Fly.io Deployment¶

Deploy the Pisama backend to Fly.io for low-latency, globally distributed hosting.

Backend Deployment¶

cd backend

# Create the app
fly launch --name pisama-api --no-deploy

# Set secrets
fly secrets set \
  JWT_SECRET=$(openssl rand -base64 32) \
  DATABASE_URL="postgresql+asyncpg://user:pass@host:5432/mao" \
  REDIS_URL="redis://host:6379" \
  ENVIRONMENT=production \
  CORS_ORIGINS="https://your-frontend.vercel.app"

# Deploy
fly deploy

Dockerfile¶

The backend/Dockerfile is production-ready:

Python 3.11-slim base image
Non-root user (mao) for security
CPU-only PyTorch for sentence-transformers (smaller image)
Built-in health check on /health
Configurable port via $PORT environment variable

Scaling¶

# Scale to 2 instances
fly scale count 2

# Increase memory (embedding models need ~1GB)
fly scale memory 1024

# View current scaling
fly scale show

Database Options¶

Fly.io does not provide managed PostgreSQL with pgvector. Options:

Provider	Recommended For	pgvector
Supabase	Hobby / Startup	Pre-installed
Neon	Serverless workloads	Available
AWS RDS	Enterprise	Manual enable
Fly Postgres	Simple setup	Manual enable

Ensure you enable the pgvector extension:

CREATE EXTENSION IF NOT EXISTS vector;

Frontend Deployment¶

Deploy the frontend to Vercel (recommended) or as a separate Fly.io app:

Vercel (Recommended)Fly.io

cd frontend
vercel --prod

Set NEXT_PUBLIC_API_URL to your Fly.io backend URL.

cd frontend
fly launch --name pisama-web --no-deploy
fly secrets set \
  NEXT_PUBLIC_API_URL="https://pisama-api.fly.dev/api/v1" \
  NEXTAUTH_SECRET=$(openssl rand -base64 32) \
  NEXTAUTH_URL="https://pisama-web.fly.dev"
fly deploy

Health Check¶

curl https://pisama-api.fly.dev/health
# {"status": "healthy", "database": "ok", "redis": "ok"}

Monitoring¶

Fly.io provides built-in logging and metrics:

# View logs
fly logs

# View metrics
fly dashboard

Pisama also exposes Prometheus metrics:

curl https://pisama-api.fly.dev/api/v1/metrics