Question 680
A large monolith must gradually move individual functions to managed services while retaining the existing public domain and supporting both old and new paths during the transition. What migration pattern reduces big-bang risk?
- Replace every public API route simultaneously with new services and remove access to the legacy monolith before validating the replacement paths.
- Disable all monitoring during migration so traffic-cutover regressions cannot stop service extraction.
- Apply a strangler pattern using API routing to shift selected capabilities behind stable interfaces incrementally.
- Keep all external traffic routed permanently to the original monolith, while developing new services that never receive production paths.
Explanation
Correct Answer
C:Incremental routing and service extraction allow tested functionality to move without cutting over the entire monolith at once.
Incorrect Answers
A:A big-bang switch defeats incremental strangler migration and increases rollback risk.
B:Removing telemetry conceals defects and makes staged rollout and rollback less reliable.
D:Without controlled routing transfer, the planned incremental extraction cannot retire old responsibilities or validate the new services.
Question 681
A new order service writes orders to a database and then emits an EventBridge event. Occasionally the DB commit succeeds but the process crashes before publishing, leaving consumers unaware of an accepted order. Which consistency pattern should the team implement?
- Use a transactional outbox or equivalent reliable event publication mechanism coupled to the committed business change.
- Publish an EventBridge event before committing the order database transaction and assume failed commits automatically retract the already sent event.
- Retry PutEvents only from the originating process memory after the database commit, without persisting an event outbox.
- Increase the CloudFront cache TTL on product images to guarantee atomic database and message-bus behavior.
Explanation
Correct Answer
A:An outbox recorded with the transaction can be delivered asynchronously, closing the gap between a durable DB commit and event publication.
Incorrect Answers
B:External event publication is not atomic with a later database commit, so this ordering can create ghost events.
C:A host crash loses the pending event even if the database commit succeeded, preserving the original dual-write failure.
D:CDN caching cannot atomically coordinate database persistence with downstream event delivery.
Question 682
An order event is already published to an SNS topic with separate SQS subscriptions for shipping, refunds and loyalty updates. Each consumer currently receives every event type and discards most messages after dequeueing; some teams cannot identify which message fields the filters apply to. Producers can reliably supply a `eventType` message attribute. What routing modernization reduces irrelevant deliveries without changing queue durability?
- Keep all SNS subscriptions unfiltered and have each queue worker discard unrelated messages after receipt, relying on local checks to reduce traffic delivered to SQS.
- Place `eventType` matching rules in each SQS dead-letter queue redrive policy, because redrive permissions determine which original SNS topics deliver normal events.
- Attach an SNS subscription filter policy for the required `eventType` values to each SQS subscription and use the appropriate MessageAttributes filter scope, while retaining existing queue consumers and failure handling.
- Put an `eventType` condition in each SQS queue resource policy and expect its permission evaluation to perform arbitrary SNS message-attribute content filtering.
Explanation
Correct Answer
C:SNS subscription filter policies evaluate message attributes or JSON message-body fields according to scope and deliver only matching events. Setting a filter for each existing SQS subscription reduces irrelevant work before enqueueing while preserving the independent durable queues. This tests filtering, not the earlier payment idempotency decision.
Incorrect Answers
A:Consumer-side filtering might avoid processing a wrong event type but does not reduce topic-to-queue deliveries, receive API load or unnecessary messages stored. Subscription filtering is evaluated before delivery and targets the requirement.
B:SQS redrive rules govern dead-letter handling or which source queues may use a DLQ, not predicate routing of SNS topic messages to ordinary SQS subscriptions. The filtering decision belongs at subscription delivery.
D:SQS queue policies authorize producer principals and queue operations; they are not SNS subscription content filters. Matching event attributes for routing must be configured on the SNS subscription.
Question 683
An EventBridge consumer had a parsing bug for two hours. Operations fixed the code and wants to replay only the affected business events back to the source event bus for controlled reprocessing, without replaying all events from producer databases. Which design is appropriate when established in advance?
- Create an EventBridge archive with a suitable event pattern and retention period, then initiate a scoped replay to its source bus and use idempotent downstream processing.
- Enable EventBridge archive only after the consumer bug is repaired and assume it contains events received before the archive existed.
- Use EventBridge dead-letter queue alone and assume it contains every event successfully delivered to the buggy consumer.
- Use a Route 53 health check to replay event messages from a bus without any archive or persisted source record.
Explanation
Correct Answer
A:EventBridge archives retain selected bus events for configured periods and allow time-scoped replay to the source bus; duplicates and non-guaranteed original order require replay-aware consumers.
