Fly.io Deployment Guide for ContextForge

⚙️ Fly.io Deployment Guide for ContextForge

This guide covers the complete deployment workflow for the ContextForge on Fly.io, including common troubleshooting steps.

Overview

Fly.io is a global app platform for running containers close to your users, with built-in TLS, persistent volumes, and managed Postgres support. It offers a generous free tier and automatic HTTPS with fly.dev subdomains.

1 - Prerequisites

2 - Quick Start (Recommended)

2.1 Initialize Fly project

fly launch --name your-app-name --no-deploy

This creates a new Fly app without deploying immediately.

2.2 Create and attach Fly Postgres

# Create postgres (choose Development configuration for testing)
fly postgres create --name your-app-db --region yyz

# Note the connection details from the output, you'll need the password

2.3 Set secrets

# Set authentication secrets
fly secrets set JWT_SECRET_KEY=$(openssl rand -hex 32)
fly secrets set PLATFORM_ADMIN_EMAIL=admin@example.com
fly secrets set PLATFORM_ADMIN_PASSWORD=your-secure-password
fly secrets set PLATFORM_ADMIN_FULL_NAME="Platform Administrator"

# Set database URL (CRITICAL: use postgresql+psycopg:// for psycopg3)
fly secrets set DATABASE_URL="postgresql+psycopg://postgres:YOUR_PASSWORD@your-app-db.flycast:5432/postgres"

⚠️ Important: Always use postgresql+psycopg:// scheme for psycopg3. Do not use postgresql:// (requires psycopg2) or postgres:// (invalid).

2.4 Deploy the app

fly deploy

3 - Containerfile Requirements

Ensure your Containerfile explicitly installs PostgreSQL dependencies:

# Create virtual environment, upgrade pip and install dependencies
RUN python3 -m venv /app/.venv && \
/app/.venv/bin/python3 -m pip install --upgrade pip setuptools pdm uv && \
/app/.venv/bin/python3 -m pip install 'psycopg[binary]' && \
/app/.venv/bin/python3 -m uv pip install ".[redis]"

The explicit psycopg[binary] (psycopg3) installation is required because uv may not properly install optional dependencies.

4 - fly.toml Configuration

Your fly.toml should look like this:

app = "your-app-name"
primary_region = "yyz"

[build]
dockerfile = "Containerfile"

[env]
HOST = "0.0.0.0"
PORT = "4444"

[http_service]
internal_port = 4444
force_https = true
auto_stop_machines = "stop"
auto_start_machines = true
min_machines_running = 0
processes = ["app"]

[[vm]]
memory = "1gb"
cpu_kind = "shared"
cpus = 1

Note: Don't put secrets like DATABASE_URL in fly.toml - use fly secrets set instead.

5 - Testing Your Deployment

5.1 Check app status

fly status
fly logs

5.2 Test endpoints

# Health check (no auth required)
curl https://your-app-name.fly.dev/health

# Protected endpoints (require auth)
export JWT_SECRET_KEY="same-value-you-set-in-fly-secrets"
export TOKEN=$(python3 -m mcpgateway.utils.create_jwt_token \
  --username admin@example.com \
  --exp 10080 \
  --secret "$JWT_SECRET_KEY" 2>/dev/null | head -1)
curl -H "Authorization: Bearer $TOKEN" https://your-app-name.fly.dev/docs
curl -H "Authorization: Bearer $TOKEN" https://your-app-name.fly.dev/tools

# Optional: enable Basic auth for docs/API
# fly secrets set DOCS_ALLOW_BASIC_AUTH=true API_ALLOW_BASIC_AUTH=true
# curl -u admin:your-password https://your-app-name.fly.dev/docs

5.3 Expected responses

• Health: {"status":"healthy"}
• Protected endpoints without auth: {"detail":"Not authenticated"}
• Protected endpoints with auth: JSON response with data

6 - Troubleshooting

Common Issue 1: SQLAlchemy postgres dialect error

sqlalchemy.exc.NoSuchModuleError: Can't load plugin: sqlalchemy.dialects:postgres

Solutions:

1. Ensure psycopg[binary] is explicitly installed in Containerfile
2. Use postgresql+psycopg:// (not postgresql:// or postgres://) in DATABASE_URL
3. Rebuild with fly deploy --no-cache

Common Issue 2: Database connection refused

Solutions:

1. Verify DATABASE_URL format: postgresql+psycopg://postgres:PASSWORD@your-db.flycast:5432/postgres
2. Check postgres app is running: fly status -a your-app-db
3. Verify password matches postgres creation output

Common Issue 3: Machines not updating

Solutions:

# Force machine updates
fly machine list
fly machine update MACHINE_ID --image your-new-image

# Or restart all machines
fly scale count 0
fly scale count 1

7 - Production Considerations

Security

• Change default BASIC_AUTH_PASSWORD to a strong password
• Consider using JWT tokens for API access
• Enable Fly's private networking for database connections

Scaling

# Scale to multiple machines for HA
fly scale count 2

# Scale machine resources
fly scale memory 2gb

Monitoring

# View real-time logs
fly logs -f

# Check machine metrics
fly machine status MACHINE_ID

8 - Clean Deployment Script

For a completely fresh deployment:

#!/bin/bash
set -e

APP_NAME="your-app-name"
DB_NAME="${APP_NAME}-db"
REGION="yyz"
PASSWORD=$(openssl rand -base64 32)

echo "🚀 Deploying ContextForge to Fly.io..."

# Create app
fly launch --name $APP_NAME --no-deploy --region $REGION

# Create postgres
fly postgres create --name $DB_NAME --region $REGION

# Set secrets
fly secrets set JWT_SECRET_KEY=$(openssl rand -hex 32)
fly secrets set BASIC_AUTH_USER=admin
fly secrets set BASIC_AUTH_PASSWORD=$PASSWORD

# Get postgres password and set DATABASE_URL
echo "⚠️  Set your DATABASE_URL manually with the postgres password:"
echo "fly secrets set DATABASE_URL=\"postgresql+psycopg://postgres:YOUR_PG_PASSWORD@${DB_NAME}.flycast:5432/postgres\""

# Deploy
echo "🏗️  Ready to deploy. Run: fly deploy"

9 - Additional Resources

• Fly.io Documentation
• Fly Postgres Guide
• Fly Secrets Management

Success indicators:

• ✅ fly status shows machines as "started"
• ✅ /health endpoint returns {"status":"healthy"}
• ✅ Protected endpoints require authentication
• ✅ No SQLAlchemy errors in logs