AWS CodePipeline
Fully managed CI/CD orchestration service that automates release pipelines from source to deploy.
Verified against AWS documentation · 2026-08-28
Key facts
- 01Pipeline = stages (Source → Build → Test → Deploy), each stage = one or more actions; artifacts pass between stages via an S3 bucket
- 02Orchestrates, doesn't execute — build work goes to CodeBuild, deploys go to CodeDeploy / ECS / Lambda / CloudFormation / Elastic Beanstalk
- 03Sources: GitHub / GitLab / Bitbucket (via connections), S3, ECR — a push or change triggers the pipeline automatically
- 04Manual approval actions gate a stage until a human clicks approve (e.g., before prod deploy)
- 05V1 pipelines: flat structure; V2 pipelines: adds triggers (Git tags, branch filters), pipeline variables, and stage conditions
Exam triggers
Phrases and scenarios that point here.
- "Automate the release process", "CI/CD pipeline", "deploy on every commit" → CodePipeline
- "Require human sign-off before production" → manual approval action
- Pipeline that builds → CodePipeline + CodeBuild; pipeline that deploys to EC2/on-prem → + CodeDeploy
- Cross-account or cross-region deployments in one release flow → CodePipeline supports both natively
Not this service
- Actually compiling/testing code → CodeBuild (CodePipeline just calls it)
- Deploying to instances with rolling/blue-green strategies → CodeDeploy
- Simple GitHub-native workflows with no AWS orchestration → GitHub Actions
Use cases
- 01Push to main → build in CodeBuild → deploy to staging → manual approval → deploy to prod, all hands-off after the commit
- 02Assembly-line mental model: each station (stage) does one job and passes the part (artifact) down the belt; a failed station stops the belt so bad parts never reach the end
- 03Infrastructure delivery: pipeline picks up a CloudFormation template change and rolls it through dev → prod with a change-set review step
- 04Multi-region rollout: one pipeline deploying the same build artifact to us-east-1 and eu-west-1 in sequence, stopping if the first region fails
- 05Container flow: new image lands in ECR → pipeline triggers → ECS service updated, no one ever SSHes anywhere
How it's charged
- 01Two pricing models by pipeline type — V1: $1/month per active pipeline (one with at least one execution that month); V2: per action execution minute
- 02V1 idle pipelines cost nothing that month, and each new pipeline is free for its first 30 days
- 03V2 doesn't bill manual approval or custom actions — only standard actions (source, build, test, deploy) accrue minutes
- 04Billing surprise: CodePipeline is rarely the bill — CodeBuild minutes, artifact storage in S3, and downstream compute usually dominate total CI/CD cost
- 05Billing surprise: on V2, failed runs and retries still burn action minutes — flaky tests literally cost money