Incorrect Answers
B:Archive capture starts when the archive is configured; earlier events cannot be reconstructed retroactively from an archive that did not exist.
C:A target DLQ typically stores delivery failures rather than every successfully delivered business event processed incorrectly downstream.
D:DNS endpoint health monitoring does not archive or replay domain events.
Question 684
A modern application accepts a long-running purchase review request. The review process sometimes pauses for human approval and must later resume with the exact stored workflow state, even if the original web server is gone. Which modernization is appropriate?
- Use Step Functions Express with a long-lived human approval wait lasting several months and assume that its short-duration execution model permits this.
- Store workflow progress only in the original API request handler memory and wait there for an approval event long after the client disconnected.
- Use an EBS provisioned IOPS increase as a substitute for approval orchestration and callback state.
- Use Step Functions with a durable state machine and approved human-approval callback or task-token pattern.
Explanation
Correct Answer
D:Durable workflow state and explicit wait/resume coordination allow long approval processes beyond a request thread or Lambda runtime.
Incorrect Answers
A:Express workflows have shorter execution limits and are not the right fit for extended human callback approval state.
B:Instance or handler memory is not durable externally resumed workflow state.
C:Faster block storage does not implement approval wait, resume or compensating workflow logic.
Question 685
A business wants to manage application authorization decisions for fine-grained resources using centralized policy evaluation rather than scattering custom policy logic across every microservice. Identity authentication already exists. Which AWS service could provide this application permissions layer?
- Use Cognito user authentication only and assume it independently resolves each application-specific relationship between principal, resource and action.
- Evaluate Amazon Verified Permissions with application policy stores, defined principals/resources/actions and tested authorization requests.
- Use only AWS Budgets action policy for every application record and treat cost thresholds as resource authorization.
- Use IAM resource policies for every business document regardless of whether the application resources are AWS IAM-addressable services.
Explanation
Correct Answer
B:Verified Permissions centralizes fine-grained application policy evaluation without replacing the identity authentication layer.
Incorrect Answers
A:Authentication identifies users but does not by itself provide the requested centralized application authorization policy decisions.
C:Budget policy is not a fine-grained application principal-resource-action authorization engine.
D:AWS IAM authorization is not automatically a custom application object-level policy evaluator.
Question 686
A shopping cart uses DynamoDB as the source of truth for cart state. Each cart is addressed by cartId, concurrent updates must be rejected when another worker advanced the version, and the cart should become unusable 24 hours after its latest valid interaction. Storage cleanup may happen asynchronously, but API reads must reject an expired cart immediately. What persistence design best satisfies the requirements?
- Store a cartId-keyed DynamoDB item, use conditional version checks, refresh a numeric Unix-epoch TTL attribute on each successful interaction, and reject logically expired carts on API reads while DynamoDB deletes them asynchronously.
- Enable DynamoDB TTL once when the cart is first created and never update its expiry time after activity, assuming that equals 24 hours after last interaction.
- Turn on DynamoDB TTL and rely solely on the service deleting every expired item at the exact expiration second before the application reads it.
- Use an unconditional PutItem for every cart write, and assume table-level on-demand capacity prevents lost updates from concurrent writers.
Explanation
Correct Answer
A:DynamoDB can conditionally enforce optimistic versions and store a refreshed epoch-seconds TTL, while application-side expiry checks are necessary because TTL garbage collection happens asynchronously, often days later.
Incorrect Answers
B:A TTL never refreshed after user activity implements time since creation, not the required inactivity-based expiry.
C:TTL is an asynchronous deletion mechanism, not an exact-second authorization or visibility boundary for expired items.
D:On-demand scaling affects request capacity, not concurrent write correctness or stale-version detection.
Question 687
An existing Amazon OpenSearch Service product index supports search correctly, but relevance tests fail for product titles: full-text queries must tokenize natural-language titles, while SKU and brand facet filters require exact-value matching and aggregations. The current mapping treats every string field identically, and the search team plans to correct the schema. Which mapping design should be used for the rebuilt index?
- Map every product title, SKU and brand as analyzed text and use match queries for exact SKU and brand facets, assuming analyzers preserve whole identifiers for exact aggregation.
- Map product title for analyzed `text` search and SKU/brand facet fields as `keyword` (or suitable multi-fields where both uses are required), then reindex and validate representative full-text searches and aggregations.
- Map all product fields only as unanalyzed `keyword`, since exact term matching automatically yields language-aware stemming, tokenization and phrase relevance for titles.
- Leave the existing OpenSearch mappings unchanged and move customer-facing text queries back to the transactional database, assuming SQL LIKE queries restore OpenSearch field-specific analyzers and facets.
