ADR-0002 is fully landed (issues #11-#32 closed): every owned image now builds on GitHub Actions and pushes to ghcr.io/viktorbarzin/<name>, with Woodpecker reduced to deploy-only. The Forgejo container registry is frozen and emptied; there are no in-cluster image builds or CI test runs anywhere. The docs still described the old hybrid topology (DockerHub builds, Woodpecker-native owned-app builds, the per-pattern migration lists, the tripit-only pilot framing), which would mislead future sessions and incident response. This brings the docs to the completed reality (closes #33): - docs/architecture/ci-cd.md: full rewrite as the canonical CI/CD reference — the fleet GHA->ghcr->Woodpecker-deploy pattern, public/private ghcr package split, infra-owned image workflows (incl. infra-ci on ghcr), the frozen Forgejo registry, what Woodpecker still runs, and the #31 decommissions. - .claude/CLAUDE.md: rewrite the "CI/CD Architecture" section to the fleet-wide final state; FIX the stale claim that claude-memory-mcp builds to DockerHub (it is GHA->ghcr); note owned images now live on ghcr and the Forgejo registry is frozen/break-glass near the image-registry bullet. - .claude/reference/service-catalog.md: f1-stream is GHA->ghcr + Woodpecker deploy-only (was "Woodpecker-native build->deploy"). - stacks/{tuya-bridge,android-emulator}/variables.tf + stacks/terminal/main.tf: cosmetic description/comment updates (forgejo -> ghcr; terminal-lobby has no CI pipeline). Description/comment text only — no stack logic changed. Historical records (docs/post-mortems/*, docs/plans/*) and ADR-0002 itself are left untouched as point-in-time records. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
17 KiB
CI/CD Pipeline
Overview
Doctrine (ADR-0002): all image builds and CI compute run OFF-infra. Every
owned image is built, tested, and linted on GitHub Actions (free on public
repos; 2000 free min/mo on private) and pushed to ghcr.io/viktorbarzin/<name>.
Woodpecker is deploy-only — a GHA job POSTs its API with the freshly-built
image tag and Woodpecker runs kubectl set image from inside the cluster.
There are no in-cluster image builds or CI test runs anywhere — the
in-cluster Woodpecker buildkit and the fallback-build pattern were removed as a
clean cut (ADR-0002, 2026-06-13). The Forgejo container registry is frozen
and emptied — break-glass only.
This breaks the old circular dependency (images needed to repair the cluster used to be built and stored inside it) and keeps build IO + registry pushes off the homelab spindle.
Architecture Diagram
graph LR
A[git push Forgejo<br/>viktor/<repo> canonical] --> B[push-mirror sync_on_commit]
B --> C[GitHub mirror<br/>ViktorBarzin/<repo>]
C --> D[GitHub Actions<br/>.github/workflows/build.yml]
D --> E[lint / test]
E --> F[buildx linux/amd64<br/>provenance:false]
F --> G[push ghcr.io/viktorbarzin/<name><br/>:sha8 + :latest]
G --> H[svu tag -> Forgejo canonical]
G --> I[POST Woodpecker deploy repo]
I --> J[.woodpecker/deploy.yml<br/>event: manual]
J --> K[kubectl set image<br/>in-cluster SA cluster-admin]
K --> L[K8s Deployment<br/>pulls from ghcr]
style D fill:#2088ff
style J fill:#4c9e47
style G fill:#f39c12
Components
| Component | Location | Purpose |
|---|---|---|
| GitHub Actions | .github/workflows/build.yml (per repo) |
Build + lint + test + push image; trigger deploy; cut semver tag |
| ghcr.io | ghcr.io/viktorbarzin/* |
Container registry for ALL owned images (public + private packages) |
