Larm·May 19, 2026·Updated November 8, 2026monitoringfly-ioplatform-guides

How to monitor a Fly.io app

Fly.io health checks decide where traffic goes inside Fly. They don't tell you what your users see. What to monitor from outside, the fly.toml checks to set up, and the Fly-specific failures to watch for.

Fly.io runs your app on Machines in the regions you pick, and the Fly Proxy sends each request to a Machine that can take it. Fly has its own health checks, and they matter, but they answer a different question than uptime monitoring does. Fly's checks decide whether a Machine should get traffic. Monitoring from outside tells you whether your users can actually use the app.

Fly's health checks

Fly has two kinds of checks in fly.toml, and they behave differently.

Service checks live under [[http_service.checks]]. The Fly Proxy uses them to decide whether a Machine is ready to serve traffic. A Machine that fails its service check doesn't get requests until it passes again.

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

Top-level checks live under [checks]. They don't affect routing. Fly's docs suggest them for apps without public services, or when you want a check that doesn't change where requests go.

Both kinds run inside Fly's network. If every Machine passes its checks but your custom domain's certificate has expired, or the app returns errors on routes the health check doesn't touch, Fly won't tell you. That's what an external monitor is for.

What to monitor

Your public hostname. The .fly.dev domain or your custom domain, over HTTPS. A check against this URL goes through the Fly Proxy, TLS and Machine selection, the same path as a real request.

A health endpoint that checks dependencies. Use the same /health path as your service check, and have it verify the database connection. Apps on Fly usually reach their database over the private network, and that connection can fail while the app itself is fine. If you're on Elixir, I wrote about a minimal health check plug for Plug and Phoenix.

More than one location. If you run Machines in several regions, a request from Europe and a request from Asia can land on different Machines. Checking from several locations shows you when one region is slow or failing while the others are fine.

Your custom domain's certificate. Fly issues custom domain certificates through Let's Encrypt and validates your domain through DNS or a TLS challenge. Wrong or conflicting DNS records can stop validation, and Cloudflare proxying can block the HTTP challenge. Larm checks certificate expiry on HTTPS monitors and alerts at 28, 14, 7 and 1 days before it runs out.

Fly-specific things to know

Machines that stop when idle. With auto_stop_machines set to "stop" or "suspend", Fly stops Machines that aren't getting traffic and starts them again on the next request. That first request waits for the Machine to boot. Set min_machines_running to keep at least one Machine running in your primary region, or give the monitor a timeout long enough for a cold start.

[http_service]
  auto_stop_machines = "stop"
  auto_start_machines = true
  min_machines_running = 1

Volumes belong to one host. A Fly volume is a slice of an NVMe drive on the same physical server as the Machine that mounts it. If that drive fails, the Machine goes down, and Fly doesn't replicate volumes for you. Fly's docs recommend at least two Machines, ideally in different regions, and replication handled by your app or database.

Rolling deploys. During a rolling deploy, some Machines run the new version while others still run the old one. If the two versions disagree, for example about a database schema, requests can behave differently depending on which Machine they hit.

Private networking. Apps in a Fly organization talk to each other over a WireGuard mesh using IPv6, called a 6PN, and find each other through .internal names. That network is separate from the public internet, so the app can be reachable from outside while it can't reach its database. A health endpoint that queries the database catches this.

Setting it up in Larm

Create an HTTP monitor in Larm for your app's public URL or its /health endpoint.

Timeout. Larm's default timeout is 10 seconds. That's enough for Machines that stay running. If Machines stop when idle and you don't keep one running, raise it to cover the boot time.

Expected status and keywords. If your health endpoint returns JSON like {"status":"ok","db":"ok"}, add "db":"ok" as an expected keyword. The check then fails when the endpoint returns 200 but reports a problem.

One monitor per app. If you run an API, a frontend and a worker as separate Fly apps, give each HTTP service its own monitor. For workers that don't serve HTTP, use a heartbeat monitor: the worker pings a URL after each run, and you get an alert when the pings stop.

Larm's free plan includes 15 monitors, checked from all probe locations.

Start monitoring in minutes.

Free plan. 15 monitors. Multi-probe voting. No credit card.

Sign Up Free