Deployment¶

Jak deployovat backend, worker, frontend, dokumentaci.

Backend + Worker (Sliplane)¶

Auto-deploy z GitHub `main`¶

Sliplane má GitHub integration enabled per service:

https://sliplane.io → service settings → Repository
  url: https://github.com/Algawebbusiness/nemoreport-ai-backend-v2
  branch: main
  autoDeploy: true

Kdykoliv někdo pushne na main: 1. Sliplane build agent (Hetzner-based) cloneuje repo 2. Build Docker image z Dockerfile 3. Push do registry-01.nbg1.ger.sliplane.io/... 4. Pull + restart container 5. Health check /health

Build čas typicky 60-90s (cached layers).

Dockerfile¶

FROM docker.io/library/python:3.12-slim
COPY --from=ghcr.io/astral-sh/uv:latest /uv /uvx /bin/
WORKDIR /app
RUN apt-get update \
    && apt-get install -y --no-install-recommends libmagic1 \
    && rm -rf /var/lib/apt/lists/*
COPY pyproject.toml uv.lock* README.md ./
RUN uv sync --frozen --no-install-project --no-dev || uv sync --no-install-project --no-dev
COPY app ./app
COPY prompts ./prompts
COPY skills ./skills
COPY fixtures ./fixtures
RUN uv sync --frozen --no-dev || uv sync --no-dev

Stejný image pro backend i worker — liší se jen cmd:

Service	CMD
`nemoreport-ai-backend-v2`	`uv run uvicorn app.main:app --host 0.0.0.0 --port 8000`
`nemoreport-ai-worker-v2`	`uv run python -m app.worker_entry`

Environment variables¶

ENV se nastavuje per service v Sliplane UI nebo via API. Strict-validated v app/config.py (fail-fast on missing).

Required pro backend + worker (oba mají stejný set):

ENV	Příklad	Secret?
`SUPABASE_URL`	`https://cubdrgjdkatyecrgckwp.supabase.co`	ne
`SUPABASE_ANON_KEY`	`sb_publishable_...`	ne
`SUPABASE_SERVICE_ROLE_KEY`	`eyJ...`	ano
`GEMINI_API_KEY`	`AIza...`	ano
`MISTRAL_API_KEY`	`...`	ano
`COHERE_API_KEY`	`...`	ano
`ADMIN_HASH`	`3bddbafdec...` (sha1 of secret)	ano
`NETTE_HMAC_SECRET`	`...`	ano (jen backend)
`EMBEDDING_MODEL`	`gemini-embedding-2`	ne
`EMBEDDING_MODEL_VERSION`	`ga-2026-04`	ne
`COHERE_RERANK_MODEL`	`rerank-v4.0-pro`	ne
`COHERE_RERANK_ENABLED`	`true`	ne
`LLM_MODEL`	`google-gla:gemini-3-flash-preview`	ne
`ALLOWED_ORIGINS`	`https://nemoreport-ai-frontend-v2.algaweb.workers.dev,http://localhost:3000`	ne
`RATE_LIMIT_STORAGE_URI`	`redis://nemoreport-redis.internal:6379/0`	ne
`REDIS_URL`	`redis://nemoreport-redis.internal:6379/1`	ne
`LOG_LEVEL`	`INFO`	ne
`APP_ENV`	`production`	ne

Update env vars¶

Per Sliplane API quirk (app/CLAUDE.md note): - updateService PUT s value="" smaže ENV var - Pro secret keys vždy passnout konkrétní hodnotu, ne "" - secret=true v getService response zobrazí jako value="" (masking, ne reálná empty)

Aktualizace přes MCP: mcp__sliplane__updateService s celým env arrayí (full replace).

Manual restart¶

# Restart backend
sliplane services restart nemoreport-ai-backend-v2

# Or via UI: Services → Click service → Restart

Logs¶

# Tail logs
sliplane services logs nemoreport-ai-backend-v2 --follow

# Or via MCP:
# mcp__sliplane__getServiceLogs(projectId, serviceId, from=<unix_ts>)

Resource scaling¶

Sliplane pricing: per-service CPU + RAM. Aktuálně každý service 1 CPU × 1 GB RAM (default).

