fly-io（飞 io）

Fly.io Deployment & Management

Core Concepts

• Fly App: Named group of Machines (VMs) + config + networking + secrets belonging to one org
• Fly Machine: Fast-launching VM (sub-second start). Ephemeral root filesystem. Lifecycle: created -> started -> stopped -> destroyed
• Fly Volume: NVMe-backed persistent storage, 1:1 with a Machine, region-pinned, encrypted at rest by default
• Fly Proxy: Edge proxy handling TLS termination, load balancing, autostop/autostart, HTTP routing
• flyctl (fly): CLI for all Fly.io operations. Install: brew install flyctl or curl -L https://fly.io/install.sh | sh
• fly.toml: App configuration file. See references/fly-toml-config.md for complete reference

Quick Start Workflow

# 1. Create app + fly.toml
fly launch

# 2. Set secrets
fly secrets set DATABASE_URL=postgres://... SECRET_KEY=...

# 3. Deploy
fly deploy

# 4. Check status
fly status
fly logs

fly.toml Essentials

Minimal web app configuration:

app = 'my-app'
primary_region = 'ord'

[build]
  dockerfile = "Dockerfile"

[http_service]
  internal_port = 8080
  force_https = true
  auto_stop_machines = "stop"
  auto_start_machines = true
  min_machines_running = 1

  [http_service.concurrency]
    type = "requests"
    soft_limit = 200
    hard_limit = 250

[[vm]]
  memory = '512mb'
  cpu_kind = 'shared'
  cpus = 1

For the complete fly.toml reference with all sections, see references/fly-toml-config.md.

Machines Management

# Scale horizontally
fly scale count 3                          # 3 Machines in primary region
fly scale count 3 --region ord,iad         # Across regions

# Scale vertically
fly scale vm performance-1x --vm-memory 2048

# Direct Machine control
fly machine run <image> --region ord
fly machine clone <machine-id> --region iad
fly machine update <machine-id> --vm-memory 512
fly machine stop <machine-id>
fly machine start <machine-id>
fly machine destroy <machine-id>

Machine sizes: shared-cpu-1x (256MB default), shared-cpu-2x, shared-cpu-4x, shared-cpu-8x, performance-1x through performance-16x. GPU options: a10, l40s, a100-pcie-40gb, a100-sxm4-80gb.

Volumes

fly volumes create mydata --region ord --size 10   # 10GB
fly volumes list
fly volumes extend <vol-id> --size 20              # Extend to 20GB (cannot shrink)
fly volumes snapshots list <vol-id>
fly volumes fork <vol-id>                          # Copy a volume

Configure in fly.toml:

[mounts]
  source = "mydata"
  destination = "/data"
  initial_size = "10gb"
  auto_extend_size_threshold = 80
  auto_extend_size_increment = "1GB"
  auto_extend_size_limit = "50GB"

Critical: Volumes are 1:1 with Machines, region-pinned, no automatic replication. Always provision at least 2 Machines with volumes for redundancy. Your app must handle data replication between volumes.

Secrets

fly secrets set KEY=value OTHER_KEY=other_value
fly secrets set KEY=value --stage               # Stage without restarting
fly secrets deploy                               # Deploy staged secrets
fly secrets list                                 # Names only, no values
fly secrets unset KEY OTHER_KEY

Secrets are encrypted at rest, injected as env vars at boot. Use [[files]] in fly.toml to mount secrets as files (value must be base64-encoded):

[[files]]
  guest_path = "/etc/ssl/private/key.pem"
  secret_name = "TLS_PRIVATE_KEY"

Logs

fly logs                                     # Continuous tail (default, streams indefinitely)
fly logs --no-tail                           # Output recent logs once and exit
fly logs --machine <machine-id>              # Filter to a specific machine
fly logs --machine <machine-id> --no-tail    # Recent logs for one machine, then exit
fly logs --region iad                        # Filter by region
fly logs --json                              # JSON-formatted output

Important: flyctl logs does NOT support -n, --tail <count>, or --lines flags. To limit output, pipe through standard tools:

fly logs --no-tail | head -50                # Last 50 lines
fly logs --no-tail | tail -20                # Most recent 20 lines
fly logs --machine <id> --no-tail | grep -i error   # Search for errors

Available flags:

Networking

For detailed networking patterns (public services, private networking, Flycast, custom domains, DNS), see references/networking.md.

