Docker auto-sets HOSTNAME to the container id; Next standalone's server.js
binds process.env.HOSTNAME, so it listened on the container hostname and the
compose healthcheck (TCP 127.0.0.1:3000) was refused → frontend container
chronically "(unhealthy)" (cosmetic: Caddy reaches it by service name and the
site serves 200). Pin HOSTNAME=0.0.0.0 — the canonical Next.js Docker fix.
Diagnosed on prod 2026-06-05.
Problem
- GlitchTip Issues = 0 за всё время, хотя backend + frontend SDK интегрированы (PR #207, #208).
- Старый Sentry.io продолжает получать события — user видит уведомления оттуда.
Root cause
- frontend/Dockerfile не имеет ARG NEXT_PUBLIC_GLITCHTIP_DSN → `npm run build` бьёт
с пустым env var → Next.js инлайнит undefined → SDK init guard `if (dsn)` skips.
Chrome-devtools check на prod bundle подтвердил: ни в одном из 9 chunks DSN-строка
не запечена; `i.env.NEXT_PUBLIC_GLITCHTIP_DSN` evaluates to undefined.
- .forgejo/workflows/deploy.yml build-frontend не передавал build-args.
- На VPS backend/.env.runtime содержит legacy SENTRY_DSN=...@sentry.io/...
config.py:_promote_legacy_sentry_dsn слепо промоутит его в glitchtip_dsn → SDK
шлёт в чужой Sentry. GLITCHTIP_DSN там не задан.
- deploy.yml SSH-скрипт никогда не редактировал SENTRY_DSN/GLITCHTIP_DSN в .env.runtime.
Solution
1. frontend/Dockerfile: ARG NEXT_PUBLIC_GLITCHTIP_DSN + NEXT_PUBLIC_ENVIRONMENT
с пустыми defaults, ENV-mirror перед `npm run build`. Локальный build без
build-args работает по-прежнему (no-op DSN).
2. .forgejo/workflows/deploy.yml:
- build-frontend: `build-args` передаёт `secrets.GLITCHTIP_FRONTEND_DSN`
+ NEXT_PUBLIC_ENVIRONMENT=production. Инвалидирует cache → frontend
image пересоберётся (expected).
- deploy step: GLITCHTIP_BACKEND_DSN через secret, в SSH-скрипте:
a) `sed -i '/^SENTRY_DSN=/d' backend/.env.runtime` — снести legacy
b) upsert GLITCHTIP_DSN (sed/printf) тем же паттерном что SENTRY_RELEASE
c) `compose up -d --force-recreate --no-deps backend worker beat` —
обычный `up -d` не перечитывает env_file без image change.
3. backend/app/core/config.py: _promote_legacy_sentry_dsn ужесточён —
принимает SENTRY_DSN только если host == errors.gendsgn.ru. Для других
URLs (sentry.io) выдаёт UserWarning и НЕ промоутит. Anti-regression на
случай если SENTRY_DSN снова окажется в .env.runtime после ручного
вмешательства.
Required Forgejo secrets (Settings → Actions → Secrets)
- GLITCHTIP_BACKEND_DSN = https://3d6e291003e142458957490c83559867@errors.gendsgn.ru/1
- GLITCHTIP_FRONTEND_DSN = https://5d7bc85e300c4e80a8554ccc818ff56d@errors.gendsgn.ru/2
DSNs публичны (видны в browser bundle) — secrets ради build-time injection,
не для конфиденциальности. Если secrets не заданы → deploy succeeds, SDK
no-op, без регрессии.
Test plan
- Verify Forgejo deploy.yml зелёный после merge
- chrome-devtools: открыть gendsgn.ru → search bundle на DSN string → должна
быть запечена строка errors.gendsgn.ru/2
- Trigger frontend error → POST к errors.gendsgn.ru/api/2/envelope/
- Backend: curl на endpoint вызывающий 500 → событие в GlitchTip backend project
- GlitchTip dashboard https://errors.gendsgn.ru/gendesign/issues — Issues > 0
References
- vault: meta/00_credentials.md (DSNs + incident notes 2026-05-16)
- vault: decisions/Dec_GlitchTip_Frontend_Sentry_SDK.md (env contract)
- vault: fixes/fixes-MOC.md (#204 backend SDK init)
Split single build-and-deploy job into 4 jobs: build-backend,
build-worker, build-frontend run in parallel (all 3 independent),
deploy job has needs:[...] on all three. Removes ~2 min of sequential
build time on warm cache runs.
Also: scope GHA cache keys per image (backend-lean / worker-chromium /
frontend) so builds don't invalidate each other's layer cache. Replace
sleep 8 health-check with a 30-iteration polling loop (fails fast on
bad deploy, doesn't over-wait on fast ones). Add --mount=type=cache to
frontend npm ci for node_modules cache across builds.
- backend/Dockerfile: split builder/runner, runtime libs only, non-root app user, curl for healthcheck, --frozen via new uv.lock
- frontend/Dockerfile: npm ci instead of npm install (deterministic), USER node
- docker-compose.prod.yml: working backend healthcheck (curl now in image), redis healthcheck, drop dead NEXT_PUBLIC_API_BASE_URL env (Next.js bakes NEXT_PUBLIC_* at build time, runtime override is no-op)
- docker-compose.yml: same redis healthcheck, BACKEND_URL for next.config rewrites, drop uv from CMD (not in new image)
- Caddyfile: handle (not handle_path) so /api prefix is preserved into FastAPI router
- next.config.ts: rewrite /api/* and /health to BACKEND_URL in dev (no Caddy locally)
- frontend/src/lib/api.ts: empty default = same-origin relative URLs
- Makefile: drop uv run from migrate target
- Add backend/uv.lock and frontend/package-lock.json for reproducible builds
Verified: docker build succeeds for both, backend container starts, /health responds, curl healthcheck works inside image, container runs as non-root.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