Pro scale-up:

sliplane services rescale nemoreport-ai-backend-v2 --cpu 2 --ram 2GB

Rollback¶

Sliplane drží předchozí Docker images v registry. Rollback: 1. Find předchozí commit hash v Sliplane events log 2. git revert na předchozí commit 3. git push origin main → auto-deploy

Redis (Sliplane)¶

service: nemoreport-redis
image: redis:7-alpine
internal_dns: nemoreport-redis.internal:6379
persistence: AOF (append-only file)

Žádné public exposure. Backend + worker komunikují přes internal Sliplane network.

Backup strategy: Redis AOF persistuje na disk volume. Pro DR je potřeba volume snapshot (Sliplane managed).

Frontend (Cloudflare Workers)¶

Manual deploy¶

CF Pages auto-deploy nepoužíváme. Frontend deploy je manual přes wrangler:

cd /home/jiri/Nextcloud/vibecode/Nemoreport_v2/nemoreport-ai-frontend-v2

# Build (Next.js + OpenNext adapter)
npm run build  # → .next/ + .open-next/

# Deploy na CF Workers
npm run deploy  # = wrangler deploy

# Output:
#   Total Upload: 6458.55 KiB / gzip: 1348.60 KiB
#   Worker Startup Time: 35 ms
#   Uploaded nemoreport-ai-frontend-v2 (22.44 sec)
#   Deployed: https://nemoreport-ai-frontend-v2.algaweb.workers.dev
#   Current Version ID: <uuid>

Konfigurace¶

wrangler.jsonc:

{
  "name": "nemoreport-ai-frontend-v2",
  "main": ".open-next/worker.js",
  "compatibility_date": "2025-09-23",
  "compatibility_flags": ["nodejs_compat"],
  "assets": {
    "directory": ".open-next/assets",
    "binding": "ASSETS"
  }
}

open-next.config.ts:

import { defineCloudflareConfig } from "@opennextjs/cloudflare";

export default defineCloudflareConfig({
  // Default config
});

Environment variables¶

Setup v wrangler.jsonc (vars, not secrets):

{
  "vars": {
    "NEXT_PUBLIC_SUPABASE_URL": "https://cubdrgjdkatyecrgckwp.supabase.co",
    "NEXT_PUBLIC_SUPABASE_ANON_KEY": "sb_publishable_...",
    "NEXT_PUBLIC_BACKEND_URL": "https://nemoreport-ai-backend-v2.sliplane.app"
  }
}

Žádné secrets na FE — SUPABASE_SERVICE_ROLE_KEY zůstává jen na BE.

Rollback¶

Cloudflare drží versions:

# List versions
wrangler versions list

# Rollback
wrangler versions rollback <version-id>

Nebo přes CF Workers Dashboard: Workers > nemoreport-ai-frontend-v2 > Deployments > Rollback.

Supabase¶

Migrace¶

# Local dev
cd /home/jiri/Nextcloud/vibecode/Nemoreport_v2/nemoreport-ai-backend-v2
supabase link --project-ref cubdrgjdkatyecrgckwp
supabase db push

# Or inline via MCP:
# mcp__claude_ai_Supabase__apply_migration(project_id, name, query)

Config (auth, redirect URLs, SMTP)¶

supabase/config.toml je source of truth. Push:

supabase config push --yes

Změní: - api.schemas (PostgREST exposed schemas) - auth.hooks (custom JWT hook function) - auth.email.template (custom magic link template) - SMTP config s RESEND_API_KEY env substituci - auth.redirect_urls (allowed redirect destinations)

Backups¶

Supabase Pro plán: - PITR (Point-in-time recovery) — 7 dní default - Daily backups — týden retention

Pro produkci doporučeno upgrade Free → Pro ($25/mo) — PITR essential pro disaster recovery.

Edge Functions (NEPOUŽÍVÁME zatím)¶

Phase B+ jsme se rozhodli use Sliplane worker místo Supabase Edge Functions kvůli: - Lepší debugging (full Python ekosystém) - Žádné cold starts - Lépe handle long-running OCR (15 min timeouty)

Dokumentace¶

Lokální dev¶

cd /home/jiri/Nextcloud/vibecode/Nemoreport_v2

# Install (jednorazově)
uv tool install mkdocs-material

# Dev server s live reload
mkdocs serve  # http://127.0.0.1:8000

# Static build
mkdocs build  # → site/

Hosting options¶

Option 1: Cloudflare Pages (recommended)

# Build
mkdocs build

# Deploy
wrangler pages deploy site/ --project-name=nemoreport-docs

Option 2: GitHub Pages

# Push docs to gh-pages branch
mkdocs gh-deploy --force

Option 3: Sliplane static

Add new service s nginx serving site/ directory.