Quick Reference

fly ips list                                 # List allocated IPs
fly ips allocate-v6                          # Dedicated IPv6 (free)
fly ips allocate-v4 --shared                 # Shared IPv4 (free)
fly ips allocate-v4                          # Dedicated IPv4 (paid)
fly certs create mydomain.com                # Add custom domain + TLS cert

Private networking: All apps in an org share a WireGuard mesh (6PN). Use <appname>.internal DNS for inter-app communication. Bind to fly-local-6pn or [::]:port to accept private connections.

Health Checks

Service-level checks affect Fly Proxy routing. Top-level [checks] are for monitoring only.

# In [http_service] or [[services]]
[[http_service.checks]]
  grace_period = "10s"
  interval = "30s"
  timeout = "5s"
  method = "GET"
  path = "/health"

# Top-level (monitoring, no routing impact)
[checks.my_check]
  type = "http"
  port = 8080
  path = "/health"
  interval = "30s"
  timeout = "5s"

Autoscaling

Autostop/autostart (Fly Proxy-based, for web services):

[http_service]
  auto_stop_machines = "stop"      # "off", "stop", or "suspend"
  auto_start_machines = true
  min_machines_running = 1

  [http_service.concurrency]
    type = "requests"
    soft_limit = 200               # Used for excess capacity calculation

Metrics-based autoscaling: Deploy the autoscaler app in your org. It polls Prometheus metrics and creates/destroys Machines. See Fly.io docs for setup.

Deployment Strategies

[deploy]
  strategy = "rolling"              # "rolling" (default), "immediate", "canary", "bluegreen"
  max_unavailable = 0.33            # For rolling: fraction or integer
  release_command = "bin/migrate"   # One-off command before deploy (e.g., DB migrations)

• rolling: One-by-one replacement (default)
• immediate: All at once, skip health check waits
• canary: Boot one Machine, verify health, then rolling (no volumes)
• bluegreen: Boot new set alongside old, migrate traffic after health checks (requires health checks, no volumes)

CI/CD with GitHub Actions

For complete CI/CD setup including deploy tokens, review apps, and multi-app deploys, see references/cicd-github-actions.md.

Quick setup:

# .github/workflows/fly.yml
name: Fly Deploy
on:
  push:
    branches: [main]
jobs:
  deploy:
    runs-on: ubuntu-latest
    concurrency: deploy-group
    steps:
      - uses: actions/checkout@v4
      - uses: superfly/flyctl-actions/setup-flyctl@master
      - run: flyctl deploy --remote-only
        env:
          FLY_API_TOKEN: ${{ secrets.FLY_API_TOKEN }}

Generate deploy token: fly tokens create deploy -x 999999h

Databases & Storage

For detailed database and storage patterns, see references/databases-storage.md.

• Managed Postgres: fly mpg create -- fully managed, automated backups
• Tigris Object Storage: S3-compatible, fly storage create
• Upstash Redis: Managed Redis, fly redis create

Production Checklist

1. Security: Enable SSO, use secrets for sensitive data, release unnecessary public IPs from private apps, use Flycast for internal services
2. Performance: Use performance CPUs for production web apps, right-size memory, enable swap (swap_size_mb) for spike absorption
3. Availability: Run 2+ Machines (ideally across regions), configure autostop/autostart, set up health checks
4. Networking: Set up custom domain + TLS cert, consider dedicated IPv4 if needed
5. Monitoring: Configure Prometheus metrics export, set up Sentry, export logs to external service
6. CI/CD: Deploy via GitHub Actions with deploy tokens, set up review app previews

Process Groups

Run multiple processes in one app (e.g., web + worker):

[processes]
  web = "bin/rails server -b 0.0.0.0 -p 8080"
  worker = "bin/sidekiq"

[http_service]
  internal_port = 8080
  processes = ["web"]

[[vm]]
  processes = ["web"]
  memory = "1gb"

[[vm]]
  processes = ["worker"]
  memory = "512mb"

Useful Commands Cheat Sheet