| Woodpecker CI | ci.viktorbarzin.me |
Deploy-only — kubectl set image in-cluster; plus infra applies + maintenance crons |
| Forgejo | forgejo.viktorbarzin.me/viktor/<repo> |
Canonical git source (push-mirrors to GitHub). Container registry FROZEN (break-glass only) |
| Pull-Through Cache | 10.0.20.10:5000/5010/5020/5030/5040 |
LAN cache for upstream registries (DockerHub, ghcr, Quay, k8s.gcr, Kyverno) |
| Kyverno | kyverno namespace |
Syncs ghcr-credentials (private-ghcr allowlist) + registry-credentials to namespaces |
| Vault | vault.viktorbarzin.me |
K8s auth for Woodpecker deploy pipelines; CI tokens in secret/ci/global + secret/viktor |
How It Works
The fleet pattern (every owned app)
- Canonical source = Forgejo
viktor/<repo>. A push-mirror (sync_on_commit) pushes every commit to the GitHub mirrorViktorBarzin/<repo>. The.github/workflows/build.ymlis committed on Forgejo and mirrors over. - GHA
buildjob (triggerson: push: branches: [master]ONLY — feature branches mirror but build/deploy nothing, the safety valve):- lint + test
svucomputes the nextvX.Y.Zfrom conventional commits and pushes the tag back to canonical Forgejo (GHA secretFORGEJO_GIT_TOKEN= write:repository PAT);VERSIONis baked into the imagedocker buildxlinux/amd64,provenance: false(single-manifest — avoids the orphaned-index-children failure class), pushghcr.io/viktorbarzin/<name>:<sha8>+:latestdelete-package-versionskeeps the newest ~10 ghcr versions
- GHA
deployjob POSTsci.viktorbarzin.me/api/repos/<id>/pipelines(the Woodpecker registration for the GitHub mirror, github-forge; GHA secretWOODPECKER_TOKEN) withIMAGE_TAG+IMAGE_NAME. .woodpecker/deploy.yml(event: manual only, so the raw Forgejo→GitHub mirror pushes don't fire a tag-less deploy) runskubectl set image deployment/<app> <container>=<image>in-cluster. Thewoodpecker-agentSA iscluster-admin, so thebitnami/kubectlstep needs no kubeconfig/RBAC. The Deployment image is inlifecycle.ignore_changes(KEEL_IGNORE_IMAGE) so the SHA tag sticks andterragrunt applydoesn't fight it. CronJobs in owned apps track:latest+imagePullPolicy: Alwaysinstead of a deploy step.
Keel stays enrolled as a redundant net (finds the deployed SHA already running → no-op).
Tooling: infra/scripts/offinfra-onboard + infra/scripts/offinfra-templates/
scaffold a repo onto this pattern (mirror, workflow, Woodpecker deploy repo,
old-pipeline removal, default-branch flip). Mirror + workflow commits go via
the Forgejo API over the internal Traefik LB
(curl --resolve forgejo.viktorbarzin.me:443:10.0.20.203) since the devvm
can't reach Forgejo's public hairpin.
ghcr package visibility
| Visibility | Packages | Pull mechanism |
|---|---|---|
| Public | beadboard, nextcloud-todos, claude-agent-service, claude-memory-mcp, kms-website, freedify, tuya_bridge, x402-gateway, chrome-service-novnc, android-emulator | Anonymous |
| Private | f1-stream, job-hunter, instagram-poster, payslip-ingest, wealthfolio-sync, fire-planner, recruiter-responder, tripit, infra-cli, infra-ci | ghcr-credentials dockerconfigjson |
Private-image pulls use the ghcr-credentials dockerconfigjson, cloned by the
kyverno stack's sync-ghcr-credentials ClusterPolicy to an explicit
ALLOWLIST of private-ghcr namespaces only (NOT cluster-wide; source
stacks/kyverno/modules/kyverno/ghcr-credentials.tf). Cred = Vault
secret/viktor/ghcr_pull_token (an alias of the admin github_pat — GitHub
has no token-mint API; swap the alias value if a scoped token is ever
UI-minted).