Doporučeno: Cloudflare Pages, custom doména docs.nemoreport.algaweb.cz.

Lokální development¶

Backend¶

cd /home/jiri/Nextcloud/vibecode/Nemoreport_v2/nemoreport-ai-backend-v2

# Install deps
uv sync

# Setup .env.local s minimum:
cat > .env.local <<EOF
SUPABASE_URL=https://cubdrgjdkatyecrgckwp.supabase.co
SUPABASE_ANON_KEY=...
SUPABASE_SERVICE_ROLE_KEY=...
GEMINI_API_KEY=...
ADMIN_HASH=test_local
RATE_LIMIT_STORAGE_URI=memory://
APP_ENV=development
EOF

# Run dev server
uv run uvicorn app.main:app --reload --port 8000

# Run worker (separate terminal)
uv run python -m app.worker_entry
# Note: needs Redis. Use Sliplane internal nebo lokální docker run redis.

Frontend¶

cd /home/jiri/Nextcloud/vibecode/Nemoreport_v2/nemoreport-ai-frontend-v2

npm install
npm run dev  # http://localhost:3000

# Setup .env.local:
cat > .env.local <<EOF
NEXT_PUBLIC_SUPABASE_URL=https://cubdrgjdkatyecrgckwp.supabase.co
NEXT_PUBLIC_SUPABASE_ANON_KEY=sb_publishable_...
NEXT_PUBLIC_BACKEND_URL=http://localhost:8000
EOF

Deploy checklist (production go-live)¶

Před production:

Disaster recovery¶

Scenario 1: Sliplane outage¶

Backend + worker down → app nefunguje
Mitigation: deploy na backup provider (Render / Fly.io) — 2 dny prep
RTO: 4 hodiny (manual restore from Docker images)
RPO: 0 (Postgres je nezávislé na Sliplane)

Scenario 2: Supabase outage¶

DB + Auth down → app nefunguje
Mitigation: vendor's responsibility (status page); Pro plán má SLA
RTO: depends on Supabase
Recovery: PITR pokud DB corruption

Scenario 3: Cloudflare outage¶

Frontend down (FE Worker)
Mitigation: rare (Cloudflare tier-1 reliability)
RTO: depends on CF
Backup: Vercel deploy ready (@opennextjs works na Vercel)

Scenario 4: Vendor data loss¶

Supabase delete project / corruption
Mitigation: PITR (7d), týdenní DB dump do CF R2 (offsite)
Implementation: cron job na týdenní pg_dump → upload na R2

Scenario 5: Code repo loss¶

GitHub + GitLab oba lost (extremely unlikely)
Mitigation: dual-push remote (git push origin main → oba simultaneously)
Local clones na dev machines + Nextcloud sync (/home/jiri/Nextcloud/vibecode/)

Deployment¶

Backend + Worker (Sliplane)¶

Auto-deploy z GitHub main¶

Dockerfile¶

Environment variables¶

Update env vars¶

Manual restart¶

Logs¶

Resource scaling¶

Rollback¶

Redis (Sliplane)¶

Frontend (Cloudflare Workers)¶

Manual deploy¶

Konfigurace¶

Environment variables¶

Rollback¶

Supabase¶

Migrace¶

Config (auth, redirect URLs, SMTP)¶

Backups¶

Edge Functions (NEPOUŽÍVÁME zatím)¶

Dokumentace¶

Lokální dev¶

Hosting options¶

Lokální development¶

Backend¶

Frontend¶

Deploy checklist (production go-live)¶

Disaster recovery¶

Scenario 1: Sliplane outage¶

Scenario 2: Supabase outage¶

Scenario 3: Cloudflare outage¶

Scenario 4: Vendor data loss¶

Scenario 5: Code repo loss¶

Auto-deploy z GitHub `main`¶