Skip to content
Tutorials Center
My learning

Topics

Lesson 30 of 36 ยท CI/CD Patterns

Running Packer in GitHub Actions

Build images in GitHub Actions: setup-packer, init, validate and build jobs, OIDC to AWS without stored keys, matrix builds per source, caching plugins and pull request versus main behavior.

Building images from a laptop does not scale: results depend on one person's tools and credentials, and nobody can audit what happened. A CI system makes builds repeatable, reviewable and permission-controlled. This lesson uses GitHub Actions for examples, but every concept maps to GitLab CI, Jenkins or any other runner.

Workflows cannot execute on this site. The YAML is read-only; JavaScript stand-ins model the decisions a pipeline makes (which jobs run, how a matrix expands). The action versions shown are examples: check each action's page for the current release and pin to a commit or tag you trust.

Pipeline shape

flowchart LR
  PR["Pull request"] --> L["fmt + validate"]
  L --> T["Optional test build"]
  M["Merge to main"] --> B["Build images"]
  B --> V["Test instance"]
  V --> R["Publish / promote"]

Pull requests only need the cheap checks; merges to main build and publish.

A pull request workflow

YAML
name: packer-validate
on:
  pull_request:
    paths: ["images/**", ".github/workflows/packer-*.yml"]

jobs:
  validate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: hashicorp/setup-packer@main
        with:
          version: "1.11.2"
      - run: packer fmt -check -diff -recursive images/
      - run: packer init images/
      - run: packer validate -var-file=images/ci.pkrvars.hcl images/

Pin the Packer version so local and CI behave the same. Use paths so unrelated changes do not trigger image jobs.

A build workflow with OIDC

Static cloud keys stored in repository secrets are long-lived and easy to leak. With OpenID Connect, the job receives a short-lived token from GitHub and exchanges it for temporary cloud credentials. For AWS:

YAML
name: packer-build
on:
  push:
    branches: [main]
    paths: ["images/**"]
  schedule:
    - cron: "0 3 * * 1"

permissions:
  id-token: write
  contents: read

jobs:
  build:
    runs-on: ubuntu-latest
    strategy:
      fail-fast: false
      matrix:
        source: ["amazon-ebs.web", "amazon-ebs.worker"]
    steps:
      - uses: actions/checkout@v4
      - uses: hashicorp/setup-packer@main
        with:
          version: "1.11.2"
      - uses: aws-actions/configure-aws-credentials@v4
        with:
          role-to-assume: arn:aws:iam::123456789012:role/packer-build
          aws-region: eu-central-1
      - run: packer init images/
      - run: >
          packer build -only="${{ matrix.source }}"
          -var "git_sha=${{ github.sha }}" images/
      - uses: actions/upload-artifact@v4
        with:
          name: manifest-${{ matrix.source }}
          path: packer-manifest.json

Notes:

  • permissions: id-token: write lets the job request the OIDC token. The IAM role's trust policy should be limited to your repository and branch.
  • The account id and role name above are placeholders.
  • matrix builds one source per job in parallel; -only selects it. fail-fast: false lets other images finish when one fails.
  • A schedule trigger rebuilds weekly to pick up OS patches (rebuild schedules).
  • Upload the manifest as an artifact so downstream jobs and humans can find the image ids.

Practical concerns

  • Concurrency. Two simultaneous builds of the same image name collide. Add concurrency: group: packer-${{ github.ref }}.
  • Timeouts. Set timeout-minutes so a hung SSH wait does not burn runner minutes.
  • Plugin caching. packer init downloads plugins each run; cache ~/.config/packer/plugins keyed on the lock file hash if it matters.
  • Secrets from forks. Pull requests from forks do not get secrets, so keep PR jobs credential-free (validate only).
  • Environments and approvals. Use a GitHub environment with required reviewers for the job that publishes to production.
  • Self-hosted runners are needed for on-premises hypervisor builds. Keep them ephemeral and isolated.
  • Cleanup on cancel. If a job is cancelled, the builder instance may remain; tag builder resources (run_tags) and run a periodic janitor job that deletes old ones.
  • Logs. Avoid PACKER_LOG=1 on public repositories; it can expose details.

Try it: decide which jobs run

JS Event router (stand-in)

Press Run (or Ctrl + Enter) to see the console output here.

No output.

Try it: expand a matrix

JS Matrix expansion (stand-in)

Press Run (or Ctrl + Enter) to see the console output here.

No output.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. Why prefer OIDC to stored access keys in CI?

    The job exchanges a short-lived token for temporary credentials.
  2. 2. What should a pull request from a fork run?

    Forks do not receive secrets and must not be able to use your cloud account.
  3. 3. Which flag selects one source from a template with several?

    packer build -only=amazon-ebs.web builds just that source.
  4. 4. Why add a scheduled trigger?

    Images age; periodic rebuilds keep them patched.

Summary

  • PRs run fmt, init, validate; merges and schedules build, test and publish.
  • Use OIDC for cloud credentials, a matrix with -only for parallel images and concurrency to avoid collisions.
  • Upload the manifest as an artifact and gate production publishing with environments.
  • Set timeouts, cache plugins and run a janitor for leaked builder resources.

Next: Image versioning, registries and promotion