cert-bank

Guides

이 글을 한국어로 읽기

DOP-C02 Study Guide: The DevOps Professional Exam Wants Closed Loops, Not Heroes

The AWS Certified DevOps Engineer – Professional exam (DOP-C02) has a favourite kind of correct answer: one where nobody has to wake up. Something notices a problem, something routes the signal, something fixes it or rolls it back, and a record is left behind. Answers that rely on a person reading an email, running a script or logging in to a server are usually the distractors. This guide teaches the exam through those loops: the release loop, the configuration loop and the operations loop.

Study guide · AWS Certified DevOps Engineer – Professional (DOP-C02)

DOP-C02 Study Guide: The DevOps Professional Exam Wants Closed Loops, Not Heroes

The exam

75 questions in 180 minutes (65 scored and 10 unscored, which are not marked), each multiple choice or multiple response; the result is a scaled score from 100 to 1,000 and the passing score is 750. Six domains: SDLC Automation (22%), Configuration Management and IaC (17%), Resilient Cloud Solutions (15%), Monitoring and Logging (15%), Incident and Event Response (14%), Security and Compliance (17%). As of November 2026 AWS has not announced a successor version of DOP-C02, but the Korean-language version of this exam is retired after December 31, 2026.

In the 395 DOP-C02 questions we wrote for this question bank, no single service dominates. CloudWatch, Lambda, CloudFormation, AWS Config and Systems Manager each appear in about one question in ten, followed closely by IAM, Organizations, CodePipeline, CodeBuild, S3 and EventBridge. That spread is the point: the exam tests how services connect, not one service in depth. It also has fewer multi-answer questions than SAP-C03 or AIP-C01 in our AWS bank (about 6%, against 9% for SAP-C03 and 11% for AIP-C01), so most of the time is spent reading long scenarios, not choosing combinations.

The loop behind most answers

Read each scenario as three slots: what detects the condition, what carries the signal, and what acts on it. The correct answer fills all three with managed pieces.

DetectRouteAct
AWS Config rule finds a noncompliant resourceConfig remediationSystems Manager Automation runbook fixes it
CloudWatch alarm during a deploymentAttached to the CodeDeploy deployment groupAutomatic rollback (only if you enable it)
GuardDuty finding or AWS Health eventEventBridge ruleLambda, Step Functions or an Automation runbook
Several related alarmsComposite alarmOne notification, or an OpsItem in Systems Manager OpsCenter
A pattern in application logsMetric filter, then an alarm on its sumNotification or automated action
An Auto Scaling instance is about to terminateTermination lifecycle hookDrain connections, ship logs, then continue

Two details catch people. A metric filter only sees log events that arrive after it is created, so for questions about logs you already have, the answer is CloudWatch Logs Insights. And event routing has failure modes of its own: EventBridge retries transient delivery errors and can send what still fails to a dead-letter queue; cross-account events need both a rule in the sending account and a bus policy in the receiving one.

DOP-C02: the loop behind most correct answers
1
Detect
Config rule, CloudWatch alarm, GuardDuty finding, or a metric filter on application logs
2
Route
Config remediation, EventBridge rule, composite alarm, or the CodeDeploy deployment group
3
Act
Automation runbook, automatic rollback, Lambda or Step Functions, or a lifecycle hook
4
Record
A record is left behind, for example an OpsItem in Systems Manager OpsCenter
The correct answer fills every slot with a managed piece; a person in any slot marks a distractor.

The release loop: choose the rollout by what you must be able to undo

SDLC Automation is the heaviest domain, and its questions often hinge on one word in the requirement: "instantly", "gradually", "validate before traffic", "without new instances".

StrategyWhat it gives you
In-place (rolling)No extra fleet; the same instances and volumes are reused. Rollback means deploying the old version again. On EC2 and on-premises, CodeDeploy offers OneAtATime, HalfAtATime and AllAtOnce.
Blue/greenA new fleet or task set, tested before traffic moves, with the old one still running for an instant switch back. Anything stored on the old instances is lost unless it was moved out first.
CanaryOne small slice of traffic, a wait, then everything (Lambda and ECS).
LinearEqual slices on a timer until 100% (Lambda and ECS).

