Flyio

Fly.io Infrastructure Skills

Version: 1.0.0 | Target Size: <25KB | Purpose: Fast reference for Fly.io deployments and global application distribution

Overview

What is Fly.io: Modern platform-as-a-service (PaaS) for deploying applications globally with minimal configuration. Fly.io transforms containers into micro-VMs that run on physical hardware across 30+ regions worldwide.

When to Use Fly.io:

• Simple to moderate applications requiring global distribution
• Fast deployments without complex Kubernetes orchestration
• Minimal operations overhead with PaaS simplicity
• Edge computing and low-latency requirements (anycast routing)
• Startup/SaaS applications with unpredictable traffic patterns
• Databases requiring multi-region active replication (Fly Postgres)

When to Use Kubernetes Instead:

• Complex microservices architectures with 10+ interdependent services
• Existing Kubernetes expertise and tooling investment
• Hybrid cloud or multi-cloud requirements (cloud-agnostic)
• Advanced orchestration needs (service mesh, custom operators, advanced scheduling)
• Enterprise compliance requirements (HIPAA, PCI-DSS on-premises)
• Extensive third-party ecosystem integrations (Helm charts, operators)

Detection Criteria:

• Auto-loads when fly.toml detected in project root
• Manual: --tools=flyio flag
• Use Case: Global application deployment with PaaS simplicity

Progressive Disclosure:

• This file (SKILL.md): Quick reference for immediate use
• REFERENCE.md: Comprehensive guide with advanced patterns and deployment strategies

Table of Contents

1. fly.toml Quick Reference
2. Deployment Patterns
3. Secrets Management
4. Networking Basics
5. Health Checks
6. Scaling Patterns
7. Common Commands Cheat Sheet
8. Quick Troubleshooting
9. Framework-Specific Examples

fly.toml Quick Reference

Minimal Configuration (Node.js Express)

# app name (must be globally unique on Fly.io)
app = "my-express-app"

[build]
  # Dockerfile-based build (default)
  dockerfile = "Dockerfile"

[env]
  # Non-sensitive environment variables
  NODE_ENV = "production"
  PORT = "8080"

[[services]]
  internal_port = 8080  # Container port
  protocol = "tcp"

  [[services.ports]]
    handlers = ["http"]
    port = 80  # External HTTP

  [[services.ports]]
    handlers = ["tls", "http"]
    port = 443  # External HTTPS

  # Health check
  [[services.http_checks]]
    interval = "10s"
    timeout = "2s"
    grace_period = "5s"
    method = "GET"
    path = "/health"

Key Concepts:

• app: Globally unique application name (DNS-safe)
• build: How to build your application (Dockerfile, buildpacks)
• env: Non-sensitive configuration (use secrets for sensitive data)
• services: Network services exposed to the internet
• internal_port: Port your app listens on inside container
• ports: External ports (80 HTTP, 443 HTTPS)
• http_checks: Health check configuration

Node.js (Express, Fastify, Koa)

app = "nodejs-api"

[build]
  dockerfile = "Dockerfile"

[env]
  NODE_ENV = "production"
  PORT = "8080"

[[services]]
  internal_port = 8080
  protocol = "tcp"
  auto_stop_machines = true
  auto_start_machines = true
  min_machines_running = 1

  [[services.ports]]
    handlers = ["http"]
    port = 80

  [[services.ports]]
    handlers = ["tls", "http"]
    port = 443

  [[services.http_checks]]
    interval = "10s"
    timeout = "2s"
    grace_period = "5s"
    method = "GET"
    path = "/health"
    protocol = "http"

Dockerfile (minimal):

FROM node:18-alpine
WORKDIR /app
COPY package*.json ./
RUN npm ci --production
COPY . .
EXPOSE 8080
CMD ["node", "server.js"]

Next.js Application

app = "nextjs-app"

[build]
  dockerfile = "Dockerfile"

[env]
  NODE_ENV = "production"
  PORT = "3000"

[[services]]
  internal_port = 3000
  protocol = "tcp"
  auto_stop_machines = true
  auto_start_machines = true

