Topic 08 Practice Test 1 covers FM APIs, Streaming, Async Processing, Resilience, Rate Limits, Routing, and Developer Tools in Domain 2 (Tasks 2.4-2.5). For broader exam preparation, review the AIP-C01 Exam Dumps. Every option includes focused technical reasoning tied to the tested decision.
Question 1
A application team must use one conversation-oriented interface across supported foundation models. Which approach most directly satisfies the stated requirement? Choose ONE.
- Bedrock Converse API
- Service Quotas monitoring and increase request
- structured response validation
- dead-letter queue for exhausted asynchronous work
Correct Answer: A
Correct Answer
Answer A is correct because Bedrock Converse API is correct because it provides a consistent message-based request and response structure for supported Bedrock models. The stem needs the team to use one conversation-oriented interface across supported foundation models. That fit makes Bedrock Converse API meet the constraint; alternatives differ materially.
Incorrect Answers
Answer B is incorrect because Use Service Quotas monitoring and increase request to address sustained demand that legitimately exceeds an account inference quota. Its mechanism measures quota consumption and requests a higher supported limit when usage growth is persistent. The current scenario requires Bedrock Converse API; Service Quotas monitoring and increase request solves a different design issue.
Answer C is incorrect because structured response validation works by parses and validates expected fields or schema and rejects malformed output before business logic executes. It fits where generated content feeds software that requires predictable machine-readable structure. The required function here is Bedrock Converse API, so this option targets the wrong condition.
Answer D is incorrect because dead-letter queue for exhausted asynchronous work belongs where one bad request must not block continuous processing or retry forever. It helps to isolate messages that repeatedly fail after bounded retries. The stem instead requires Bedrock Converse API, so dead-letter queue for exhausted asynchronous work does not meet the decisive condition.
Question 2
A enterprise architect must send a direct model-specific inference request when the native model payload is required. Which approach most directly satisfies the stated requirement? Choose ONE.
- cross-region inference profile
- SQS buffering for burst smoothing
- InvokeModel API
- Bedrock Converse API
Correct Answer: C
Correct Answer
Answer C is correct because InvokeModel API is correct because it invokes a selected model through Bedrock Runtime using that model’s request schema. The stem needs the team to send a direct model-specific inference request when the native model payload is required. That fit makes InvokeModel API meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because cross-region inference profile is meant to route supported Bedrock inference across eligible Regions for additional capacity and resilience. Use it when the workload needs managed cross-Region inference routing rather than a single fixed Region endpoint. This stem calls for InvokeModel API; cross-region inference profile addresses another operational need.
Answer B is incorrect because SQS buffering for burst smoothing can be appropriate because it queues work durably so consumers can process it with bounded concurrency. Apply it where traffic arrives in bursts that exceed instantaneous inference throughput but can be processed asynchronously. This case depends on InvokeModel API, leaving SQS buffering for burst smoothing mismatched to the requirement.
Answer D is incorrect because The role of Bedrock Converse API is to use one conversation-oriented interface across supported foundation models. It is useful when an application wants portable multi-turn interaction without model-specific payload formats. Here the design needs InvokeModel API; this choice instead solves an adjacent problem.
Question 3
A site reliability team must show partial generated output to the user before the full response is complete. Which approach most directly satisfies the stated requirement? Choose ONE.
- response streaming
- model fallback policy
- bounded worker concurrency
- InvokeModel API
Correct Answer: A
Correct Answer
Answer A is correct because response streaming is correct because it returns generated chunks progressively so perceived latency is reduced for long outputs. The stem needs the team to show partial generated output to the user before the full response is complete. That fit makes response streaming meet the constraint; alternatives differ materially.
Incorrect Answers
Answer B is incorrect because model fallback policy belongs where the application can tolerate a different approved model to preserve availability. It helps to continue service when the preferred model is unavailable or unsuitable for a request class. The stem instead requires response streaming, so model fallback policy does not meet the decisive condition.
