovery does not create meaningful date partitions when objects remain unpartitioned. Athena cannot skip whole S3 file groups based solely on a row-level date predicate.

Pass Your Amazon AWS Certified Solutions Architect - Associate SAA-C03 Exam Easy!

Question 620

Explanation

Correct Answer

Incorrect Answers

Question 621

Explanation

Correct Answer

Incorrect Answers

Question 622

Explanation

Correct Answer

Incorrect Answers

Question 623

Explanation

Correct Answer

Incorrect Answers

Question 624

Explanation

Correct Answer

Incorrect Answers

Question 625

Explanation

Correct Answer

Incorrect Answers

Question 626

Explanation

Correct Answer

Incorrect Answers

Question 627

Explanation

Correct Answer

Incorrect Answers

Question 628

Explanation

Correct Answer

Incorrect Answers

Question 629

Explanation

Correct Answer

Incorrect Answers

Question 630

Explanation

Correct Answer

Incorrect Answers

AWS Certified Solutions Architect - Associate SAA-C03 Practice Tests