What each monitoring stack catches

A service that fails quietly is found by its users. These stacks find it first: some by asking your services whether they answer, one by noticing that a job has stopped reporting, one by watching the hosts themselves. Every one of them sends its alerts somewhere — and where that somewhere runs decides whether the alert arrives on the day it matters.

None of this is required to run an application. Each is a stack like any other, behind Traefik, and each keeps its interface on the VPN by default.

Is it up?

Uptime Kuma and Gatus both probe your services on a schedule — HTTP, TCP, DNS, certificate expiry — and alert when an answer is wrong. They differ in where the configuration lives. Uptime Kuma is configured in its web interface and keeps everything in its database. Gatus is configured in one YAML file, which can sit in version control beside the rest of your setup. One of the two is enough.

Did it run?

Healthchecks works the other way round. Each job pings a URL when it runs; Healthchecks alerts when an expected ping does not arrive. An alarm wired like that detects its own failure: a backup that stopped running, a timer disabled weeks ago, a host that went down, or a monitor that died — all of which look like silence to the probes above. borgmatic reports to it for that reason.

What is the host doing?

Beszel shows CPU, memory, disk and load per host and per container, with threshold alerts. A disk-usage threshold warns before the outage rather than during it. Further hosts run the standalone agent, which connects out to the hub and opens no port.

Where the alerts go

ntfy receives alerts over HTTP and pushes them to phones, desktops and scripts. Every stack above can publish to it.

A receiver on the monitored host stops with it. When the host goes down, its monitoring goes down — and a receiver, or an SMTP server, next to it stops too. The outage that most needs an alert then produces none. ntfy is built to run on a second host on its own; the ntfy project also runs a public instance, which gives an off-host path without a second machine.

A channel that has never fired proves nothing. For each one you rely on, trigger it once on purpose — stop a monitored container, let a check expire, cross a threshold — and confirm the alert arrives on the device that is supposed to get it.

The rest of the category

The repository also ships ciao, Checkmate, Grafana with Prometheus, Zabbix, Scrutiny for disk health, changedetection.io and Gotify. They are in the catalogue, each with what it is and its pinned version.

Where to go from here

• Traefik not running yet: what every server needs
• Backups that report when they stop: backup and restore
• Something already broken: when something is broken