SAP-C02 domain - 25% of the exam

Continuous Improvement for Existing Solutions

Continuous Improvement for Existing Solutions is 25% of the SAP-C02 exam. These are the objectives it covers, each with practice questions, with every answer explained.

The domain in numbers

70
Practice questions
8
Objectives
34
Medium
36
Hard

Where people go wrong

  • Offering in-place or all-at-once updates as a safe rollback

    The wrong answers change the running release in place or all at once: smaller in-place batches in CodeDeploy, alarm-based rollback on an in-place deployment, the ECS rolling update with a circuit breaker, or a Lambda alias repointed fully to the new version. None keeps an untouched original to return to. The correct answers run the new version in parallel and shift traffic to it: CodeDeploy blue/green, linear shifting on an ECS listener with a baking period, or a Lambda canary with a pre-traffic hook and an alarm as the rollback trigger.

    See it in a sample question
  • Treating Parameter Store as a rotating secret store

    The wrong answers hold the credential in Systems Manager Parameter Store and add a Lambda function and an EventBridge schedule to change it, or leave the key as a plaintext String parameter. Parameter Store has no built-in rotation, and a plaintext String parameter gives no customer-managed key encryption or per-retrieval audit. The correct answers use AWS Secrets Manager, with managed RDS rotation or a custom rotation Lambda function where no managed one exists, and the application reads the secret at runtime. Parameter Store fits the non-sensitive endpoints and flags.

    See it in a sample question
  • Adjusting a fixed alarm threshold instead of learning the pattern

    The wrong answers keep a static line and tune around it: a fixed percentile threshold, a lower threshold, more breaching datapoints, a long smoothing window, or a week-over-week comparison in custom code. Each one pages on normal evening peaks or reacts late to a slow climb. The correct answers use CloudWatch anomaly detection, which learns the daily and weekly pattern and alarms when the metric leaves the expected band. A composite alarm only combines existing alarm states, so the thresholds underneath it stay static.

    See it in a sample question
  • Fronting TCP and UDP services with CloudFront or DNS routing

    The wrong answers put CloudFront or Route 53 in front of a raw TCP or UDP service: CloudFront for an order API and market-data feed, a custom game protocol or a payments gateway, latency-based DNS, or a shorter time to live. CloudFront handles HTTP and cacheable content, and DNS answers depend on resolver caching and public-internet paths. The correct answers place AWS Global Accelerator in front of the Network Load Balancers, giving static anycast addresses, entry at the nearest edge and routing over the AWS backbone to the closest healthy Region.

    See it in a sample question
  • Using GuardDuty to find vulnerabilities and unused permissions

    The wrong answers choose Amazon GuardDuty for an inventory question. GuardDuty detects threats and anomalous behaviour from logs, traffic and runtime activity. It does not catalogue known software vulnerabilities in EC2 instances, container images or Lambda packages, and it does not list which granted permissions a role has left unused. The correct answers use Amazon Inspector to scan EC2, ECR and Lambda for known vulnerabilities, and an IAM Access Analyzer unused access analyzer for permissions that a role has never used.

    See it in a sample question

Objectives in this domain

What to study

The official documents this domain's practice questions cite most.

Sample questions from this domain

Free sampleContinuous Improvement for Existing Solutionshard

A payments company deploys a stateful Java application onto an Amazon EC2 Auto Scaling group fronted by an Application Load Balancer, orchestrated by AWS CodeDeploy with an in-place deployment that updates instances in batches. The most recent release introduced a regression that only surfaced under production load, and the team had to drain customers off a degraded fleet while the old version was redeployed, causing a long recovery. Leadership now wants the next version stood up on a completely separate set of instances, validated, then cut over with the ability to revert to the untouched original fleet in seconds if metrics degrade. Which CodeDeploy change BEST delivers this fast, low-risk cutover and rollback?

