stwaidele/dorfteich
1
0
Fork 0
Code Issues 2 Pull Requests Actions Packages Projects Releases 22 Wiki Activity
64f2deb40f
dorfteich/docs/architecture/adr/0014-gitea-actions-cicd.md
Claude Fable 5 69e7b1406b
Some checks failed
CD / Build and push images (push) Successful in 1m9s
Details
CI / Lint, typecheck, test (push) Failing after 1m11s
Details
CI / Auth e2e pack (push) Has been skipped
Details
CI / Import/export fidelity gate (push) Has been skipped
Details
CI / Build container images (push) Has been skipped
Details
CD / Deploy to Test (push) Successful in 9s
Details
CD / Smoke tests against Test (push) Successful in 1m13s
Details
CD / Promote to Int (push) Successful in 11s
Details
Move the stage directories to /srv/DOCKER (host consolidation) 
ONE consolidated every Docker stack under /srv/DOCKER (BASEL
convention, tracked in stwaidele/infrastructure-one); the three
Dorfteich stages follow. Deploy targets in cd.yml (test/int) and
prod-deploy.yml plus the docs now point at /srv/DOCKER/dorfteich-<stage>.
Data lives in named volumes keyed by the unchanged compose project
name, so the directory move carries no data migration. The go-live
checklist's uptime-kuma path already lives under /srv/DOCKER — the doc
just catches up.

Deliberately committed together with the host-side move: this commit
must not deploy before the directories exist at the new path.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-14 12:43:29 +02:00

2.9 KiB
Raw Blame History

ADR 0014: CI/CD with Gitea Actions, staged promotion

  • Status: accepted
  • Date: 2026-07-04

Context

The project is hosted on a self-managed Gitea (gitea.101010.cloud). Environments (kickoff): Dev runs locally on contributors' machines; Test and Int run as separate Compose stacks on the operator's dedicated server ONE (one.101010.cloud, domains test.dorfteich.cloud / int.dorfteich.cloud), which also hosts Gitea itself; Prod (dorfteich.online) goes live on the same host at M8 — the architecture must keep any later move cheap. Contributors include AI coding sessions — gates must be automated and objective wherever possible. (Originally the stages and runner lived on a shared 4-GB VPS, 188.245.116.44; everything moved to ONE on 2026-07-11.)

Decision

  • Gitea Actions is the CI/CD system (GitHub-Actions-compatible syntax); an act_runner runs on ONE (one.101010.cloud) with Docker access.
  • Images are built once per change and promoted, never rebuilt per stage: pushed to the Gitea container registry (gitea.101010.cloud/stwaidele/dorfteich-{web,api,collab}), tagged with the git SHA plus moving tags test, int, and semver tags for releases.
  • Pipeline (details in deployment.md):
    1. PR / push: lint, typecheck, unit tests, build; PRs must be green to merge into main.
    2. Merge to main: build + push images (SHA tag) → deploy to Test automatically → run the Playwright e2e suite against Test.
    3. Promotion to Int: automatic when e2e on Test is green — Int is the stable preview environment (same images, int tag).
    4. Promotion to Prod: manual gate — creating a release tag (vX.Y.Z) triggers the Prod deploy after a required manual approval in the workflow. Database migrations run automatically on container start (ADR 0006); releases with breaking migrations must say so in the release notes.
  • Deploy mechanism: the runner executes docker compose pull && docker compose up -d in the stage's directory (/srv/DOCKER/dorfteich-<stage>/) over SSH (deploy key per stage). Moving Prod to a dedicated host later changes only that SSH target.
  • Self-hosters consume the same release images via a published docker-compose.yml pinned to semver tags (update strategy in operations.md).

Consequences

  • One objective quality bar (green e2e on Test) decides Int promotion; the only human gate is the Prod release — matching the risk profile.
  • The registry, runner, and repo live on the same Gitea — no third-party CI dependency.
  • e2e stability becomes load-bearing; flaky tests block the pipeline and must be treated as defects (stated in the contributor conventions).

Alternatives considered

  • GitHub Actions with repo mirror: splits source of truth; rejected.
  • Manual deploys per stage: does not scale to many small stories and invites drift between stages.