Amazon AWS Certified Solutions Architect - Associate SAA-C03 Bundle

The option Enforce scheduled stop/scale-in with environment exceptions addresses a separate condition: a dev-test environment runs 24/7 without any activity overnight. Nonproduction scheduling cuts unused compute hours. The actual requirement is an autoscaling group runs too many hosts because scaling targets use an unrelated metric. Recommended corrective action: Tune scaling around real workload demand and performance objectives. Misaligned scaling signals cause excess capacity.

For the alternative failure cost Explorer shows a consistent compute floor across the past year, the control Match eligible committed Savings Plans to measured baseline usage is relevant: commitments suit predictable long-running usage. Here the evidence instead is an autoscaling group runs too many hosts because scaling targets use an unrelated metric. The selected technical mechanism is Tune scaling around real workload demand and performance objectives: misaligned scaling signals cause excess capacity.

Prefer flexible Compute Savings Plans over narrow instance-family commitments solves a different operational problem where a service frequently changes regions and instance types due to expansion. Flexible commitments accommodate changing deployment footprints. It does not directly address an autoscaling group runs too many hosts because scaling targets use an unrelated metric. The case calls for Tune scaling around real workload demand and performance objectives, since misaligned scaling signals cause excess capacity.

For the alternative failure a scheduled container needs privileged OS customization unavailable to Fargate, the control Use EC2-backed ECS or native EC2 with carefully sized instances is relevant: runtime requirements determine whether managed tasks are suitable. Here the evidence instead is traffic is near zero most nights yet EC2 hosts are kept for a lightweight REST API. The selected technical mechanism is Use API Gateway plus Lambda when function limits permit: invocation-based service can avoid idle fleet spend.

Run on scheduled serverless/batch service appropriate to runtime solves a different operational problem where an interactive report runs every hour for five minutes and otherwise needs no server. Scheduled pay-as-used resources avoid 55 minutes idle. It does not directly address traffic is near zero most nights yet EC2 hosts are kept for a lightweight REST API. The case calls for Use API Gateway plus Lambda when function limits permit, since invocation-based service can avoid idle fleet spend.

The competing option Use AWS Batch/Glue managed capacity for the bounded job would be useful when a data engineer needs parallel one-off ETL across thousands of records. Managed job services provision for processing periods. The present issue is traffic is near zero most nights yet EC2 hosts are kept for a lightweight REST API; it is more directly addressed by Use API Gateway plus Lambda when function limits permit. Invocation-based service can avoid idle fleet spend.

Use EC2-backed ECS or native EC2 with carefully sized instances solves a different operational problem where a scheduled container needs privileged OS customization unavailable to Fargate. Runtime requirements determine whether managed tasks are suitable. It does not directly address a streaming worker continuously processes high-throughput events 24/7. The case calls for Use appropriately sized ECS/EC2 workers rather than assuming Lambda is cheapest, since steady sustained throughput can favor provisioned processing.

The competing option Run on scheduled serverless/batch service appropriate to runtime would be useful when an interactive report runs every hour for five minutes and otherwise needs no server. Scheduled pay-as-used resources avoid 55 minutes idle. The present issue is a streaming worker continuously processes high-throughput events 24/7; it is more directly addressed by Use appropriately sized ECS/EC2 workers rather than assuming Lambda is cheapest. Steady sustained throughput can favor provisioned processing.

The option Use AWS Batch/Glue managed capacity for the bounded job addresses a separate condition: a data engineer needs parallel one-off ETL across thousands of records. Managed job services provision for processing periods. The actual requirement is a streaming worker continuously processes high-throughput events 24/7. Recommended corrective action: Use appropriately sized ECS/EC2 workers rather than assuming Lambda is cheapest. Steady sustained throughput can favor provisioned processing.

Use API Gateway plus Lambda when function limits permit solves a different operational problem where traffic is near zero most nights yet EC2 hosts are kept for a lightweight REST API. Invocation-based service can avoid idle fleet spend. It does not directly address a data engineer needs parallel one-off ETL across thousands of records. The case calls for Use AWS Batch/Glue managed capacity for the bounded job, since managed job services provision for processing periods.

The option Use EC2-backed ECS or native EC2 with carefully sized instances addresses a separate condition: a scheduled container needs privileged OS customization unavailable to Fargate. Runtime requirements determine whether managed tasks are suitable. The actual requirement is a data engineer needs parallel one-off ETL across thousands of records. Recommended corrective action: Use AWS Batch/Glue managed capacity for the bounded job. Managed job services provision for processing periods.