Answer C is incorrect because The role of bounded worker concurrency is to prevent a worker fleet from overwhelming Bedrock Runtime or downstream systems. It is useful when too many parallel workers create avoidable throttling and contention. Here the design needs response streaming; this choice instead solves an adjacent problem.
Answer D is incorrect because InvokeModel API can be appropriate because it invokes a selected model through Bedrock Runtime using that model’s request schema. Apply it where the application needs explicit control of a model-specific inference body. This case depends on response streaming, leaving InvokeModel API mismatched to the requirement.
Question 4
A security engineering group must decouple long-running generation from an interactive request path. Which approach most directly satisfies the stated requirement? Choose ONE.
- circuit breaker around repeated failures
- request classification before model routing
- response streaming
- asynchronous inference workflow
Correct Answer: D
Correct Answer
Answer D is correct because asynchronous inference workflow is correct because it accepts work for background completion and lets the application track or retrieve results later. The stem needs the team to decouple long-running generation from an interactive request path. That fit makes asynchronous inference workflow meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because Use circuit breaker around repeated failures to stop hammering a failing dependency and allow recovery. Its mechanism opens after repeated failures, rejects or reroutes requests temporarily, then probes for recovery. The current scenario requires asynchronous inference workflow; circuit breaker around repeated failures solves a different design issue.
Answer B is incorrect because request classification before model routing is meant to choose the inference path from measurable task requirements. Use it when different request types should not all be forced through one expensive model. This stem calls for asynchronous inference workflow; request classification before model routing addresses another operational need.
Answer C is incorrect because The role of response streaming is to show partial generated output to the user before the full response is complete. It is useful when interactive applications benefit from displaying tokens or events as they arrive. Here the design needs asynchronous inference workflow; this choice instead solves an adjacent problem.
Question 5
A branch deployment team must recover safely from throttling and transient service errors without synchronized retry storms. Which approach most directly satisfies the stated requirement? Choose ONE.
- client timeout aligned with inference SLA
- CloudWatch throttling alarm for API clients
- exponential backoff with jitter
- asynchronous inference workflow
Correct Answer: C
Correct Answer
Answer C is correct because exponential backoff with jitter is correct because it increases retry delay over successive attempts and randomizes timing. The stem needs the team to recover safely from throttling and transient service errors without synchronized retry storms. That fit makes exponential backoff with jitter meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because client timeout aligned with inference SLA can be appropriate because it sets connection/read deadlines from observed model latency and the application service-level objective. Apply it where requests succeed on the service but clients cancel because timeout values are unrealistically short. This case depends on exponential backoff with jitter, leaving client timeout aligned with inference SLA mismatched to the requirement.
Answer B is incorrect because CloudWatch throttling alarm for API clients works by alarms on relevant client or service error metrics and triggers operational response. It fits where the team needs early warning that inference traffic is approaching a rate or concurrency boundary. The required function here is exponential backoff with jitter, so this option targets the wrong condition.
Answer D is incorrect because asynchronous inference workflow is meant to decouple long-running generation from an interactive request path. Use it when generation duration is incompatible with a synchronous web request. This stem calls for exponential backoff with jitter; asynchronous inference workflow addresses another operational need.
Question 6
A support organization must avoid duplicate business side effects when an inference workflow is retried. Which approach most directly satisfies the stated requirement? Choose ONE.
- prompt and output token budget
- correlation ID across asynchronous steps
- exponential backoff with jitter
- idempotency at the application layer
Correct Answer: D
Correct Answer
Answer D is correct because idempotency at the application layer is correct because it records a request or operation key so repeated processing can return the prior result or skip the side effect. The stem needs the team to avoid duplicate business side effects when an inference workflow is retried. That fit makes idempotency at the application layer meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because prompt and output token budget can be appropriate because it limits context and requested output tokens so inference stays within model and application constraints. Apply it where oversized prompts or unconstrained output are causing latency, cost, or context-window pressure. This case depends on idempotency at the application layer, leaving prompt and output token budget mismatched to the requirement.
Answer B is incorrect because correlation ID across asynchronous steps is meant to trace one logical request through queues, workers, retries, and model calls. Use it when failures are hard to diagnose because one user request fans out through multiple services. This stem calls for idempotency at the application layer; correlation ID across asynchronous steps addresses another operational need.