Explanation
Correct Answer
B:Analyzed text mappings support full-text queries, while keyword mappings preserve whole values suitable for exact filtering and aggregations. Rebuilding and validating this mapping solves a search-schema error, distinct from another practice scenario search service selection, another practice scenario tenant security and another practice scenario alias-based rollout mechanics.
Incorrect Answers
A:Analyzed text is appropriate for relevance but may tokenize identifiers and brand values. Exact-match filters and stable aggregations need an unanalyzed keyword representation; one all-text mapping loses the requested distinction.
C:A keyword-only field treats the complete value as a term and is therefore useful for exact SKU or brand matches, not the analyzed language-aware title search requested. It sacrifices expected relevance behavior.
D:External SQL searches do not repair the search index's title analysis or exact faceting mappings, and they reintroduce query load on the transactional store. The requirement is to correct the existing search engine schema.
Question 688
An application is splitting order processing into two services. The primary service commits an order transaction in Aurora; a change-capture pipeline delivers database changes to downstream projections. A business dashboard must expose a denormalized read model without running expensive aggregate queries on the order writer. Which modernization pattern fits?
- Run all analytical dashboard queries directly against the Aurora order writer and increase writer size rather than maintaining the approved read projection.
- Use the transactional outbox only to publish events but never build a materialized read model, assuming event delivery alone makes dashboard joins fast.
- Build a CQRS-style read projection from committed change events into a query-optimized store and monitor lag, idempotency, and replay/reconciliation.
- Use a public S3 website endpoint as a transactional SQL replica that automatically joins every source table at request time.
Explanation
Correct Answer
C:CQRS separates the write model from a query-specific projection, trading bounded eventual consistency for isolation of dashboard read traffic; replay and deduplication protect correctness.
Incorrect Answers
A:This repeats the identified read/write coupling and does not create a query-optimized downstream model.
B:Outbox delivery addresses lost events, not the separate CQRS projection necessary for efficient denormalized reads.
D:S3 website hosting is object delivery rather than a live transactional relational projection with joins, replay and state management.
Question 689
A manufacturer already uses Amazon Kinesis Data Streams with per-device partition keys and correctly checkpointed independent consumers. Following a software defect, analysts discover that a historical event from four days ago is needed for replay, but the stream retained records for only the default 24 hours. The team plans for a future ten-day debugging and replay window without changing the streaming service. Which control should it adjust?
- Increase the number of consumer applications, assuming additional consumers restore records already expired from the stream.
- Enable enhanced fan-out on each consumer and assume throughput provisioning retroactively recovers data beyond the existing retention window.
- Change the stream’s shard partition keys after the loss and assume that reconstructs all expired device readings automatically.
- Increase Kinesis stream data retention to at least ten days (within supported limits), assess storage charges, and verify consumers can replay from a permitted earlier position.
Explanation
Correct Answer
D:Kinesis Data Streams supports configured retention beyond the default day (up to 365 days); increasing it before future incidents extends the valid replay window at an additional cost.
Incorrect Answers
A:More consumers cannot read data that has already expired from the configured stream retention period.
B:Enhanced fan-out addresses consumer read throughput isolation, not historical retention of expired records.
C:Partition-key behavior controls future routing and ordering, not persistence of stream records after their expiry.
Question 690
A subscription SaaS product accepts a cancellation request now but must send one reminder exactly seven days before the recorded end date, even if no application server is running at that time. The request may be rescheduled or canceled, and each reminder needs an audit trail. What managed design fits?
- Create a one-time Amazon EventBridge Scheduler schedule per subscription reminder, point it at an authorized target and update or delete it when the deadline changes.
- Use a recurring daily Scheduler rule for every subscription and let the reminder fire on all dates, ignoring the one-time end-date requirement.
- Configure a CloudFront cache invalidation and assume it invokes the reminder at the subscription deadline.
- Keep the reminder deadline only in an EC2 instance memory timer for the next year, assuming it survives instance replacement and rescheduling.
Explanation
Correct Answer
A:EventBridge Scheduler supports one-time time-based invocations with target authorization, configurable retry and schedule lifecycle; the application should handle idempotency and record delivery results.
Incorrect Answers
B:A recurring schedule does not correspond to a one-time reminder relative to each subscriber-specific termination date.
C:A CDN invalidation manages cached content, not arbitrary future business reminders and their cancellation state.
D:An in-process timer is not durable or centrally editable after instance failures or cancellation.
AWS Certified Solutions Architect - Professional SAP-C02 Practice Tests