在打印窗口把「目标」选为「另存为 PDF」即可保存,页面已带 awsgood.com 版权水印。 返回原文

www.awsgood.com

DOP-C02 Exam Domain Deep Dive: SDLC Automation Traps

2026-11-13 DOP-C02 DevOps Engineer - Professional 约 5 分钟

TL;DR

The six domains, and why 22% is bigger than it sounds

DOP-C02 splits into six task areas. AWS publishes the weights in the exam guide, and they matter because a 65-question scored set means each percentage point is roughly two-thirds of a question.

# Domain Weight ~Scored questions Where candidates bleed points
1 SDLC Automation 22% ~14 Cross-account artifacts, deployment strategy choice, pipeline trigger semantics
2 Configuration Management and IaC 17% ~11 CloudFormation drift, StackSets, change sets
3 Resilient Cloud Solutions 15% ~10 Multi-AZ vs multi-Region trade-offs, backup/restore targets
4 Monitoring and Logging 15% ~10 Metric filters, CloudWatch agent config, log retention cost
5 Incident and Event Response 14% ~9 EventBridge routing, auto-remediation, runbook automation
6 Security and Compliance 17% ~11 Cross-account IAM, secrets rotation, policy boundaries

There are 75 questions total: 65 scored and 10 unscored items AWS is trialling, and you cannot tell them apart. That is worth remembering when a question feels strangely worded or trivial — it may not count.

Who should book it: you've run CodePipeline or an equivalent in production, you've debugged a failed deploy at an inconvenient hour, and you're comfortable with IAM trust policies. Two years of hands-on AWS ops is a reasonable bar. If your day job is single-account EC2 administration with deploys done by SSH, the SDLC and Resilience domains will hurt.

SDLC automation, decoded: what the 22% actually asks

Strip out the marketing and the domain asks four things. How do you get code from a repo into an artifact reproducibly? How do you build it with a build spec that fails when it should? How do you shift traffic to the new version without a rollback fire drill? And how do artifacts, state and permissions move between accounts and Regions?

The CodeBuild half is mostly reading output correctly. A build that exits non-zero yields COMMAND_EXECUTION_ERROR: Error while executing command: npm ci. Reason: exit status 1 in the build log, which is a build failure, not an infrastructure failure — the distinction shows up in "why did the pipeline stop here" questions. A minimal spec that actually produces something deployable:

version: 0.2
phases:
  install:
    runtime-versions:
      nodejs: 20
  pre_build:
    commands:
      - echo "building $CODEBUILD_RESOLVED_SOURCE_VERSION"
      - npm ci
  build:
    commands:
      - npm run build
reports:
  jest-report:
    files:
      - 'clover.xml'
    base-directory: 'coverage'
    file-format: CLOVERXML
artifacts:
  files:
    - appspec.yml
    - scripts/**/*
    - dist/**/*


The CodeDeploy half is strategy selection, and the predefined config names are worth memorising because they appear verbatim in answer options.

| Config | Traffic shift | Pick it when |
|---|---|---|
| `CodeDeployDefault.OneAtATime` | One instance at a time, stops on failure | A bad host must not affect the rest of the fleet |
| `CodeDeployDefault.HalfAtATime` | 50% then 50% | Large EC2 fleet, you want a fast but staged cutover |
| `CodeDeployDefault.LambdaCanary10Percent5Minutes` | 10% for 5 minutes, then the rest | Lambda alias deploys with real traffic |
| `CodeDeployDefault.ECSCanary10Percent5Minutes` | 10% then 100% | ECS services behind an ALB |
| `CodeDeployDefault.ECSLinear10PercentEvery1Minutes` | +10% each minute | Slow burn where you want alarm-based rollback |

Note the odd plural in the ECS linear configs — `Every1Minutes`, not `Every1Minute` as Lambda uses. That one trips people in labs and it's exactly the kind of detail the exam leaves in a plausible-looking distractor.

Two other things show up repeatedly. CodePipeline V2 pipeline types add execution modes (`QUEUED`, `PARALLEL`, `SUPERSEDED`) and trigger filters on branches and file paths — V1 pipelines don't have those, and a question asking for "only run when files under /infra change" is pointing at V2. Then there's the cross-account shape: an artifact bucket policy, a KMS key policy, and a role in the target account. Candidates reliably remember the bucket policy and forget the key.

## Booking, cost, and the six weeks before you sit it

Registration runs through aws.amazon.com/certification into Pearson VUE: pick a test centre or online proctoring, choose a slot, pay. The exam is available in English and several other languages, and you get your result on screen at the end.

| Item | What to budget |
|---|---|
| Exam fee | USD 300 list price; regional pricing and tax differ |
| Retake | Full fee again — no bundled second attempt |
| Study hours, existing production CI/CD experience | 40–60 |
| Study hours, no pipeline experience | 75–90, mostly hands-on |
| Practice exams | 2–3 full-length, timed, before booking the real slot |
| Validity | 3 years from the pass date |

A six-week plan that fits around a job: week 1 read the exam guide task statements and list everything you can't explain out loud. Weeks 2–3 build one small pipeline per domain — CodeCommit to CodeBuild to CodeDeploy on EC2, then a CodeDeploy ECS blue/green, then a cross-account deploy. Weeks 4–5 do configuration/IaC and monitoring, since those overlap heavily. Week 6 is practice exams plus re-reading your weak list only.

### Exam technique on a 180-minute clock

Seventy-five questions in 180 minutes is 2.4 minutes each, and the scenario stems are long. My rule: first pass, answer only what you can decide in under 90 seconds and flag the rest. You should be at question 40 around the 95-minute mark — if you're past 120 minutes there, you're borrowing time from review you'll need.

Read the qualifier before the scenario. "LEAST operational overhead" and "MOST cost-effective" often point at different answers for the same architecture. And on multiple-response items, partial selection scores zero, so count the checkboxes the prompt implies before you commit.

## The 01:37 deploy failure that explains half the domain

We moved a pipeline into a hardened toolchain account. At 01:37 on a Tuesday the Deploy stage failed with this:


Action execution failed
AccessDeniedException: User: arn:aws:sts::444455556666:assumed-role/CrossAccountDeploy/ci
is not authorized to perform: kms:Decrypt on resource:
arn:aws:kms:us-east-1:111122223333:key/8f2c1a70-... because no resource-based policy
allows the kms:Decrypt action


The first configuration was wrong in a specific, boring way. Two months earlier a security review moved the artifact bucket from SSE-S3 to a customer managed key in `us-east-1`. The key policy got a statement for the pipeline's own role, which is why builds and same-account depl

Sources