  [[services.ports]]
    handlers = ["http"]
    port = 80

  [[services.ports]]
    handlers = ["tls", "http"]
    port = 443

  [[services.http_checks]]
    interval = "15s"
    timeout = "5s"
    grace_period = "10s"
    method = "GET"
    path = "/"

Dockerfile (Next.js standalone):

FROM node:18-alpine AS builder
WORKDIR /app
COPY package*.json ./
RUN npm ci
COPY . .
RUN npm run build

FROM node:18-alpine
WORKDIR /app
COPY --from=builder /app/next.config.js ./
COPY --from=builder /app/public ./public
COPY --from=builder /app/.next/standalone ./
COPY --from=builder /app/.next/static ./.next/static
EXPOSE 3000
CMD ["node", "server.js"]

Python (Django, FastAPI, Flask)

app = "python-api"

[build]
  dockerfile = "Dockerfile"

[env]
  PYTHONUNBUFFERED = "1"
  PORT = "8000"

[[services]]
  internal_port = 8000
  protocol = "tcp"

  [[services.ports]]
    handlers = ["http"]
    port = 80

  [[services.ports]]
    handlers = ["tls", "http"]
    port = 443

  [[services.http_checks]]
    interval = "10s"
    timeout = "2s"
    grace_period = "5s"
    method = "GET"
    path = "/health"

Dockerfile (FastAPI example):

FROM python:3.11-slim
WORKDIR /app
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt
COPY . .
EXPOSE 8000
CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]

Go Microservice

app = "go-service"

[build]
  dockerfile = "Dockerfile"

[env]
  PORT = "8080"

[[services]]
  internal_port = 8080
  protocol = "tcp"
  auto_stop_machines = true
  auto_start_machines = true

  [[services.ports]]
    handlers = ["http"]
    port = 80

  [[services.ports]]
    handlers = ["tls", "http"]
    port = 443

  [[services.tcp_checks]]
    interval = "10s"
    timeout = "2s"
    grace_period = "5s"

Dockerfile (multi-stage Go build):

FROM golang:1.21-alpine AS builder
WORKDIR /app
COPY go.* ./
RUN go mod download
COPY . .
RUN CGO_ENABLED=0 go build -o /app/server

FROM alpine:latest
RUN apk --no-cache add ca-certificates
COPY --from=builder /app/server /app/server
EXPOSE 8080
CMD ["/app/server"]

Ruby on Rails

app = "rails-app"

[build]
  dockerfile = "Dockerfile"

[env]
  RAILS_ENV = "production"
  PORT = "3000"

[[services]]
  internal_port = 3000
  protocol = "tcp"

  [[services.ports]]
    handlers = ["http"]
    port = 80

  [[services.ports]]
    handlers = ["tls", "http"]
    port = 443

  [[services.http_checks]]
    interval = "10s"
    timeout = "5s"
    grace_period = "10s"
    method = "GET"
    path = "/health"

Elixir Phoenix

app = "phoenix-app"

[build]
  dockerfile = "Dockerfile"

[env]
  PORT = "4000"
  MIX_ENV = "prod"

[deploy]
  release_command = "/app/bin/migrate"

[[services]]
  internal_port = 4000
  protocol = "tcp"

  [[services.ports]]
    handlers = ["http"]
    port = 80

  [[services.ports]]
    handlers = ["tls", "http"]
    port = 443

  [[services.http_checks]]
    interval = "10s"
    timeout = "2s"
    grace_period = "5s"
    method = "GET"
    path = "/health"

Key Feature: release_command runs database migrations before deployment.