Run on scheduled serverless/batch service appropriate to runtime would help if an interactive report runs every hour for five minutes and otherwise needs no server. Scheduled pay-as-used resources avoid 55 minutes idle. That is not the symptom described here: a data engineer needs parallel one-off ETL across thousands of records. The matching response is Use AWS Batch/Glue managed capacity for the bounded job, because managed job services provision for processing periods.

The competing option Use EC2-backed ECS or native EC2 with carefully sized instances would be useful when a scheduled container needs privileged OS customization unavailable to Fargate. Runtime requirements determine whether managed tasks are suitable. The present issue is an interactive report runs every hour for five minutes and otherwise needs no server; it is more directly addressed by Run on scheduled serverless/batch service appropriate to runtime. Scheduled pay-as-used resources avoid 55 minutes idle.

The option Use appropriately sized ECS/EC2 workers rather than assuming Lambda is cheapest addresses a separate condition: a streaming worker continuously processes high-throughput events 24/7. Steady sustained throughput can favor provisioned processing. The actual requirement is an interactive report runs every hour for five minutes and otherwise needs no server. Recommended corrective action: Run on scheduled serverless/batch service appropriate to runtime. Scheduled pay-as-used resources avoid 55 minutes idle.

Use AWS Batch/Glue managed capacity for the bounded job would help if a data engineer needs parallel one-off ETL across thousands of records. Managed job services provision for processing periods. That is not the symptom described here: an interactive report runs every hour for five minutes and otherwise needs no server. The matching response is Run on scheduled serverless/batch service appropriate to runtime, because scheduled pay-as-used resources avoid 55 minutes idle.

The option Use API Gateway plus Lambda when function limits permit addresses a separate condition: traffic is near zero most nights yet EC2 hosts are kept for a lightweight REST API. Invocation-based service can avoid idle fleet spend. The actual requirement is a scheduled container needs privileged OS customization unavailable to Fargate. Recommended corrective action: Use EC2-backed ECS or native EC2 with carefully sized instances. Runtime requirements determine whether managed tasks are suitable.

Use AWS Batch/Glue managed capacity for the bounded job would help if a data engineer needs parallel one-off ETL across thousands of records. Managed job services provision for processing periods. That is not the symptom described here: a scheduled container needs privileged OS customization unavailable to Fargate. The matching response is Use EC2-backed ECS or native EC2 with carefully sized instances, because runtime requirements determine whether managed tasks are suitable.

For the alternative failure a streaming worker continuously processes high-throughput events 24/7, the control Use appropriately sized ECS/EC2 workers rather than assuming Lambda is cheapest is relevant: steady sustained throughput can favor provisioned processing. Here the evidence instead is a scheduled container needs privileged OS customization unavailable to Fargate. The selected technical mechanism is Use EC2-backed ECS or native EC2 with carefully sized instances: runtime requirements determine whether managed tasks are suitable.

Scale GPUs to zero and provision by job queue demand would help if a GPU training fleet remains active when the training queue is empty. Queue-driven provisioning reduces unused acceleration cost. That is not the symptom described here: an engineering team has no ownership tags for chargeback on EC2 fleets. The matching response is Enforce cost allocation tags and enable reporting in billing tools, because tags attribute resource charges to responsible teams.

Add time-to-live cleanup and budget alerts to migration environments solves a different operational problem where a short-term migration spawns large temporary compute that remains allocated afterward. Expiry enforcement limits abandoned spend. It does not directly address an engineering team has no ownership tags for chargeback on EC2 fleets. The case calls for Enforce cost allocation tags and enable reporting in billing tools, since tags attribute resource charges to responsible teams.

The competing option Use scheduled Auto Scaling capacity reduction with monitoring would be useful when a web app is consistently underutilized during weekends. Scheduled scale-in matches recurring low demand. The present issue is an engineering team has no ownership tags for chargeback on EC2 fleets; it is more directly addressed by Enforce cost allocation tags and enable reporting in billing tools. Tags attribute resource charges to responsible teams.

For the alternative failure a cluster uses On-Demand for long predictable baseline jobs without commitments, the control Purchase appropriately scoped Savings Plans based on usage evidence is relevant: reservations/commitments can discount recurring use. Here the evidence instead is a web app is consistently underutilized during weekends. The selected technical mechanism is Use scheduled Auto Scaling capacity reduction with monitoring: scheduled scale-in matches recurring low demand.