Migrated apps (issues #13–#27)
f1-stream, job-hunter, tuya_bridge, beadboard, nextcloud-todos,
claude-agent-service, claude-memory-mcp, kms-website, Freedify,
instagram-poster, payslip-ingest, broker-sync (image name wealthfolio-sync),
fire-planner, recruiter-responder, x402-gateway — plus tripit (the original
pilot, 2026-06-09). Earlier public-repo apps already on GHA (Website,
k8s-portal, apple-health-data, audiblez-web, plotting-book, insta2spotify,
audiobook-search, council-complaints) now also land on ghcr.
Infra-owned images (issues #29 / #30)
Images owned by the infra repo build on GHA workflows in the infra repo's own
.github/workflows/ (the github↔forgejo divergence was deliberately NOT
reconciled — the workflows were added to the GitHub lineage via PR):
| Image | Workflow | Destination |
|---|---|---|
| chrome-service-novnc | build-chrome-service-novnc.yml |
public ghcr.io/viktorbarzin/chrome-service-novnc |
| android-emulator | build-android-emulator.yml |
public ghcr.io/viktorbarzin/android-emulator |
| infra CLI | build-cli.yml |
DockerHub viktorbarzin/infra (kept) + ghcr.io/viktorbarzin/infra-cli |
| infra-ci | build-infra-ci.yml |
private ghcr.io/viktorbarzin/infra-ci |
infra-ci is the image the .woodpecker/default.yml apply step and
drift-detection.yml run in (proven by pipelines 165/166). chatterbox-tts is
already built by tripit's GHA → ghcr.
The Woodpecker build-ci-image.yml and build-cli.yml pipelines were
REMOVED. Break-glass for infra-ci is now a manual
.woodpecker/breakglass-infra-ci.yml (ghcr pull-and-save to the registry VM).
Forgejo container registry — FROZEN
Issue #32 wiped all viktor/* container packages (~19G reclaimed, /data
58%→20%). The registry is break-glass-only now; nothing pushes to it. The
forgejo-cleanup CronJob stays in DRY_RUN (nothing to clean). Pull-through
caches on the registry VM (10.0.20.10) are unchanged. See
docs/runbooks/forgejo-registry-breakglass.md.
Image registry / pull path
- Containerd
hosts.tomlredirects pulls from docker.io and ghcr.io to the pull-through cache at10.0.20.10(5000 = docker.io, 5010 = ghcr.io). - Pull-through cache serves cached images from the LAN, fetches upstream on a miss.
- Kyverno ClusterPolicies sync
ghcr-credentials(private-ghcr allowlist) andregistry-credentialsto namespaces.
Woodpecker — what it still runs
Woodpecker is deploy + cluster-touching steps only:
| Pipeline | File | Purpose |
|---|---|---|
| per-app deploy | .woodpecker/deploy.yml (each repo) |
kubectl set image + Slack notify (event: manual) |
| terragrunt apply | .woodpecker/default.yml |
Changed-stacks apply on push to master (runs in infra-ci) |
| certbot | .woodpecker/renew-tls.yml |
TLS renewal cron |
| drift-detection | .woodpecker/drift-detection.yml |
Nightly Terraform drift (runs in infra-ci) |
| provision-user | .woodpecker/provision-user.yml |
Add namespace-owner user from Vault spec |
| registry-config-sync | .woodpecker/registry-config-sync.yml |
SCP modules/docker-registry/* → 10.0.20.10 on change |
| pve-nfs-exports-sync | .woodpecker/pve-nfs-exports-sync.yml |
Sync scripts/pve-nfs-exports → /etc/exports on PVE |
| issue-automation | .woodpecker/issue-automation.yml |
Triage + respond to ViktorBarzin/infra GitHub issues |
| postmortem-todos | .woodpecker/postmortem-todos.yml |
Auto-resolve safe TODOs from new post-mortems |
| k8s-portal | .woodpecker/k8s-portal.yml |
Path-filtered deploy for the portal |
| breakglass-infra-ci | .woodpecker/breakglass-infra-ci.yml |
Manual ghcr pull-and-save of infra-ci to the registry VM |
No build/test pipeline exists on any repo. Do not (re)introduce one.