Answer C is incorrect because Use exponential backoff with jitter to recover safely from throttling and transient service errors without synchronized retry storms. Its mechanism increases retry delay over successive attempts and randomizes timing. The current scenario requires idempotency at the application layer; exponential backoff with jitter solves a different design issue.
Question 7
A cloud engineering team must use supported client retry behavior for transient Bedrock Runtime failures. Which approach most directly satisfies the stated requirement? Choose ONE.
- structured response validation
- AWS SDK retry configuration
- dead-letter queue for exhausted asynchronous work
- idempotency at the application layer
Correct Answer: B
Correct Answer
Answer B is correct because AWS SDK retry configuration is correct because it applies SDK retry modes, bounded attempts, and backoff rather than hand-coded tight loops. The stem needs the team to use supported client retry behavior for transient Bedrock Runtime failures. That fit makes AWS SDK retry configuration meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because structured response validation works by parses and validates expected fields or schema and rejects malformed output before business logic executes. It fits where generated content feeds software that requires predictable machine-readable structure. The required function here is AWS SDK retry configuration, so this option targets the wrong condition.
Answer C is incorrect because dead-letter queue for exhausted asynchronous work belongs where one bad request must not block continuous processing or retry forever. It helps to isolate messages that repeatedly fail after bounded retries. The stem instead requires AWS SDK retry configuration, so dead-letter queue for exhausted asynchronous work does not meet the decisive condition.
Answer D is incorrect because Use idempotency at the application layer to avoid duplicate business side effects when an inference workflow is retried. Its mechanism records a request or operation key so repeated processing can return the prior result or skip the side effect. The current scenario requires AWS SDK retry configuration; idempotency at the application layer solves a different design issue.
Question 8
A infrastructure team must address sustained demand that legitimately exceeds an account inference quota. Which approach most directly satisfies the stated requirement? Choose ONE.
- SQS buffering for burst smoothing
- Bedrock Converse API
- AWS SDK retry configuration
- Service Quotas monitoring and increase request
Correct Answer: D
Correct Answer
Answer D is correct because Service Quotas monitoring and increase request is correct because it measures quota consumption and requests a higher supported limit when usage growth is persistent. The stem needs the team to address sustained demand that legitimately exceeds an account inference quota. That fit makes Service Quotas monitoring and increase request meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because SQS buffering for burst smoothing can be appropriate because it queues work durably so consumers can process it with bounded concurrency. Apply it where traffic arrives in bursts that exceed instantaneous inference throughput but can be processed asynchronously. This case depends on Service Quotas monitoring and increase request, leaving SQS buffering for burst smoothing mismatched to the requirement.
Answer B is incorrect because The role of Bedrock Converse API is to use one conversation-oriented interface across supported foundation models. It is useful when an application wants portable multi-turn interaction without model-specific payload formats. Here the design needs Service Quotas monitoring and increase request; this choice instead solves an adjacent problem.
Answer C is incorrect because AWS SDK retry configuration is meant to use supported client retry behavior for transient Bedrock Runtime failures. Use it when the application uses an AWS SDK and needs standardized handling of retryable API errors. This stem calls for Service Quotas monitoring and increase request; AWS SDK retry configuration addresses another operational need.
Question 9
A architecture review board must route supported Bedrock inference across eligible Regions for additional capacity and resilience. Which approach most directly satisfies the stated requirement? Choose ONE.
- bounded worker concurrency
- cross-region inference profile
- InvokeModel API
- Service Quotas monitoring and increase request
Correct Answer: B
Correct Answer
Answer B is correct because cross-region inference profile is correct because it uses a profile that can distribute requests to supported destination Regions. The stem needs the team to route supported Bedrock inference across eligible Regions for additional capacity and resilience. That fit makes cross-region inference profile meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because The role of bounded worker concurrency is to prevent a worker fleet from overwhelming Bedrock Runtime or downstream systems. It is useful when too many parallel workers create avoidable throttling and contention. Here the design needs cross-region inference profile; this choice instead solves an adjacent problem.