Whenever two versions run at the same time, they share the same database. Schema changes must therefore be backward compatible: add first, switch code, remove old columns last (often called expand and contract). An answer that renames a column in the same release as the code change breaks the version still running.

Around the deployment: CodePipeline's manual approval action for human sign-off, CodeConnections for GitHub sources without stored tokens, and Secrets Manager references in the buildspec rather than secrets in environment variables. CodeDeploy does not deploy to EKS at all (its compute platforms are EC2/on-premises, Lambda and ECS); on EKS, canaries come from Kubernetes controllers such as Argo Rollouts or Flagger.

Beginner tip: deploy a small Lambda function with AWS SAM using DeploymentPreference set to Canary10Percent5Minutes and a CloudWatch alarm on the function's errors listed under its Alarms. Publish a second version that throws an exception, invoke it a few times during the five minutes, and watch CodeDeploy shift traffic back on its own. One working rollback teaches the alarm-to-rollback loop better than any table, and the whole stack can be removed with one command.

The configuration loop: what infrastructure as code will not do for you

  • Drift is detected, not repaired by a normal update. CloudFormation compares the stack with reality when asked; the AWS Config managed rule for stack drift makes that check recurring. To keep a manual change, put it into the template and review a change set. (Newer drift-aware change sets, --deployment-mode REVERT_DRIFT, can also put drifted resources back in line with the template; a standard update does not.)
  • Capabilities must be acknowledged. IAM resources with custom names need CAPABILITY_NAMED_IAM; SAM and other macros need CAPABILITY_AUTO_EXPAND.
  • Secrets stay out of templates. Use dynamic references to Secrets Manager; NoEcho only hides a value from display. For RDS, let the service manage the master password in Secrets Manager.
  • A new launch template version changes only future launches. Use an instance refresh to roll it through the running Auto Scaling group.
  • Many accounts at once: StackSets with service-managed permissions through Organizations, run from a delegated administrator; Service Catalog when developers should launch approved products through a launch role that holds the permissions they do not.
Going deeper: why CloudFormation leaves drift alone

A standard stack update decides what to change by comparing the new template with the template CloudFormation last deployed, not with the live resources. If someone edits a security group in the console, the stored template has not changed, so the next stack update sees nothing to do for that resource and the manual edit survives. Drift detection is a separate, read-only comparison of the live resource with the expected properties. When a later template change does touch that resource, CloudFormation can overwrite the manual change or fail the update, depending on the property, and a standard change set does not warn you in advance (only a drift-aware change set compares with the live state). That is why the exam's answer is always to move the change into the template (or revert it) rather than to "run an update to fix drift".

The operations loop: resilience without manual steps

Resilient Cloud Solutions questions favour designs where failure is handled by the platform: multi-AZ Auto Scaling groups with load balancer health checks; session state in ElastiCache or DynamoDB instead of sticky sessions; Lambda provisioned concurrency to remove cold starts (reserved concurrency only caps and reserves); Aurora Global Database for cross-Region relational recovery, and DynamoDB global tables when every Region must accept writes. Applications must reconnect by DNS name after a Multi-AZ database failover, because the name stays and the address changes.

A study order

  1. CodePipeline, CodeBuild and CodeDeploy end to end, including every deployment configuration and rollback setting.
  2. CloudFormation in depth (change sets, drift, stack policies, StackSets, dynamic references), then CDK and SAM.
  3. CloudWatch metrics, alarms, Logs Insights and metric filters; EventBridge patterns and cross-account buses.
  4. Systems Manager (Automation, Run Command, Patch Manager, Parameter Store, OpsCenter) and Config remediation.
  5. Organizations, Identity Center and multi-account security; then mixed sets with a 180-minute clock.

Practise

Practise in the question bank: every question explains each option, right and wrong. Start with the domain you just studied, then take a timed mock exam.

Exam facts as published in the AWS exam guide for DOP-C02 and on the AWS exam page; service shares refer to our own practice bank. Deployment options and service names (for example CodeConnections, formerly CodeStar Connections) change; check the current documentation and the official exam page.