CI/CD¶
GitHub Actions with self-hosted runners.
Workflow Structure¶
.github/workflows/
├── deploy.yml # Cloud orchestrator (change detection, dependency chain)
├── _quality.yml # Linting, security scanning
├── _core.yml # Core infrastructure (ECR, budgets)
├── _range.yml # Range VPC infrastructure
├── _shifter-engine.yml # Engine container build and push
├── _shifter-platform.yml # Portal infrastructure and app deployment
├── _gcp-dev.yml # GCP CI wrapper over the gcp-deploy commands
├── release-gcp.yml # GCP release: images, attestations, deployer image (release-please.yml calls it)
├── packer-gcp.yml # GCP guest image build, validation call, and publication
├── packer-gcp-validate.yml # GCP guest candidate-boot validation (packer-gcp.yml calls it)
├── packer.yml # AMI builds (AWS)
└── packer-promote.yml # AMI promotion to prod (AWS)
Deployment Chain¶
graph LR
Quality --> Core
Core --> Range
Core --> ShifterEngine["Shifter Engine"]
Range --> ShifterPlatform["Shifter Platform"]
ShifterEngine --> ShifterPlatform
Jobs run only when relevant files change. deploy.yml detects changes and triggers appropriate workflows.
Change Detection¶
| Job | Triggers On |
|---|---|
| core | platform/terraform/modules/ecr/**, platform/terraform/environments/*/*.tf |
| range | platform/terraform/modules/range/**, platform/terraform/environments/*/range/** |
| shifter_engine | shifter/engine/provisioner/**, platform/terraform/environments/*/eks/** |
| shifter_platform | platform/terraform/modules/portal/**, platform/terraform/environments/*/portal/** (Terraform only) |
| portal_image | shifter/shifter_platform/**, shifter/installation/** (portal image build/deploy, no Terraform) |
Environment Targeting¶
Environment deploys are manual (workflow_dispatch with an environment input);
push and pull_request run validation only (#730).
- Pull request → Quality only
- Push to
dev/main→ Quality only; no deploy - Manual dispatch
environment=aws-dev→ AWS dev deploy
Run a deploy with gh workflow run deploy.yml --ref <branch> -f environment=<env>.
Authentication¶
OIDC federation per cloud. No long-lived credentials.
| Secret | Purpose |
|---|---|
AWS_ROLE_ARN |
AWS prod IAM role |
AWS_ROLE_ARN_DEV |
AWS dev IAM role |
GCP_DEPLOY_SERVICE_ACCOUNT |
Purpose-scoped GCP deploy service account email |
GCP_PACKER_BUILD_SERVICE_ACCOUNT |
Purpose-scoped GCE image build service account email (gcp-image-build) |
GCP_PACKER_VALIDATE_SERVICE_ACCOUNT |
Purpose-scoped GCE image validation service account email (gcp-image-validate) |
GCP_RELEASE_SCAN_SERVICE_ACCOUNT |
Purpose-scoped GCP release-scan service account email |
GCP_WORKLOAD_IDENTITY_PROVIDER |
GCP Workload Identity Federation provider |
AWS roles defined in platform/terraform/global/iam/github-oidc.tf. The GCP
WIF provider and the purpose identities come from the
platform/terraform/gcp/global/cicd-oidc root. GitHub federation is optional
there: a GCP deployment that runs without GitHub has no WIF provider, and its
operators impersonate the purpose identities instead (ADR-070).
GCP Current State¶
No GCP tenant deploys through CI/CD after #2668. GCP tenants install published
releases with the GitHub-less commands (ADR-070). The GCP jobs stay in
deploy.yml, but no environment option selects them; #2672 decides the future
of that path. When a tenant is wired in, its deployment Environment and the
purpose-scoped gcp-release-scan-<deployment suffix> Environment provide their
own exact WIF identity variables. The reusable workflow does not reuse the
deploy identity for release scanning.
The developer build project holds the only GitHub Actions WIF identities in use.
packer-gcp.yml builds in the gcp-image-build Environment and calls
packer-gcp-validate.yml in the gcp-image-validate Environment (see
docs/architecture/gcp-guest-images.md).
_gcp-dev.yml is a thin wrapper over the GitHub-less deploy commands
(ADR-070). Each job renders the deployment input bundle from the GitHub
secrets and variables with scripts/gcp/render_ci_deployment_inputs.py, then
runs scripts/bootstrap/deploy.py:
| Job | Command |
|---|---|
validate |
terraform fmt and terraform validate |
prepare |
preflight, gcp-deploy --stage infrastructure, then the image build and attestation with push-to-registry: true |
release_scan |
gcp-release-scan as the release-scan identity |
deploy |
gcp-deploy --stage rollout |
post-deploy-smoke |
gcp-smoke; a failure opens a GitHub issue |
No GitHub workflow runs gcp-destroy. An operator runs it as the destroy
identity (ADR-070).
Every deploy dispatch runs the quality gate before an apply. All GCP
deployments use the shared Terraform root environments/platform with the
deployment's own state prefix. The command keeps TF_DATA_DIR in a private
directory for each run. The AWS application lifecycle uses EKS and the shared
Helm chart.
The deploy command sets public_hostname from deployment.domain and sets
enable_managed_tls = true. It fails closed unless
gke_master_authorized_cidrs is empty for Connect Gateway or contains only
connected RFC1918 networks.
A GCP deployment can also run without GitHub: an operator runs the same
commands from the deployer image of a published release. See the
GitHub-less GCP deployment guide.
gdc-bootstrap builds only the legacy GDC VM Runtime substrate for the gdc
range backend.