Answer C is incorrect because InvokeModel API can be appropriate because it invokes a selected model through Bedrock Runtime using that model’s request schema. Apply it where the application needs explicit control of a model-specific inference body. This case depends on cross-region inference profile, leaving InvokeModel API mismatched to the requirement.
Answer D is incorrect because Service Quotas monitoring and increase request is meant to address sustained demand that legitimately exceeds an account inference quota. Use it when throttling is caused by stable production demand rather than a short transient burst. This stem calls for cross-region inference profile; Service Quotas monitoring and increase request addresses another operational need.
Question 10
A production operations group must continue service when the preferred model is unavailable or unsuitable for a request class. Which approach most directly satisfies the stated requirement? Choose ONE.
- request classification before model routing
- response streaming
- model fallback policy
- cross-region inference profile
Correct Answer: C
Correct Answer
Answer C is correct because model fallback policy is correct because it selects an approved alternate model according to capability, quality, latency, or availability rules. The stem needs the team to continue service when the preferred model is unavailable or unsuitable for a request class. That fit makes model fallback policy meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because request classification before model routing is meant to choose the inference path from measurable task requirements. Use it when different request types should not all be forced through one expensive model. This stem calls for model fallback policy; request classification before model routing addresses another operational need.
Answer B is incorrect because The role of response streaming is to show partial generated output to the user before the full response is complete. It is useful when interactive applications benefit from displaying tokens or events as they arrive. Here the design needs model fallback policy; this choice instead solves an adjacent problem.
Answer D is incorrect because The role of cross-region inference profile is to route supported Bedrock inference across eligible Regions for additional capacity and resilience. It is useful when the workload needs managed cross-Region inference routing rather than a single fixed Region endpoint. Here the design needs model fallback policy; this choice instead solves an adjacent problem.
Question 11
A platform team must stop hammering a failing dependency and allow recovery. Which approach most directly satisfies the stated requirement? Choose ONE.
- circuit breaker around repeated failures
- CloudWatch throttling alarm for API clients
- asynchronous inference workflow
- model fallback policy
Correct Answer: A
Correct Answer
Answer A is correct because circuit breaker around repeated failures is correct because it opens after repeated failures, rejects or reroutes requests temporarily, then probes for recovery. The stem needs the team to stop hammering a failing dependency and allow recovery. That fit makes circuit breaker around repeated failures meet the constraint; alternatives differ materially.
Incorrect Answers
Answer B is incorrect because CloudWatch throttling alarm for API clients works by alarms on relevant client or service error metrics and triggers operational response. It fits where the team needs early warning that inference traffic is approaching a rate or concurrency boundary. The required function here is circuit breaker around repeated failures, so this option targets the wrong condition.
Answer C is incorrect because asynchronous inference workflow is meant to decouple long-running generation from an interactive request path. Use it when generation duration is incompatible with a synchronous web request. This stem calls for circuit breaker around repeated failures; asynchronous inference workflow addresses another operational need.
Answer D is incorrect because model fallback policy works by selects an approved alternate model according to capability, quality, latency, or availability rules. It fits where the application can tolerate a different approved model to preserve availability. The required function here is circuit breaker around repeated failures, so this option targets the wrong condition.
Question 12
A network operations group must avoid abandoning valid requests too early while still bounding user wait time. Which approach most directly satisfies the stated requirement? Choose ONE.
- correlation ID across asynchronous steps
- exponential backoff with jitter
- circuit breaker around repeated failures
- client timeout aligned with inference SLA
Correct Answer: D
Correct Answer
Answer D is correct because client timeout aligned with inference SLA is correct because it sets connection/read deadlines from observed model latency and the application service-level objective. The stem needs the team to avoid abandoning valid requests too early while still bounding user wait time. That fit makes client timeout aligned with inference SLA meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because correlation ID across asynchronous steps is meant to trace one logical request through queues, workers, retries, and model calls. Use it when failures are hard to diagnose because one user request fans out through multiple services. This stem calls for client timeout aligned with inference SLA; correlation ID across asynchronous steps addresses another operational need.