Woodpecker API
Uses numeric repo IDs (/api/repos/<id>/pipelines), NOT owner/name paths
(those return HTML). The deploy registration for each app is the GitHub
mirror repo (registered github-forge). IDs are stable across renames and must
be looked up from the Woodpecker UI/DB.
Woodpecker YAML gotchas
- Commands with
${VAR}:${VAR}must be quoted — an unquoted:triggers YAML map parsing when the vars are empty. - Use
bitnami/kubectl:latest(not pinned versions — entrypoint compatibility). - Global secrets must include
manualin their events list for API-triggered pipelines.
GitHub repo secrets
Per repo: WOODPECKER_TOKEN (POST the deploy pipeline), FORGEJO_GIT_TOKEN
(write:repository PAT for the svu tag push). ghcr push uses the workflow's
built-in GITHUB_TOKEN (packages: write).
Infra repo CI topology
The infra repo runs on Woodpecker via two forge registrations: the Forgejo
forge (repo id 82, registered 2026-06-08) and the legacy GitHub forge (repo id
1). Pushes to Forgejo master fire .woodpecker/default.yml
(changed-stacks terragrunt apply, in infra-ci) plus the notify-nonadmin-push
Slack audit step. Operational facts (2026-06-10):
- Webhook URL is the IN-CLUSTER service:
http://woodpecker-server.woodpecker.svc.cluster.local/api/hook?...(PATCHed via the Forgejo API). The Woodpecker default (https://ci.viktorbarzin.me/...) resolves to the non-proxied public A record from pods → NAT hairpin → intermittentcontext deadline exceeded, silently dropping push events. If Woodpecker "repairs" the repo it rewrites the hook back toci.viktorbarzin.me— re-apply the in-cluster URL. - Repo-scoped secrets must exist on BOTH repos: pipelines reference
repo-level secrets (
registry_ssh_key,pve_ssh_key,CLOUDFLARE_TOKEN, …). When registering a new forge repo for infra, clone the secret set too. - Empty commits defeat path filters: a commit with no changed files makes Woodpecker include ALL workflow files (path conditions can't exclude), so every repo secret must resolve. Normal commits with real files only compile the matching workflows.
The Forgejo trigger is not fully dependable — land infra changes by pushing
Forgejo master (as viktor), use [ci skip] for docs/no-op commits, and verify
deploys via scripts/tg + live cluster state rather than trusting the CI
checkmark. The two remotes have diverged (parallel histories under
different SHAs); expect github pushes to reject non-fast-forward and leave them
— never force-push.
Configuration
GitHub Actions (per-app .github/workflows/build.yml)
name: build
on:
push:
branches: [master]
jobs:
build:
runs-on: ubuntu-latest
permissions:
contents: write # svu tag push
packages: write # ghcr push
steps:
- uses: actions/checkout@v4
- name: lint + test
run: make lint test
- name: svu tag -> Forgejo
run: |
VERSION=$(svu next)
# ... push tag to canonical Forgejo with FORGEJO_GIT_TOKEN
- uses: docker/setup-buildx-action@v3
- uses: docker/build-push-action@v6
with:
platforms: linux/amd64
provenance: false
push: true
tags: |
ghcr.io/viktorbarzin/<name>:${{ github.sha }}
ghcr.io/viktorbarzin/<name>:latest
deploy:
needs: build
runs-on: ubuntu-latest
steps:
- name: Trigger Woodpecker deploy
run: |
curl -X POST https://ci.viktorbarzin.me/api/repos/<DEPLOY_REPO_ID>/pipelines \
-H "Authorization: Bearer ${{ secrets.WOODPECKER_TOKEN }}" \
-d '{"branch":"master","variables":{"IMAGE_TAG":"...","IMAGE_NAME":"..."}}'
Woodpecker deploy pipeline (per-app .woodpecker/deploy.yml)
when:
event: manual
steps:
deploy:
image: bitnami/kubectl:latest # uses the in-cluster woodpecker-agent SA (cluster-admin)
commands:
- "kubectl set image deployment/app app=${IMAGE_NAME}:${IMAGE_TAG} -n <ns>"
- "kubectl rollout status deployment/app -n <ns> --timeout=300s"
notify:
image: plugins/slack
when:
status: [success, failure]
CI/CD secrets sync
A CronJob in the woodpecker namespace pushes secret/ci/global from Vault →
the Woodpecker API every 6h, keeping global secrets in sync. Woodpecker deploy
pipelines authenticate to the cluster via the in-cluster woodpecker-agent SA
(cluster-admin); Vault K8s auth backs any secret reads.
