Files
AndreaSlopFotografie/docs/deploy-and-ci.md
T
SluijsensandClaude Sonnet 5.5 6475696709 Pipeline: new names for repo variables and secrets; no OFFERINGS_URL; the post-deploy check always runs as CMS host
TEST_DEPLOY_PATH, TEST_HEALTH_URL, PROD_HEALTH_URL, PROD_DEPLOY_PATH and
PROD_HOST_* replace the old names (decision of Bryan, 2026-11-18). The test
build no longer fetches packages (the host injects them) and ASF_TEST_SERVED_BY_CMS
is gone: test is always served by the CMS host. PI_MAIN_* stay until the
TEST_HOST_* secrets and the SSH key exist.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-11-18 19:20:19 +01:00

8.4 KiB

Deploy and CI

Pipeline: .gitea/workflows/continuous_integration.yaml. Branch strategy and rollback are in aidlc-docs/features/frontend-shell/operations/deployment/ and aidlc-docs/features/prerendering/.

Triggers

Trigger Result
Pull request Checks only, no deploy.
Push to master Checks, then deploy to test.
Manual (workflow_dispatch), any branch Checks, and a deploy to test if deploy_test is on.
Manual from a release/* branch Deploy to production if deploy_production is on.

Workflow inputs

  • deploy_test (boolean, default false): "Deploy to Test". Builds the test artifact and deploys it to ASF test.
  • deploy_production (boolean, default false): "Deploy to Production". Only works from a release/* branch and only with Bryan's explicit go.

Dispatch from the CLI (the token comes from your git credentials, never print it): POST /api/v1/repos/SLP_Software/AndreaSlopFotografie/actions/workflows/continuous_integration.yaml/dispatches with {"ref": "<branch>", "inputs": {"deploy_test": "true"}}.

Jobs

  1. check: lint, format, unit/property/axe tests (coverage threshold 100% on src/lib/**), build (with the committed package snapshot), audit. Uploads artifact dist (this is what production gets).
  2. build-test: builds again with VITE_CONTACT_FORM_ENABLED=true and VITE_SITE_URL=https://test.andreaslop.nl (see "Own address of the test site") and uploads dist-test.
  3. deploy-test: needs check and build-test, downloads dist-test, uploads to the ASF test host and runs the deployment check (served by the CMS host, admin files, POST status).
  4. lighthouse: against the built site with the preview stub (--audit-route --contact-stub).
  5. deploy-production: manual, from release/* only, uses artifact dist.

There is one Raspberry Pi runner plus a desktop runner, so jobs can wait in the queue for a long time (for example 30 minutes before build-test); a deploy that only just started is not hung. Check started_at (UTC) before cancelling anything.

Deploying a feature branch to test

Committing, pushing and deploying a feature branch to test is allowed for the sessions that work on it. Merging, deploying to production and changing repo variables are not (Bryan decides). After a test deploy, check the page in a real browser against the real test CMS; web/e2e/ holds the checks that run in CI.

Contact form switch (VITE_CONTACT_FORM_ENABLED)

The contact form is closed by default. Only the test build (build-test) sets VITE_CONTACT_FORM_ENABLED=true, so the ASF test host shows the real form; the production artifact (job check) is built without it and stays closed. Only the exact value true opens the form (src/lib/contactFormFlag.ts); anything else keeps it closed, so a typo cannot open it in production. This replaced the former TEST ONLY commit d449e76 (which hard-coded enabled: true); it was reverted before the merge, so nothing has to be undone any more. Opening the form in production needs Bryan's explicit go after the captcha decision (see go-live-checklist.md). Locally: VITE_CONTACT_FORM_ENABLED=true pnpm build.

The check after a deploy and the host injection

deploy/verify-deployment.sh compares what the server sends with the files in dist. Once the CMS host injects (packages, banners, prices, JSON-LD price, site address), the pages differ from dist by design, so both sides are first normalised with config/normalizeInjected.ts (the contents of the CMS regions, the data block, offers in structured data, inline prices and the site address are taken out). Everything else must still be equal byte for byte, so an old build or a wrong server config is still caught; the 404 page is not injected and is compared as it is. The preview server can stand in for the host (node config/previewServer.ts --cms-data=<json>), which is how the script is tried locally.

Own address of the test site

The test site stands on its own: in the test build VITE_SITE_URL is https://test.andreaslop.nl, so canonical, og:url, og:image, the JSON-LD ids and urls and sitemap.xml all point to the test host. A link shared from the test site (WhatsApp, Discord) therefore shows its own picture, not one from production. The production artifact (job check) is built without the variable and keeps https://andreaslop.nl. src/lib/siteUrl.ts only accepts an origin (https, or http on localhost); anything else falls back to production, so a typo cannot move production to another domain. This is a build setting, not TEST ONLY code: nothing has to be reverted before a merge. robots.txt on the test host is the nginx one (Disallow: /, plus X-Robots-Tag: noindex), not public/robots.txt, so the test site stays out of search engines. To try it locally: VITE_SITE_URL=http://localhost:4173 pnpm build.

Generated files

web/deploy/iis/web.config and web/deploy/nginx/*.conf are generated by pnpm server-config from web/config/serverConfig.ts. Change the generator, rerun it and commit the result; a test checks that they match.

No nightly build (since the host injection)

There used to be a nightly test build (schedule, 03:15 UTC) that baked the CMS data into the HTML for search engines. It is gone: the CMS host writes the packages, banners and hero into the first HTML of every request (cache 60 s, refreshed at once on a change; NFR-10), so a content change is visible without a rebuild. A rebuild is only needed for code and design. The nightly run also put master over ASF-test every night, which would undo a feature branch deployed there.

The production site gets the same only when the ASF production site is served by the CMS host (the IIS pipeline, later; see go-live-checklist.md). Until then production shows the committed snapshot. The test build no longer sets OFFERINGS_URL (removed 2026-11-18, once the host injection was confirmed on ASF-test); it still sets VITE_SITE_URL (own address of the test site) and VITE_CONTACT_FORM_ENABLED.

Names of the repo variables and secrets

Format {ENV}_{GROUP}_{NAME} with ENV = TEST or PROD, no app part (this repo only deploys the ASF site) and no VITE (the pipeline maps a value to VITE_... in the build). Decision of Bryan, 2026-11-18; the full list is in the SLS repo (aidlc-docs/_shared/gitea-naamgeving-variabelen.md). Repo variables and secrets are created by SLS, not by feature sessions.

Name Kind Used by
TEST_DEPLOY_PATH variable deploy-test: target folder on the Pi
TEST_HEALTH_URL variable deploy-test: check after the deploy
PROD_DEPLOY_PATH variable deploy-production: remote folder (default wwwroot)
PROD_HEALTH_URL variable deploy-production: check after the deploy
PROD_HOST_ADDRESS, PROD_HOST_USERNAME, PROD_HOST_PASSWORD secrets deploy-production: FTP (plain FTP, no FTPS, at the hosting partner)
PI_MAIN_ADDRESS, _PORT, _USERNAME, _PASSWORD secrets deploy-test: still the old names; they become TEST_HOST_* with an SSH key (SLS arranges the key)

Removed: DEPLOY_PATH_TEST, HEALTH_CHECK_URL_TEST, HEALTH_CHECK_URL_PRODUCTION, FTP_REMOTE_DIR, FTP_HOST, FTP_USERNAME, FTP_PASSWORD (renamed), ASF_OFFERINGS_URL_TEST and ASF_TEST_SERVED_BY_CMS (no longer needed). The old names in the repo are removed by SLS after a green run.