Answer B is incorrect because Use exponential backoff with jitter to recover safely from throttling and transient service errors without synchronized retry storms. Its mechanism increases retry delay over successive attempts and randomizes timing. The current scenario requires client timeout aligned with inference SLA; exponential backoff with jitter solves a different design issue.
Answer C is incorrect because circuit breaker around repeated failures is meant to stop hammering a failing dependency and allow recovery. Use it when a downstream inference or tool endpoint is persistently failing rather than intermittently throttled. This stem calls for client timeout aligned with inference SLA; circuit breaker around repeated failures addresses another operational need.
Question 13
A application team must bound latency and cost by controlling request and response size. Which approach most directly satisfies the stated requirement? Choose ONE.
- prompt and output token budget
- dead-letter queue for exhausted asynchronous work
- idempotency at the application layer
- client timeout aligned with inference SLA
Correct Answer: A
Correct Answer
Answer A is correct because prompt and output token budget is correct because it limits context and requested output tokens so inference stays within model and application constraints. The stem needs the team to bound latency and cost by controlling request and response size. That fit makes prompt and output token budget meet the constraint; alternatives differ materially.
Incorrect Answers
Answer B is incorrect because dead-letter queue for exhausted asynchronous work belongs where one bad request must not block continuous processing or retry forever. It helps to isolate messages that repeatedly fail after bounded retries. The stem instead requires prompt and output token budget, so dead-letter queue for exhausted asynchronous work does not meet the decisive condition.
Answer C is incorrect because Use idempotency at the application layer to avoid duplicate business side effects when an inference workflow is retried. Its mechanism records a request or operation key so repeated processing can return the prior result or skip the side effect. The current scenario requires prompt and output token budget; idempotency at the application layer solves a different design issue.
Answer D is incorrect because client timeout aligned with inference SLA belongs where requests succeed on the service but clients cancel because timeout values are unrealistically short. It helps to avoid abandoning valid requests too early while still bounding user wait time. The stem instead requires prompt and output token budget, so client timeout aligned with inference SLA does not meet the decisive condition.
Question 14
A enterprise architect must treat model output as untrusted data before downstream automation uses it. Which approach most directly satisfies the stated requirement? Choose ONE.
- Bedrock Converse API
- structured response validation
- AWS SDK retry configuration
- prompt and output token budget
Correct Answer: B
Correct Answer
Answer B is correct because structured response validation is correct because it parses and validates expected fields or schema and rejects malformed output before business logic executes. The stem needs the team to treat model output as untrusted data before downstream automation uses it. That fit makes structured response validation meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because The role of Bedrock Converse API is to use one conversation-oriented interface across supported foundation models. It is useful when an application wants portable multi-turn interaction without model-specific payload formats. Here the design needs structured response validation; this choice instead solves an adjacent problem.
Answer C is incorrect because AWS SDK retry configuration is meant to use supported client retry behavior for transient Bedrock Runtime failures. Use it when the application uses an AWS SDK and needs standardized handling of retryable API errors. This stem calls for structured response validation; AWS SDK retry configuration addresses another operational need.
Answer D is incorrect because prompt and output token budget belongs where oversized prompts or unconstrained output are causing latency, cost, or context-window pressure. It helps to bound latency and cost by controlling request and response size. The stem instead requires structured response validation, so prompt and output token budget does not meet the decisive condition.
Question 15
A site reliability team must absorb request spikes before workers call the model at a controlled rate. Which approach most directly satisfies the stated requirement? Choose ONE.
- InvokeModel API
- Service Quotas monitoring and increase request
- SQS buffering for burst smoothing
- structured response validation
Correct Answer: C
Correct Answer
Answer C is correct because SQS buffering for burst smoothing is correct because it queues work durably so consumers can process it with bounded concurrency. The stem needs the team to absorb request spikes before workers call the model at a controlled rate. That fit makes SQS buffering for burst smoothing meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because InvokeModel API can be appropriate because it invokes a selected model through Bedrock Runtime using that model’s request schema. Apply it where the application needs explicit control of a model-specific inference body. This case depends on SQS buffering for burst smoothing, leaving InvokeModel API mismatched to the requirement.
