AWS Certified Solutions Architect - Associate SAA-C03 Practice Test 81
Question 800
A Firehose pipeline receives JSON application logs, but the downstream Athena queries repeatedly scan raw uncompressed lines. The analytics team wants columnar Parquet files in S3. Which transformation should be evaluated?
- Enable only an ALB HTTP listener and expect it to convert JSON records into Parquet.
- Change object names from `.json` to `.parquet` without encoding the actual content.
- Increase Kinesis shard retention while leaving the Firehose delivery payload raw JSON.
- Configure supported Firehose data-format conversion or transformations and a valid Glue schema/catalog arrangement before S3 delivery.
Explanation
Correct Answer
D:Firehose can convert supported record formats for efficient analytic storage when schemas and destination compatibility are configured.
Incorrect Answers
A:A load balancer does not transform streaming payload formats.
B:A file extension does not make bytes conform to Parquet columnar format.
C:Retention does not transform the stored object serialization.
Question 801
An analytics platform wants custom consumers to reread historical event offsets at different rates for real-time models. A proposal uses Firehose only because it delivers events to S3 with minimal code. Which distinction matters?
- Use Kinesis Data Streams or another replay-capable streaming log when consumers need controlled offsets and independent replay.
- Assume Firehose provides every consumer an indefinite seekable shard offset log identical to Kinesis Data Streams.
- Use an S3 lifecycle rule to simulate real-time consumer offsets for records still arriving.
- Change Firehose object prefix names and assume this implements low-latency ordered record replay across all consumers.
Explanation
Correct Answer
A:Data Streams provides shard records and consumer position semantics; Firehose is optimized for managed onward delivery to destinations.
Incorrect Answers
B:Delivery streams are not the same general-purpose multi-consumer replay abstraction.
C:Object transitions do not expose stream checkpoint and shard consumption semantics.
D:Destination object keys are not an offset-tracking streaming API.
Question 802
An MSK Kafka topic uses customer ID as the record key. One customer produces nearly all events and all of its writes bottleneck on one partition. Order among independent messages is not required. What should change?
- Force every producer to use the same customer key so Kafka can spread that key across all partitions automatically.
- Revisit partition count and key distribution or split independent hot-tenant work across partition keys.
- Increase topic retention without considering key distribution or broker load.
- Move the consumer group to another VPC subnet while producers keep sending the same hot key.
Explanation
Correct Answer
B:Kafka keys determine partition placement, and a hot key can limit one partition even if other brokers are idle.
Incorrect Answers
A:A fixed key normally hashes to one partition, reinforcing skew.
C:Retention changes history lifespan, not a hot partition's ingest rate.
D:Network placement does not alter the key-to-partition skew driving ingest.
Question 803
A Kafka-based billing journal must replay a week's events after an outage and tolerate one broker failure. Its current MSK topic has a short retention and replication factor one. Which design dimensions must change?
- Keep replication factor one and assume managed MSK permanently protects the only broker's partition bytes.
- Leave retention at one hour and assume consumers can replay events from five days ago.
- Set appropriate topic replication, minimum in-sync replica/ack behavior and retention for the recovery guarantees.
- Use a CloudFront cache to serve Kafka broker logs as the primary replicated queue.
Explanation
Correct Answer
C:Kafka log availability and replay depend on replicated partitions, producer durability settings and retained records.
Incorrect Answers
A:Managed cluster operation does not remove the failure risk of a single partition replica.
B:Expired Kafka records cannot be reconsumed through old offsets.
D:A CDN cache is not a Kafka replicated commit log.
Question 804
A large platform relies on separate Kafka consumer groups replaying historical records and preserving per-partition order. An architect proposes replacing the Kafka backbone with one SQS standard queue with no app redesign. What breaks?
- Assume SQS standard queues expose Kafka partitions and arbitrary offsets to every consumer group.
- Assume an SQS standard queue delivers strictly ordered once-only records to all independent consumer groups.
- Replace Kafka with one SQS FIFO queue and label workers as consumer groups without preserving distinct replay offsets for each group.
- Retain Kafka/MSK or redesign around another log with the needed partition, replay and consumer-group semantics.
Explanation
Correct Answer
D:A single SQS queue's delivery and retention model is not a transparent replacement for a seekable Kafka partitioned log.
Incorrect Answers
A:SQS and Kafka have distinct consumption and ordering/replay contracts.
B:Standard queues are at-least-once and can deliver out of order, and consumption is not Kafka-style group replay.
C:SQS FIFO preserves ordering within message groups but a consumed and deleted message is not independently replayable by arbitrarily many consumer groups using separate offsets. Kafka-style retained logs need a matching streaming design.
Question 805
A data lake ingests CSV from many departments into distinct S3 paths. Analysts cannot discover schemas consistently and keep hand-editing table definitions for every new batch. Which metadata service flow helps?
- Use AWS Glue crawlers or managed catalog registration to populate and maintain Data Catalog table schemas as appropriate.
- Create extra S3 buckets alone and assume every query engine automatically infers the correct schema and partitions.
- Use IAM password rotation to infer CSV column types across datasets.
- Add more EC2 instances to serve as hardcoded schema files without a central catalog.
Explanation
Correct Answer
A:Glue Data Catalog gives analytics services a shared schema/metastore rather than per-user ad hoc file definitions.
Incorrect Answers
B:Object storage does not itself establish a governed shared table metadata catalog.
C:Identity hygiene is separate from schema discovery and metadata curation.
D:Ad hoc server-side files perpetuate the discovery/consistency problem.
Question 806
A company repeatedly runs Athena queries that read only two fields from a 4-TB raw CSV lake. It pays for large scans even after indexing other systems. Which storage representation can improve I/O economics?