  • AKeep the in-place deployment but reduce the batch size to one instance at a time and raise the minimum healthy hosts threshold so that only a small portion of the fleet ever runs the new version at any single moment during the gradual rolling replacement of the existing group.
  • BAdd automatic rollback on a CloudWatch alarm to the existing in-place deployment, so when the alarm watching the application error rate breaches its threshold CodeDeploy redeploys the last known good revision back onto the same instances and restores service to affected customers automatically.
  • CReplace CodeDeploy with a Route 53 weighted routing policy that splits traffic between two target groups and shift the weight gradually toward the new version, lowering it again if the application begins to return elevated error responses to clients.
  • DSwitch the CodeDeploy deployment to the blue/green type so a replacement Auto Scaling group is provisioned with the new version, traffic is shifted to it behind the load balancer after validation, and the original group is kept to allow an immediate reroute back on failure. Correct
Use CodeDeploy blue/green for an EC2 Auto Scaling group to validate a separate fleet and reroute traffic back to the untouched original on failure. CodeDeploy blue/green provisions a brand-new replacement Auto Scaling group running the new version, validates it, and registers it with the load balancer before shifting traffic, while the original group is retained. If post-cutover metrics degrade, CodeDeploy reroutes the load balancer back to the original group in seconds because that fleet was never modified. Tuning in-place batches, alarm rollback on in-place, or Route 53 weighting all either mutate the original hosts or shift traffic slowly through DNS, so none gives the near-instant revert to a preserved fleet.

Why A is wrong: A smaller in-place batch still mutates the original instances in place, so there is no untouched fleet to fall back to and rollback means redeploying the previous version over the same hosts, which is the slow recovery the team is trying to eliminate.

Why B is wrong: Alarm-based rollback on an in-place deployment reinstalls the prior revision onto the same mutated hosts, which takes as long as a deployment rather than seconds, so it does not provide the instant reroute to a preserved original fleet that blue/green offers.

Why C is wrong: Route 53 weighting shifts traffic at DNS resolution and is subject to client-side caching of records, so cutover and revert are imprecise and slow, and abandoning CodeDeploy loses the managed deployment lifecycle and health-driven automation the team already relies on.

Why D is correct: Blue/green provisions a separate replacement Auto Scaling group, registers it behind the load balancer once health checks pass, and leaves the original group running so CodeDeploy can reroute traffic back to it almost instantly if the new version degrades, exactly matching the fast cutover and rollback requirement.

Free sampleContinuous Improvement for Existing Solutionshard

A healthcare company manages configuration values for an application that runs on Amazon EC2 across development, staging, and production accounts. Today, database endpoints, feature flags, and an external API key are hardcoded into AMIs, so any change forces a full rebuild and redeploy. Auditors require that the API key be encrypted at rest with a customer-managed key and that its retrieval be logged, while the non-sensitive endpoints and flags should be editable centrally without rebuilding instances, with each environment reading its own values. The team wants to minimise operational overhead and avoid building a custom configuration service. Which approach BEST meets these requirements?

  • AStore the endpoints, feature flags, and the external API key together as plaintext String parameters in Systems Manager Parameter Store under a per-environment path prefix, and restrict who is able to read the production path by attaching IAM policies scoped to that parameter hierarchy.
  • BKeep all three categories of value baked into per-environment AMIs as before but encrypt each AMI's EBS snapshots with a customer-managed AWS KMS key, and rebuild and redeploy the instances whenever any endpoint, flag, or key value needs to change.
  • CStore the endpoints and feature flags as Systems Manager Parameter Store parameters per environment path for central editing, and keep the API key in AWS Secrets Manager encrypted with a customer-managed key, reading both at runtime so retrievals are logged in CloudTrail. Correct
  • DRun a small self-hosted configuration service on an EC2 instance in each account that serves endpoints and flags over a TLS endpoint and stores the API key in an encrypted file on disk, and have the application query that service at startup to fetch its environment values.
Split configuration by sensitivity: Parameter Store for central non-secret values and Secrets Manager with a customer-managed key for encrypted, CloudTrail-audited secrets. Parameter Store keeps the endpoints and feature flags as per-environment parameters that the application reads at runtime, so they can be changed centrally without rebuilding AMIs. Secrets Manager encrypts the API key with a customer-managed KMS key and records every GetSecretValue call in CloudTrail, meeting the encryption-at-rest and retrieval-logging mandate. Plaintext parameters, encrypted AMIs, or a self-hosted server each miss either the audited secret handling or the rebuild-free central editing, so the split-by-sensitivity design is the best fit.