Deployment Patterns

Zero-Downtime Deployments

Fly.io's rolling deployment ensures zero downtime:

[deploy]
  strategy = "rolling"  # Default
  max_unavailable = 0.33  # 33% of machines can be down during deploy

How it works:

1. New machines start with updated code
2. Health checks pass on new machines
3. Old machines drain connections (graceful shutdown)
4. Old machines stop after connections close

Graceful Shutdown:

// Express.js example
const server = app.listen(PORT);

process.on('SIGTERM', () => {
  console.log('SIGTERM received, closing server gracefully');
  server.close(() => {
    console.log('Server closed');
    process.exit(0);
  });
});

Health Check Requirements:

[[services.http_checks]]
  interval = "10s"
  timeout = "2s"
  grace_period = "5s"  # Wait 5s before first check
  method = "GET"
  path = "/health"

Blue-Green Deployment

# Deploy to new machines without affecting current ones
fly deploy --strategy=immediate --no-cache

# Traffic still routes to old machines
# Validate new deployment
fly status

# Switch traffic to new machines (instant cutover)
fly deploy --strategy=bluegreen

When to use:

• Critical production deployments requiring validation
• Database migrations requiring downtime window
• High-risk changes requiring instant rollback capability

Canary Releases

Gradual traffic shifting:

# Deploy new version to 10% of machines
fly deploy --strategy=canary:10

# Monitor metrics, errors, logs
fly logs

# Increase to 50%
fly deploy --strategy=canary:50

# Full rollout
fly deploy --strategy=rolling

When to use:

• New features with uncertain performance impact
• Gradual rollout to detect issues early
• A/B testing scenarios

Rollback Procedures

# View release history
fly releases

# Example output:
# VERSION  STATUS   DESCRIPTION
# v3       current  Deploy by user@example.com
# v2       complete Deploy by user@example.com
# v1       complete Deploy by user@example.com

# Rollback to previous version (v2)
fly deploy --image registry.fly.io/my-app:v2

# Immediate rollback (no health checks)
fly deploy --strategy=immediate --image registry.fly.io/my-app:v2

Emergency Rollback:

# Rollback and bypass all health checks
fly deploy --strategy=immediate --no-cache --image registry.fly.io/my-app:v2

Secrets Management

Setting Secrets

# Set single secret
fly secrets set DATABASE_URL="postgres://user:pass@host/db"

# Set multiple secrets
fly secrets set \
  API_KEY="abc123" \
  JWT_SECRET="xyz789" \
  STRIPE_KEY="sk_live_..."

# Set secret from file
fly secrets set SSL_CERT="$(cat cert.pem)"

# Import from .env file
fly secrets import < .env

Important: Secrets are encrypted at rest and never logged.

Multi-Environment Segregation

Development:

fly secrets set -a my-app-dev \
  DATABASE_URL="postgres://dev-host/myapp_dev" \
  STRIPE_KEY="sk_test_..."

Staging:

fly secrets set -a my-app-staging \
  DATABASE_URL="postgres://staging-host/myapp_staging" \
  STRIPE_KEY="sk_test_..."

Production:

fly secrets set -a my-app-prod \
  DATABASE_URL="postgres://prod-host/myapp_prod" \
  STRIPE_KEY="sk_live_..."

Best Practice: Use separate apps for each environment (-dev, -staging, -prod).

Viewing and Removing Secrets

# List secret names (values hidden)
fly secrets list

# Unset secret
fly secrets unset API_KEY

# Unset multiple secrets
fly secrets unset API_KEY JWT_SECRET

Security Note: Secret values cannot be retrieved after setting. Store backups securely.

Secrets Rotation Strategy

# 1. Set new secret with different name
fly secrets set NEW_DATABASE_URL="postgres://new-host/db"

# 2. Update application code to use NEW_DATABASE_URL
fly deploy

# 3. Verify application works with new secret
fly logs

# 4. Remove old secret
fly secrets unset DATABASE_URL

# 5. Rename secret (optional)
fly secrets set DATABASE_URL="$(fly secrets list | grep NEW_DATABASE_URL)"
fly secrets unset NEW_DATABASE_URL

Networking Basics

Internal Service Communication

Fly.io provides internal DNS for private service communication:

<app-name>.internal

Example (microservices):

# api-service fly.toml
app = "api-service"

[[services]]
  internal_port = 8080
  protocol = "tcp"

# worker-service fly.toml
app = "worker-service"

[env]
  API_URL = "http://api-service.internal:8080"

Key Benefits:

• No external traffic (faster, more secure)
• Automatic service discovery
• Load balancing across all machines