Answer B is incorrect because Service Quotas monitoring and increase request is meant to address sustained demand that legitimately exceeds an account inference quota. Use it when throttling is caused by stable production demand rather than a short transient burst. This stem calls for SQS buffering for burst smoothing; Service Quotas monitoring and increase request addresses another operational need.
Answer D is incorrect because structured response validation can be appropriate because it parses and validates expected fields or schema and rejects malformed output before business logic executes. Apply it where generated content feeds software that requires predictable machine-readable structure. This case depends on SQS buffering for burst smoothing, leaving structured response validation mismatched to the requirement.
Question 16
A security engineering group must prevent a worker fleet from overwhelming Bedrock Runtime or downstream systems. Which approach most directly satisfies the stated requirement? Choose ONE.
- response streaming
- cross-region inference profile
- SQS buffering for burst smoothing
- bounded worker concurrency
Correct Answer: D
Correct Answer
Answer D is correct because bounded worker concurrency is correct because it caps simultaneous inference operations and scales within known service and cost limits. The stem needs the team to prevent a worker fleet from overwhelming Bedrock Runtime or downstream systems. That fit makes bounded worker concurrency meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because The role of response streaming is to show partial generated output to the user before the full response is complete. It is useful when interactive applications benefit from displaying tokens or events as they arrive. Here the design needs bounded worker concurrency; this choice instead solves an adjacent problem.
Answer B is incorrect because The role of cross-region inference profile is to route supported Bedrock inference across eligible Regions for additional capacity and resilience. It is useful when the workload needs managed cross-Region inference routing rather than a single fixed Region endpoint. Here the design needs bounded worker concurrency; this choice instead solves an adjacent problem.
Answer C is incorrect because SQS buffering for burst smoothing belongs where traffic arrives in bursts that exceed instantaneous inference throughput but can be processed asynchronously. It helps to absorb request spikes before workers call the model at a controlled rate. The stem instead requires bounded worker concurrency, so SQS buffering for burst smoothing does not meet the decisive condition.
Question 17
A branch deployment team must choose the inference path from measurable task requirements. Which approach most directly satisfies the stated requirement? Choose ONE.
- asynchronous inference workflow
- request classification before model routing
- model fallback policy
- bounded worker concurrency
Correct Answer: B
Correct Answer
Answer B is correct because request classification before model routing is correct because it classifies requests by capability, latency, risk, or context needs before selecting an approved model or workflow. The stem needs the team to choose the inference path from measurable task requirements. That fit makes request classification before model routing meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because asynchronous inference workflow is meant to decouple long-running generation from an interactive request path. Use it when generation duration is incompatible with a synchronous web request. This stem calls for request classification before model routing; asynchronous inference workflow addresses another operational need.
Answer C is incorrect because model fallback policy works by selects an approved alternate model according to capability, quality, latency, or availability rules. It fits where the application can tolerate a different approved model to preserve availability. The required function here is request classification before model routing, so this option targets the wrong condition.
Answer D is incorrect because Use bounded worker concurrency to prevent a worker fleet from overwhelming Bedrock Runtime or downstream systems. Its mechanism caps simultaneous inference operations and scales within known service and cost limits. The current scenario requires request classification before model routing; bounded worker concurrency solves a different design issue.
Question 18
A support organization must detect sustained rate-limit pressure before user-visible failures grow. Which approach most directly satisfies the stated requirement? Choose ONE.
- exponential backoff with jitter
- CloudWatch throttling alarm for API clients
- circuit breaker around repeated failures
- request classification before model routing
Correct Answer: B
Correct Answer
Answer B is correct because CloudWatch throttling alarm for API clients is correct because it alarms on relevant client or service error metrics and triggers operational response. The stem needs the team to detect sustained rate-limit pressure before user-visible failures grow. That fit makes CloudWatch throttling alarm for API clients meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because Use exponential backoff with jitter to recover safely from throttling and transient service errors without synchronized retry storms. Its mechanism increases retry delay over successive attempts and randomizes timing. The current scenario requires CloudWatch throttling alarm for API clients; exponential backoff with jitter solves a different design issue.