Why A is wrong: Holding the API key as a plaintext String parameter leaves the secret unencrypted with a customer-managed key and provides no per-retrieval audit, so although Parameter Store handles the non-sensitive values well it fails the encryption and logging requirement for the key.

Why B is wrong: Encrypting the AMI snapshots protects the volume at rest but still hardcodes the values, so every change forces the rebuild the team wants to eliminate and there is no central editing or per-retrieval logging of the API key as the auditors require.

Why C is correct: Parameter Store holds the non-sensitive endpoints and flags under per-environment paths so they are edited centrally without rebuilding AMIs, while Secrets Manager encrypts the API key with a customer-managed key and logs each GetSecretValue retrieval in CloudTrail, satisfying both the central-config and the encrypted, audited-secret requirements with managed services.

Why D is wrong: A self-hosted configuration service is exactly the custom service and added operational overhead the team wants to avoid, and a hand-rolled encrypted file gives no customer-managed-key integration or built-in per-retrieval audit trail compared with the managed AWS options.

Free sampleContinuous Improvement for Existing Solutionsmedium

An online retailer has noticed that its checkout error rate climbs gradually over several hours before customers complain, by which time the team is already firefighting. The application emits a per-minute error-rate metric to Amazon CloudWatch, but the rate is naturally noisy and a fixed threshold either pages too early on harmless spikes or too late on a real degradation. The team wants alerting that adapts to normal daily and weekly patterns and notifies on-call before the error rate becomes severe. Which approach BEST delivers earlier, lower-noise alerting?

  • ASet a static CloudWatch alarm threshold at the historical ninety-fifth percentile of the error rate and notify Amazon SNS, lowering the threshold by hand each time the team finds that an incident was missed.
  • BConfigure a metric math expression that averages the error rate over a rolling sixty-minute window and alarms when that smoothed value exceeds a fixed level, paging on-call through Amazon SNS once it is breached.
  • CForward the error-rate metric to a scheduled AWS Lambda function that compares the current value with the same minute one week earlier and publishes to Amazon SNS when the difference is large enough to matter.
  • DCreate a CloudWatch alarm that uses anomaly detection on the checkout error-rate metric so the band learns the expected daily and weekly pattern, then notify Amazon SNS when the metric moves outside the band. Correct
Apply CloudWatch anomaly detection to alarm on a metric's learned daily and weekly band rather than a static threshold for earlier, quieter alerts. CloudWatch anomaly detection trains a model on a metric's historical pattern, including daily and weekly seasonality, and produces an expected band. An alarm on that band fires when the error rate drifts outside normal behaviour, catching a gradual degradation earlier than a static threshold and suppressing routine spikes, without the custom code or constant manual retuning the other options require.

Why A is wrong: A fixed percentile threshold does not adapt to time of day or week, so it keeps producing the early or late pages the team is trying to avoid, and manual retuning after each miss is reactive and ongoing toil.

Why B is wrong: Smoothing over a long window reduces noise but adds latency and still relies on a single fixed level, so it tends to alert later on a slow climb and does not adapt to the normal daily and weekly shape of the metric.

Why C is wrong: A bespoke week-over-week Lambda is custom code to build and maintain and compares against a single past point, so a one-off anomaly last week skews it, unlike a trained anomaly-detection band over the full history.

Why D is correct: CloudWatch anomaly detection builds a model of the metric's normal daily and weekly behaviour and alarms when values fall outside the learned band, so it catches a gradual drift earlier than a fixed line while ignoring routine spikes.

Free sampleContinuous Improvement for Existing Solutionshard

A payments company runs an active-active TCP gateway in two Regions and currently steers users with Amazon Route 53 weighted records. During a recent regional impairment, the agreed 60-second failover KPI was missed because client resolvers and on-device caches held the failed Region's records well past the configured time to live. The clients cannot be updated to change their DNS behaviour, and the team needs near-immediate failover and the ability to shift traffic weight between Regions without depending on DNS propagation. Which change MOST effectively meets these requirements?