Decisions & Rationale
Why all builds off-infra (ADR-0002)?
- Breaks the circular dependency — the images needed to repair the cluster no longer live inside it (they're on ghcr, an external registry).
- Removes build IO + registry push load from the contended homelab spindle.
- GHA is free on public repos and generous on private; buildx provenance:false sidesteps the orphaned-index-children failure class that plagued the in-cluster registry.
- Clean cut — no in-cluster fallback builds anywhere; one pattern, fleet-wide.
Why ghcr (not push back to Forgejo)?
Forgejo's container registry repeatedly orphaned OCI index children (2026-04-13/19, 2026-05-04, 2026-06-10) and its retention is not container-aware. ghcr is external (DR-safe), free for this scale, and has native multi-arch handling. The Forgejo registry was frozen + emptied (issue #32).
Why Woodpecker stays for deploy?
kubectl set image needs in-cluster privileged access; doing it from GHA would
mean exposing kube-apiserver or a long-lived kubeconfig. Woodpecker's
woodpecker-agent SA is already cluster-admin in-cluster — the deploy step
needs no credentials.
Why event: manual on deploy.yml?
The Forgejo→GitHub push-mirror sends raw, tag-less pushes to the GitHub mirror.
If deploy.yml fired on push, every mirror sync would trigger a deploy with no
image tag. manual means only the GHA deploy job's explicit API POST (with
IMAGE_TAG) deploys.
Why linux/amd64 only?
The cluster runs on x86_64 nodes only; ARM builds waste time and storage.
Troubleshooting
GHA build fails: ghcr push "denied"
The workflow GITHUB_TOKEN needs packages: write permission and the package
must allow the repo to push. Check the workflow permissions: block and the
package's "Manage Actions access" settings.
Image pull fails: "ErrImagePull" / "ImagePullBackOff"
# Public image — check the pull-through cache is up
curl http://10.0.20.10:5010/v2/_catalog
# Private image — verify the ghcr-credentials Secret exists in the namespace
kubectl get secret ghcr-credentials -n <namespace>
# It's Kyverno-synced to an allowlist; if missing, the namespace isn't on the
# allowlist in stacks/kyverno/modules/kyverno/ghcr-credentials.tf
If the cause is the internal-DNS hairpin (fresh pulls timing out on the public
Forgejo path), see the CoreDNS viktorbarzin.me carve-out in
docs/architecture/networking.md and docs/runbooks/registry-vm.md.
Deploy didn't happen after a push
Confirm the push was to master (feature branches build/deploy nothing).
Check the GHA run completed the deploy job, then check Woodpecker received the
manual pipeline (ci.viktorbarzin.me, the GitHub-mirror deploy repo). Verify
live with kubectl rollout status — not the CI checkmark.
Woodpecker deploy fails: "YAML: did not find expected key"
Unquoted command with ${VAR}:${VAR} syntax when a VAR is empty. Quote the
command (see the deploy.yml example above).
Related
- ADR:
../adr/0002-all-image-builds-off-infra-gha-ghcr.md— the decision - Databases Architecture — database credentials via Vault
- Multi-Tenancy — per-user Woodpecker access
- Runbook:
../runbooks/forgejo-registry-breakglass.md— using the frozen registry - Runbook:
../runbooks/registry-vm.md— pull-through cache VM + image-pull debugging - Onboarding tool:
../../scripts/offinfra-onboard+../../scripts/offinfra-templates/