The competing option Scale GPUs to zero and provision by job queue demand would be useful when a GPU training fleet remains active when the training queue is empty. Queue-driven provisioning reduces unused acceleration cost. The present issue is a web app is consistently underutilized during weekends; it is more directly addressed by Use scheduled Auto Scaling capacity reduction with monitoring. Scheduled scale-in matches recurring low demand.

The option Enforce cost allocation tags and enable reporting in billing tools addresses a separate condition: an engineering team has no ownership tags for chargeback on EC2 fleets. Tags attribute resource charges to responsible teams. The actual requirement is a web app is consistently underutilized during weekends. Recommended corrective action: Use scheduled Auto Scaling capacity reduction with monitoring. Scheduled scale-in matches recurring low demand.

The competing option Enforce cost allocation tags and enable reporting in billing tools would be useful when an engineering team has no ownership tags for chargeback on EC2 fleets. Tags attribute resource charges to responsible teams. The present issue is a GPU training fleet remains active when the training queue is empty; it is more directly addressed by Scale GPUs to zero and provision by job queue demand. Queue-driven provisioning reduces unused acceleration cost.

The option Use scheduled Auto Scaling capacity reduction with monitoring addresses a separate condition: a web app is consistently underutilized during weekends. Scheduled scale-in matches recurring low demand. The actual requirement is a GPU training fleet remains active when the training queue is empty. Recommended corrective action: Scale GPUs to zero and provision by job queue demand. Queue-driven provisioning reduces unused acceleration cost.

Purchase appropriately scoped Savings Plans based on usage evidence would help if a cluster uses On-Demand for long predictable baseline jobs without commitments. Reservations/commitments can discount recurring use. That is not the symptom described here: a GPU training fleet remains active when the training queue is empty. The matching response is Scale GPUs to zero and provision by job queue demand, because queue-driven provisioning reduces unused acceleration cost.

The option Scale GPUs to zero and provision by job queue demand addresses a separate condition: a GPU training fleet remains active when the training queue is empty. Queue-driven provisioning reduces unused acceleration cost. The actual requirement is a cluster uses On-Demand for long predictable baseline jobs without commitments. Recommended corrective action: Purchase appropriately scoped Savings Plans based on usage evidence. Reservations/commitments can discount recurring use.

Add time-to-live cleanup and budget alerts to migration environments would help if a short-term migration spawns large temporary compute that remains allocated afterward. Expiry enforcement limits abandoned spend. That is not the symptom described here: a cluster uses On-Demand for long predictable baseline jobs without commitments. The matching response is Purchase appropriately scoped Savings Plans based on usage evidence, because reservations/commitments can discount recurring use.

For the alternative failure a web app is consistently underutilized during weekends, the control Use scheduled Auto Scaling capacity reduction with monitoring is relevant: scheduled scale-in matches recurring low demand. Here the evidence instead is a cluster uses On-Demand for long predictable baseline jobs without commitments. The selected technical mechanism is Purchase appropriately scoped Savings Plans based on usage evidence: reservations/commitments can discount recurring use.

The option Enforce cost allocation tags and enable reporting in billing tools addresses a separate condition: an engineering team has no ownership tags for chargeback on EC2 fleets. Tags attribute resource charges to responsible teams. The actual requirement is a short-term migration spawns large temporary compute that remains allocated afterward. Recommended corrective action: Add time-to-live cleanup and budget alerts to migration environments. Expiry enforcement limits abandoned spend.

For the alternative failure a GPU training fleet remains active when the training queue is empty, the control Scale GPUs to zero and provision by job queue demand is relevant: queue-driven provisioning reduces unused acceleration cost. Here the evidence instead is a short-term migration spawns large temporary compute that remains allocated afterward. The selected technical mechanism is Add time-to-live cleanup and budget alerts to migration environments: expiry enforcement limits abandoned spend.

Purchase appropriately scoped Savings Plans based on usage evidence solves a different operational problem where a cluster uses On-Demand for long predictable baseline jobs without commitments. Reservations/commitments can discount recurring use. It does not directly address a short-term migration spawns large temporary compute that remains allocated afterward. The case calls for Add time-to-live cleanup and budget alerts to migration environments, since expiry enforcement limits abandoned spend.