External Access Configuration

HTTP/HTTPS:

[[services]]
  internal_port = 8080
  protocol = "tcp"

  [[services.ports]]
    handlers = ["http"]
    port = 80

  [[services.ports]]
    handlers = ["tls", "http"]
    port = 443

TCP/UDP:

[[services]]
  internal_port = 5432
  protocol = "tcp"

  [[services.ports]]
    port = 5432

WebSocket:

[[services]]
  internal_port = 8080
  protocol = "tcp"

  [[services.ports]]
    handlers = ["http"]
    port = 80
    # WebSocket support included automatically

Private Networking

Secure communication between apps:

# Allocate private IP
fly ips private

# Example: Connect database to API
# Database app
fly ips private -a my-postgres

# API app connects via internal DNS
# my-postgres.internal:5432

Use Cases:

• Database connections (Postgres, Redis)
• Backend-to-backend communication
• Admin tools accessing production systems

Fly Proxy and Anycast Routing

How it works:

1. User request hits nearest Fly edge location (anycast)
2. Fly Proxy routes to nearest healthy machine
3. Machine processes request
4. Response returns to user

Benefits:

• Ultra-low latency (edge routing)
• Automatic SSL/TLS termination
• Global load balancing
• DDoS protection

Regions:

# List available regions
fly platform regions

# Add region
fly regions add lhr  # London Heathrow

# Remove region
fly regions remove lhr

# Backup regions (fallback)
fly regions set iad ord sjc --backup lhr

Health Checks

HTTP Health Checks

[[services.http_checks]]
  interval = "10s"        # Check every 10 seconds
  timeout = "2s"          # Timeout after 2 seconds
  grace_period = "5s"     # Wait 5s before first check (app startup)
  method = "GET"
  path = "/health"
  protocol = "http"
  tls_skip_verify = false
  headers = {}

Health Check Endpoint (Express.js):

app.get('/health', (req, res) => {
  // Check database connection
  db.ping()
    .then(() => res.status(200).json({ status: 'healthy' }))
    .catch(() => res.status(503).json({ status: 'unhealthy' }));
});

TCP Health Checks

For non-HTTP services (databases, caches):

[[services.tcp_checks]]
  interval = "10s"
  timeout = "2s"
  grace_period = "5s"

When to use:

• Redis, Postgres, MongoDB
• gRPC services
• Custom TCP protocols

Custom Script-Based Health Checks

Use HTTP endpoint that runs comprehensive checks:

app.get('/health', async (req, res) => {
  const checks = {
    database: false,
    redis: false,
    external_api: false
  };

  try {
    // Database check
    await db.query('SELECT 1');
    checks.database = true;

    // Redis check
    await redis.ping();
    checks.redis = true;

    // External API check (optional)
    await axios.get('https://api.example.com/ping', { timeout: 1000 });
    checks.external_api = true;

    // All checks passed
    res.status(200).json({ status: 'healthy', checks });
  } catch (error) {
    res.status(503).json({ status: 'unhealthy', checks, error: error.message });
  }
});

Zero-Downtime Deployment Health Checks

Requirements:

1. grace_period: Allow app startup time
2. Fast response: <2s timeout (avoid slow queries)
3. Critical dependencies only: Don't fail on non-critical services

[[services.http_checks]]
  interval = "10s"
  timeout = "2s"
  grace_period = "10s"  # Allow 10s for app startup
  method = "GET"
  path = "/health"

Best Practice:

// Fast health check (critical dependencies only)
app.get('/health', async (req, res) => {
  try {
    await db.query('SELECT 1');  # Database only
    res.status(200).json({ status: 'ok' });
  } catch (error) {
    res.status(503).json({ status: 'error' });
  }
});

// Comprehensive readiness check (separate endpoint)
app.get('/ready', async (req, res) => {
  // Check all dependencies
  // Caches, external APIs, etc.
});

Scaling Patterns

Horizontal Scaling

# Scale to 3 machines
fly scale count 3

# Scale per region
fly scale count 2 --region iad  # US East
fly scale count 2 --region lhr  # London

# View current scale
fly status

When to scale:

• Increased traffic (CPU >70%, memory >80%)
• Global distribution requirements
• High availability (multiple machines per region)

Auto-Scaling Configuration

[[services]]
  internal_port = 8080
  protocol = "tcp"
  auto_stop_machines = true   # Stop idle machines
  auto_start_machines = true  # Start on traffic
  min_machines_running = 1    # Always keep 1 running

[http_service.concurrency]
  type = "requests"
  soft_limit = 200   # Start new machine at 200 concurrent requests
  hard_limit = 250   # Reject requests at 250 concurrent requests

How it works:

1. Traffic increases → new machines start automatically
2. Traffic decreases → machines stop after idle timeout
3. Cost savings: Pay only for running machines

Best Practice:

• Set min_machines_running = 1 for production (avoid cold starts)
• Set min_machines_running = 0 for dev/staging (cost optimization)

Regional Distribution

# Add regions for global distribution
fly regions add iad ord sjc  # US East, Central, West
fly regions add lhr ams       # London, Amsterdam
fly regions add nrt syd       # Tokyo, Sydney

# View current regions
fly regions list

# Remove region
fly regions remove nrt

Strategy:

• Start with 1-2 regions near users
• Expand based on latency metrics
• Use backup regions for disaster recovery

Resource Limits (CPU, Memory)

# View available machine types
fly platform vm-sizes

# Scale machine size
fly scale vm shared-cpu-1x  # 1 shared CPU, 256MB RAM
fly scale vm shared-cpu-2x  # 2 shared CPUs, 512MB RAM
fly scale vm dedicated-cpu-1x  # 1 dedicated CPU, 2GB RAM

# Custom memory
fly scale memory 512  # 512MB RAM

Machine Types:

• shared-cpu-1x: Small apps, low traffic (256MB)
• shared-cpu-2x: Medium apps (512MB)
• dedicated-cpu-1x: Production apps, consistent performance (2GB)
• dedicated-cpu-2x: High-traffic apps (4GB)

Common Commands Cheat Sheet

Deployment

fly deploy                    # Deploy application
fly deploy --strategy rolling # Rolling deployment (zero downtime)
fly deploy --strategy immediate  # Immediate deployment (downtime)
fly deploy --remote-only      # Build on Fly.io (not locally)
fly deploy --no-cache         # Rebuild without cache

Scaling

fly scale count 3             # Scale to 3 machines
fly scale vm shared-cpu-2x    # Change machine type
fly scale memory 512          # Set memory to 512MB
fly regions add iad           # Add region
fly regions remove iad        # Remove region
fly autoscale standard min=1 max=10  # Enable autoscaling

Secrets

fly secrets set KEY=value     # Set secret
fly secrets list              # List secret names
fly secrets unset KEY         # Remove secret
fly secrets import < .env     # Import from file

Monitoring

fly logs                      # View logs
fly logs -a my-app            # Logs for specific app
fly logs --region iad         # Logs for specific region
fly status                    # Application status
fly vm status                 # Machine status
fly checks list               # Health check status

Debugging

fly ssh console               # SSH into machine
fly ssh console -C "ps aux"   # Run command via SSH
fly ssh sftp shell            # SFTP access
fly proxy 5432                # Port forward to local machine
fly doctor                    # Diagnose issues

App Management

fly apps list                 # List all apps
fly apps create my-app        # Create new app
fly apps destroy my-app       # Delete app
fly open                      # Open app in browser
fly info                      # Application info

Quick Troubleshooting

Top 10 Common Issues

1. Deployment fails → Check health checks: fly logs, adjust grace_period
2. App not starting → Review logs: fly logs, check Dockerfile CMD/ENTRYPOINT
3. Network timeout → Verify internal_port matches app port, check firewall rules
4. Database connection fails → Check secrets: fly secrets list, verify DATABASE_URL
5. High memory usage → Scale machine: fly scale vm shared-cpu-2x, optimize app
6. SSL certificate issues → Verify domain: fly certs show, add certificate: fly certs add example.com
7. Regional latency → Add regions: fly regions add lhr ams, monitor with fly logs
8. Health check failures → Adjust grace period: grace_period = "10s", simplify health endpoint
9. Build failures → Review Dockerfile: fly deploy --local-only, check dependencies
10. Cost concerns → Right-size machines: fly scale vm shared-cpu-1x, enable auto-stop: auto_stop_machines = true