Answer C is incorrect because circuit breaker around repeated failures is meant to stop hammering a failing dependency and allow recovery. Use it when a downstream inference or tool endpoint is persistently failing rather than intermittently throttled. This stem calls for CloudWatch throttling alarm for API clients; circuit breaker around repeated failures addresses another operational need.
Answer D is incorrect because The role of request classification before model routing is to choose the inference path from measurable task requirements. It is useful when different request types should not all be forced through one expensive model. Here the design needs CloudWatch throttling alarm for API clients; this choice instead solves an adjacent problem.
Question 19
A cloud engineering team must trace one logical request through queues, workers, retries, and model calls. Which approach most directly satisfies the stated requirement? Choose ONE.
- idempotency at the application layer
- client timeout aligned with inference SLA
- correlation ID across asynchronous steps
- CloudWatch throttling alarm for API clients
Correct Answer: C
Correct Answer
Answer C is correct because correlation ID across asynchronous steps is correct because it propagates a stable request identifier in logs and messages across each component. The stem needs the team to trace one logical request through queues, workers, retries, and model calls. That fit makes correlation ID across asynchronous steps meet the constraint; alternatives differ materially.
Incorrect Answers
Answer A is incorrect because Use idempotency at the application layer to avoid duplicate business side effects when an inference workflow is retried. Its mechanism records a request or operation key so repeated processing can return the prior result or skip the side effect. The current scenario requires correlation ID across asynchronous steps; idempotency at the application layer solves a different design issue.
Answer B is incorrect because client timeout aligned with inference SLA belongs where requests succeed on the service but clients cancel because timeout values are unrealistically short. It helps to avoid abandoning valid requests too early while still bounding user wait time. The stem instead requires correlation ID across asynchronous steps, so client timeout aligned with inference SLA does not meet the decisive condition.
Answer D is incorrect because CloudWatch throttling alarm for API clients can be appropriate because it alarms on relevant client or service error metrics and triggers operational response. Apply it where the team needs early warning that inference traffic is approaching a rate or concurrency boundary. This case depends on correlation ID across asynchronous steps, leaving CloudWatch throttling alarm for API clients mismatched to the requirement.
Question 20
A infrastructure team must isolate messages that repeatedly fail after bounded retries. Which approach most directly satisfies the stated requirement? Choose ONE.
- dead-letter queue for exhausted asynchronous work
- AWS SDK retry configuration
- prompt and output token budget
- correlation ID across asynchronous steps
Correct Answer: A
Correct Answer
Answer A is correct because dead-letter queue for exhausted asynchronous work is correct because it moves poison or persistently failing work out of the main queue for inspection and replay. The stem needs the team to isolate messages that repeatedly fail after bounded retries. That fit makes dead-letter queue for exhausted asynchronous work meet the constraint; alternatives differ materially.
Incorrect Answers
Answer B is incorrect because AWS SDK retry configuration is meant to use supported client retry behavior for transient Bedrock Runtime failures. Use it when the application uses an AWS SDK and needs standardized handling of retryable API errors. This stem calls for dead-letter queue for exhausted asynchronous work; AWS SDK retry configuration addresses another operational need.
Answer C is incorrect because prompt and output token budget belongs where oversized prompts or unconstrained output are causing latency, cost, or context-window pressure. It helps to bound latency and cost by controlling request and response size. The stem instead requires dead-letter queue for exhausted asynchronous work, so prompt and output token budget does not meet the decisive condition.
Answer D is incorrect because The role of correlation ID across asynchronous steps is to trace one logical request through queues, workers, retries, and model calls. It is useful when failures are hard to diagnose because one user request fans out through multiple services. Here the design needs dead-letter queue for exhausted asynchronous work; this choice instead solves an adjacent problem.