- Increase CSV text indentation and expect Athena to skip unused columns without reading their bytes.
- Convert reusable lake datasets to compressed columnar Parquet with suitable schema and partitioning.
- Store each CSV file in a separate S3 bucket and assume bucket count reduces query-scanned bytes.
- Partition the original row-oriented CSV solely by filename but continue requiring Athena to read every column of every matched CSV object.
Explanation
Correct Answer
B:Columnar formats permit engines to read fewer needed columns and often compress repetitive values effectively.
Incorrect Answers
A:Text CSV remains row-oriented; whitespace changes do not create column pruning.
C:Bucket names do not change the amount of data the engine must read.
D:Partition pruning can reduce the files scanned when predicates match partitions, but CSV remains row-oriented. For repeated queries selecting only two fields, compressed columnar Parquet permits column projection and typically reduces scanned bytes more directly.
Question 807
An S3 lake stores five years of device records in one unpartitioned prefix. Every query filters for a single day and customer but reads every historical file. What dataset organization should be tested?
- Combine all records into one giant unpartitioned CSV file to make every date filter faster.
- Refresh the AWS Glue Data Catalog crawler every hour but leave all records physically unpartitioned under the same S3 prefix.
- Partition data by frequently filtered dimensions and enable partition pruning in the query/catalog layer.
- Increase Lambda timeout for the query caller and accept the same excessive scan as optimal.
Explanation
Correct Answer
C:Partition pruning skips irrelevant object sets when the filter aligns with dataset partition keys.
Incorrect Answers
A:A monolithic file usually prevents skipping unrelated historical partitions.
B:Updated catalog metadata may describe the dataset accurately but will not by itself reorganize underlying files into date/customer partitions. Partition pruning depends on an appropriate physical organization and predicates.
D:Longer client waits do not improve storage scan efficiency.
Question 808
Analysts run the same deterministic Athena SELECT query against a read-only daily snapshot many times during a 45-minute reporting meeting. The data will not change during that meeting, and all callers use the same workgroup and have access to the prior S3 results. Partitioning and columnar formatting are already optimized; however, repeated executions still consume billed scanned bytes. What should the architect configure to avoid unnecessary re-execution while keeping results sufficiently fresh?
- Increase Athena workgroup bytes-scanned limits so each identical query can scan the full table more quickly without charges.
- Enable Glue crawler runs every five minutes, because continuously refreshing table metadata automatically caches every Athena SQL result.
- Upgrade the Athena query engine version and assume every previously run SELECT result is automatically reused for all workgroups regardless of configured age and permissions.
- Enable Athena query result reuse for eligible matching queries within the same workgroup and set a maximum reuse age covering the stable reporting window.
Explanation
Correct Answer
D:Athena can return a previous eligible SELECT result instead of rerunning the query when the workgroup, query/database/catalog configuration, permissions and result age match. The snapshot is stable for 45 minutes, so an appropriate reuse age saves repeated scans without violating the stated freshness requirement.
Incorrect Answers
A:A workgroup scan limit is a cost guardrail, not a result cache. Raising the limit does not eliminate scanned bytes or the charges from repeated SELECT executions.
B:Glue crawlers discover metadata and partitions. They do not cache Athena query execution results and can generate unnecessary catalog work for static data.
C:An Athena engine version alone does not enable result reuse without eligible query settings and matching workgroup, result location, access rights and age. The reuse option must be configured.
Question 809
An Athena table reads S3 logs stored under predictable year/month/day prefixes. A new day arrives every morning, but automation to add each partition to the Glue Data Catalog is fragile and creates query startup delays. Queries filter on dates. Which table configuration can infer the date partitions without registration?
- Configure Athena partition projection with an appropriate date range and S3 location template in the table properties.
- Convert each log file to Parquet but continue relying on manual registration of every date partition.
- Apply an S3 Intelligent-Tiering lifecycle configuration that automatically creates Glue partition objects.
- Use CloudFront invalidations to refresh AWS Glue partition metadata for each date prefix.
Explanation
Correct Answer
A:Athena partition projection calculates eligible partition values and paths at query time, avoiding Glue GetPartitions and daily registration when the layout is predictable.
Incorrect Answers
B:Parquet reduces scanned bytes for columnar queries but does not alone eliminate the need to track date partitions.
C:An S3 storage class does not register or project table metadata in the Glue Data Catalog.
D:CloudFront invalidations manage CDN caches, not Athena’s projected date-partition metadata.
Question 810
Multiple teams share an Athena workgroup. One accidentally submits a very expensive full-lake scan, and another tries to query sensitive columns outside its role. Which controls should be configured?
- Create separate Athena workgroups for each team but continue granting all groups unrestricted S3 and Lake Formation access to the sensitive tables.
- Use Athena workgroup query limits/cost controls with appropriate IAM, Lake Formation or data-access policies.
- Set a workgroup-wide bytes-scanned SNS alert while leaving the single-query scan cutoff and underlying data permissions unchanged.
- Enable query-result encryption and restrict the results bucket, but keep unrestricted table permissions and no scan limits for the analysts.
Explanation
Correct Answer
B:Workgroup settings can constrain query spending, while identity/catalog permissions determine what data a principal may read.
Incorrect Answers
A:Workgroup separation can isolate workloads and settings, but it cannot repair overbroad permissions to confidential data. The scenario also needs per-query scan limits or equivalent workgroup governance for expensive SQL requests.
C:An aggregate Athena workgroup alert signals threshold overruns; it is not the same as a per-query byte cutoff that cancels an oversized scan. Neither alert grants or denies access to sensitive columns, which requires data permissions.
D:Encryption and a private results bucket reduce result exposure, yet do not stop unauthorized source-table queries or bound scan-heavy SQL. Identity/data controls and bytes-scanned limits remain necessary.
