DeadmanOpen source, from Veloccy
You should not be paying for an uptime checker.
One costs almost nothing to run. Deadman is a dead-man switch and uptime checker that runs on your own Cloudflare account, inside the free plan. There is no hosted version and nothing to buy.
A tram’s dead-man’s handle. When the driver lets go, the tram stops.
Why it is free.
An uptime checker is one small request a minute. With one heartbeat source, Deadman makes about 4,320 requests a day: 1,440 checks, 1,440 heartbeats and 1,440 reads of its own status. Cloudflare’s free plan allows 100,000.
4,320 a day, used100,000 a day, free
Paying a monthly fee for that is absurd. It is just as absurd that there was nothing you could take and run on your own account with one button. So we wrote it.
Here is what it does. Stop the cluster and see.
This is Deadman’s own judging, running in your browser on a fast clock, with the settings it ships with. Stop the cluster, park the domain behind the health route, or break a channel.
cluster heartbeat
platform api.veloccy.com/health
push
00:00 UTCA minute passes each second.
What Deadman sent
Nothing. While everything answers, Deadman says nothing.
One alert, then one recovery.
Each source pings Deadman as a heartbeat. When one has been silent for too long you get one alert, and one more on the first ping back. There is nothing in between. A source that has never pinged still alerts.
A page that answers 200 is not always up.
Every minute Deadman fetches your public URLs from outside. A health route must answer with exactly the body you expect, and a redirect counts as a failure, so a login page or a parked domain is down. Three failures in a row make one alert.
Every channel is tried until it takes the alert.
Deadman records what each channel has been told. One that fails is tried again every minute, and the other is not sent the alert twice. If the outage ends before a channel heard of it, that channel is told how long it lasted.
Nothing inside a cluster can report the cluster’s own death.
An alert that a dead cluster has to deliver is no alert. So Deadman lives outside, on Cloudflare, and listens from there.
Your cluster
Deadman
POST /ping/clusterThe cluster sends a heartbeat every minute.
GET /statusAnd every minute it checks that Deadman is still running.
Deadman cannot report its own death either.
So the two watch each other. Your Prometheus reads Deadman’s /status every minute. It answers 200 only while a run has finished in full in the last three minutes: every URL fetched, every check judged, every alert delivered. When it goes stale, your cluster is the one that alerts you.
If your cluster and Deadman are both down at once, nothing alerts. Nothing watches the pair.
HEARTBEATS: bindings.json({ cluster: { silentAfterSeconds: 300 } }),
UPTIME: bindings.json({
failuresBeforeAlert: 3,
timeoutSeconds: 10,
targets: [
{ name: "web app", url: "https://app.veloccy.com/" },
{ name: "platform", url: "https://api.veloccy.com/health", expectBody: "ok" },
],
}), curl -X POST https://deadman.example.com/ping/cluster \
-H "Authorization: Bearer $PING_SECRET_CLUSTER" curl -i https://deadman.example.com/status After the button, the setup is a few lines of config and a curl.
You list what should ping and what should answer. This is the config it ships with; change the names and URLs to yours.
- There are no defaults.
- A missing or malformed setting refuses every ping and fails every run, and the logs name each problem.
- A ping and a check never race.
- All state is in one Durable Object, which handles one call at a time, so two runs cannot both send the same alert.
- Ping secrets are compared in constant time.
- A wrong secret or an unknown source is refused and does not count as a ping.
It is open source. Fork it and make it yours.
Deadman is MIT licensed. Nothing in it is clever. It is a small Worker that someone had to write once, so that nobody has to pay for one again.
Deploy to CloudflareDeadman notices. If you want the alert to wake someone, Vigil is our open-source pager for iPhone, and it keeps paging until a person answers.