  • AFront both regional endpoints with AWS Global Accelerator using its static anycast addresses, configure endpoint health checks, and adjust the traffic dial and endpoint weights to shift and fail over traffic without any DNS change on the client. Correct
  • BLower the time to live on the Amazon Route 53 weighted records to a few seconds and enable health checks, so resolvers refresh the records quickly and stop returning the impaired Region soon after a health check fails.
  • CReplace the weighted records with Amazon Route 53 failover routing combined with an Amazon CloudFront distribution placed in front of each Region, so cached edge responses are expected to cover the outage gap while DNS slowly converges on the surviving Region during an impairment.
  • DDeploy AWS Transit Gateway peering between the two Regions and advertise a shared address range, so the network reconverges to the surviving Region during an impairment without the clients needing any DNS record change at all.
Use AWS Global Accelerator anycast addresses and the traffic dial to achieve fast Regional failover that does not depend on client DNS caching. Because client and resolver DNS caches can outlast a record's time to live, DNS-based failover cannot guarantee a tight failover KPI; Global Accelerator keeps clients on stable anycast IPs and performs the failover and weight shift at the edge in seconds, removing the dependency on DNS propagation entirely.

Why A is correct: Global Accelerator keeps the client pointed at fixed anycast addresses and reroutes at the edge when an endpoint becomes unhealthy, so failover and weight shifts take effect in seconds independent of resolver and client DNS caching.

Why B is wrong: A shorter time to live narrows the window but still depends on resolvers and client caches honouring it, and the incident showed those caches outlast the configured value, so the failover KPI remains at risk.

Why C is wrong: CloudFront fronts HTTP and caches content, which does not suit a stateful TCP payments gateway, and the design still relies on DNS convergence for the actual failover the resolvers were caching past.

Why D is wrong: Transit Gateway connects VPCs and on-premises networks for internal routing and does not provide a public global front door or client-facing failover, so it does not address how external payment clients reach a healthy Region.

Free sampleContinuous Improvement for Existing Solutionsmedium

A software firm runs container workloads on Amazon ECS using images pulled from Amazon ECR, alongside a fleet of Amazon EC2 instances. Security must continuously detect known software vulnerabilities and unintended network exposure across both the running EC2 instances and the container images in the registry, with findings prioritised by severity and as little agent and scanner management as possible. Which approach BEST provides this continuous vulnerability detection?

  • AInstall and licence a third-party vulnerability scanner on every EC2 instance and add a dedicated scan stage to each container build pipeline, then aggregate the resulting reports into an Amazon S3 bucket and triage the highest-severity items with a scheduled Athena query each week.
  • BEnable Amazon Inspector for the account so it automatically and continuously scans eligible EC2 instances and ECR container images for known vulnerabilities and network reachability, presenting prioritised findings without managing standalone scanners. Correct
  • CTurn on Amazon GuardDuty across the account so its threat intelligence feeds flag malicious activity, treating the runtime alerts it raises as the inventory of software vulnerabilities in the EC2 and container estate.
  • DUse AWS Config with managed rules to record instance and image configuration changes, relying on the resulting compliance timeline to reveal which workloads are carrying vulnerable software packages.
Enable Amazon Inspector to continuously and automatically scan EC2 instances and ECR images for vulnerabilities and exposure with prioritised findings. Amazon Inspector automatically discovers eligible EC2 instances and ECR container images, continuously assesses them against vulnerability databases, evaluates network reachability for unintended exposure, and assigns a severity score to each finding so teams can prioritise. This managed, agent-light model avoids running standalone scanners, which a per-host third-party scanner, GuardDuty threat detection and Config configuration tracking each cannot provide for vulnerability assessment.

Why A is wrong: Bolting on a third-party scanner per host and per pipeline gives results but is exactly the scanner management and weekly manual triage the requirement asks to minimise, and it is not continuous.

Why B is correct: Amazon Inspector continuously and automatically scans eligible EC2 instances and ECR images for software vulnerabilities and network reachability, scoring findings by severity, which removes the burden of running separate scanners.

Why C is wrong: GuardDuty detects active threats and anomalous behaviour from logs and traffic, not the catalogue of known software vulnerabilities in instances and images, so it does not deliver vulnerability assessment.

Why D is wrong: Config tracks configuration state and drift but does not analyse package versions against a vulnerability database, so it cannot identify which workloads contain known CVEs.

Other domains in this exam

See also the SAP-C02 cert hub, the study guide, and the cheat sheet.

Examworthy is not affiliated with or endorsed by Amazon Web Services. Original, blueprint-aligned practice material only.