Deployment Troubleshooting

Failed health checks:

# View health check status
fly checks list

# Increase grace period
# fly.toml
[[services.http_checks]]
  grace_period = "15s"  # Increase from 5s

# Check logs during deployment
fly logs --region iad

Build failures:

# Build locally to debug
fly deploy --local-only

# Clear cache and rebuild
fly deploy --no-cache

# View build logs
fly builds

Runtime Troubleshooting

High memory usage:

# Check current usage
fly vm status

# Scale up memory
fly scale memory 1024

# Or upgrade machine type
fly scale vm dedicated-cpu-1x

Connection issues:

# Test internal networking
fly ssh console -C "curl http://other-app.internal:8080/health"

# Check external connectivity
fly ssh console -C "curl https://api.example.com"

# Port forward for local debugging
fly proxy 8080

Logs not appearing:

# Ensure app logs to stdout/stderr (not files)
# Check log level
fly logs --region iad

# SSH to machine and check logs
fly ssh console
tail -f /var/log/app.log

Framework-Specific Examples

NestJS Backend

app = "nestjs-api"

[build]
  dockerfile = "Dockerfile"

[env]
  NODE_ENV = "production"
  PORT = "3000"

[[services]]
  internal_port = 3000
  protocol = "tcp"

  [[services.ports]]
    handlers = ["http"]
    port = 80

  [[services.ports]]
    handlers = ["tls", "http"]
    port = 443

  [[services.http_checks]]
    interval = "10s"
    timeout = "2s"
    grace_period = "10s"
    method = "GET"
    path = "/health"

NestJS Health Check:

// health.controller.ts
@Controller('health')
export class HealthController {
  @Get()
  check() {
    return { status: 'ok', timestamp: Date.now() };
  }
}

Phoenix LiveView

app = "phoenix-liveview"

[build]
  dockerfile = "Dockerfile"

[env]
  PORT = "4000"
  MIX_ENV = "prod"
  SECRET_KEY_BASE = "use-fly-secrets-set-for-this"

[deploy]
  release_command = "/app/bin/migrate"

[[services]]
  internal_port = 4000
  protocol = "tcp"

  [[services.ports]]
    handlers = ["http"]
    port = 80

  [[services.ports]]
    handlers = ["tls", "http"]
    port = 443

Phoenix Health Check:

# router.ex
get "/health", HealthController, :check

# health_controller.ex
defmodule MyApp.HealthController do
  use MyAppWeb, :controller

  def check(conn, _params) do
    json(conn, %{status: "ok"})
  end
end

Rails Backend

app = "rails-api"

[build]
  dockerfile = "Dockerfile"

[env]
  RAILS_ENV = "production"
  PORT = "3000"

[deploy]
  release_command = "bin/rails db:migrate"

[[services]]
  internal_port = 3000
  protocol = "tcp"

  [[services.ports]]
    handlers = ["http"]
    port = 80

  [[services.ports]]
    handlers = ["tls", "http"]
    port = 443

Next Steps

For Advanced Patterns:

• See REFERENCE.md for comprehensive guide with deployment strategies, Fly Postgres, Redis, monitoring
• Covers: Multi-region databases, CI/CD integration, metrics, logging, cost optimization

Common Use Cases:

• Multi-region apps → REFERENCE.md § Global Distribution
• Database setup → REFERENCE.md § Fly Postgres
• CI/CD pipelines → REFERENCE.md § GitHub Actions Integration
• Production monitoring → REFERENCE.md § Observability

Progressive Disclosure: Start here for quick reference, load REFERENCE.md for comprehensive patterns and production examples.

Performance Target: <100ms skill loading (this file ~22KB)

Last Updated: 2025-10-25 | Version: 1.0.0