Skip to content

DEPLOY.md

Latest commit

 

History

History
155 lines (115 loc) · 3.92 KB

DEPLOY.md

File metadata and controls

155 lines (115 loc) · 3.92 KB
Raw

🚀 Deployment Guide

Deploy the Demand Forecasting API to production in minutes.

Option 1: Render (Recommended — Free Tier)

One-click deploy with PostgreSQL, Redis, and automatic HTTPS.

Quick Deploy

  1. Fork this repo to your GitHub account
  2. Go to render.com → Blueprints → New Blueprint Instance
  3. Connect your GitHub repo — Render auto-detects render.yaml
  4. Click Apply — deploys in ~5 minutes

What gets created:

  • ✅ FastAPI web service (512MB, auto-scaling ready)
  • ✅ PostgreSQL with TimescaleDB (1GB free, upgradeable)
  • ✅ Redis cache (25MB free)
  • ✅ Automatic HTTPS via *.onrender.com
  • ✅ Auto-deploy on every git push to master
  • ✅ Health checks + zero-downtime deploys
  • ✅ Admin API key auto-generated

After Deploy

# Your API is live at:
https://demand-forecast-api.onrender.com

# Check health
curl https://demand-forecast-api.onrender.com/api/v1/health

# Train the model (once deployed)
curl -X POST https://demand-forecast-api.onrender.com/api/v1/admin/retrain \
  -H "Authorization: Bearer YOUR_ADMIN_KEY"

# Make a forecast
curl -X POST https://demand-forecast-api.onrender.com/api/v1/forecast/demand \
  -H "Content-Type: application/json" \
  -d '{"product_id": "SKU-12345", "horizon_days": 14}'

# Dashboard
curl https://demand-forecast-api.onrender.com/api/v1/dashboard/summary?days=30

Option 2: Docker Compose (Self-Hosted)

Full stack: API + Postgres + Kafka + Redis + Grafana.

git clone https://github.com/twomathematicians-code/demand-forecasting.git
cd demand-forecasting

# Copy and configure environment
cp .env.example .env
# Edit .env: set DF_ADMIN_API_KEY, DF_KATZILLA_API_KEY, etc.

# Start all services
docker compose up -d

# Verify
curl http://localhost:8000/api/v1/health

Services:

  • API: http://localhost:8000
  • Grafana: http://localhost:3000 (admin/admin)
  • Postgres: localhost:5432
  • Kafka: localhost:9092
  • Redis: localhost:6379

Option 3: Kubernetes

# Apply manifests
kubectl apply -f k8s/deployment.yaml

# Create secrets
kubectl create secret generic demand-forecast-secrets \
  --from-literal=DF_DB_PASSWORD=your_db_pass \
  --from-literal=DF_ADMIN_API_KEY=your_admin_key \
  -n demand-forecasting

# Check deployment
kubectl get pods -n demand-forecasting
kubectl port-forward svc/demand-forecast-svc 8000:80 -n demand-forecasting

Security Configuration

Required Environment Variables

Variable Purpose Required
DF_ADMIN_API_KEY Protects /admin/retrain endpoint ✅ Production
DF_DB_PASSWORD PostgreSQL password ✅
DF_KATZILLA_API_KEY Katzilla data integration Optional
DF_CORS_ORIGINS Allowed CORS domains (comma-separated) Recommended

Production Checklist

  • Set DF_ENVIRONMENT=production
  • Set a strong DF_ADMIN_API_KEY (32+ chars)
  • Configure DF_CORS_ORIGINS to your frontend domain
  • Enable Redis: DF_REDIS_ENABLED=true
  • Review rate limits (default: 200 req/min)
  • Set up a custom domain in Render
  • Configure Grafana alerting (if self-hosting)

Environment-Specific Configs

Development (local)

DF_ENVIRONMENT=development
DF_REDIS_ENABLED=false
DF_KAFKA_CONSUMER_ENABLED=false
DF_DRIFT_CHECK_ENABLED=false

Staging (Render free tier)

DF_ENVIRONMENT=staging
DF_REDIS_ENABLED=true
DF_RATE_LIMIT_MAX=50

Production (Render paid / self-hosted)

DF_ENVIRONMENT=production
DF_REDIS_ENABLED=true
DF_KAFKA_CONSUMER_ENABLED=true
DF_DRIFT_CHECK_ENABLED=true
DF_RATE_LIMIT_MAX=200
DF_API_WORKERS=4

Monitoring

  • Render Dashboard: CPU, memory, request latency, error rate
  • Health endpoint: GET /api/v1/health — model status + uptime
  • Prometheus metrics: GET /metrics — request counts, latency percentiles
  • Logs: Structured JSON logging with request IDs — tail via render logs or docker logs