The big outages of late 2025 had a pattern. When AWS's us-east-1 went down in October, and Cloudflare followed in November, plenty of teams learned that their monitoring lived on the same infrastructure as the things it monitored. Status pages lagged. Alerting SaaS tools were unreachable. For an hour or two, the honest answer to "what exactly is broken?" was a shrug.
The fix is old-fashioned and cheap: keep one monitor that depends on nothing but a box in your own building. It won't tell you why a hyperscaler is on fire, but it will tell you, from the outside, what is reachable and what isn't, while everyone else refreshes a stale status page.
What such a watchdog should check, roughly in this order:
- Your internet uplink, against more than one anchor, so a single provider blip doesn't page you.
- DNS for the domains you depend on.
- Your public services, with real HTTP assertions (status code, a keyword in the body), not just "port 443 answers".
- TLS certificates and domain expiry, the two outages everyone causes themselves.
- The local infrastructure the cloud can't see anyway: containers, the hypervisor, the NAS that quietly lost a disk, temperatures.
Any self-hosted monitor covers some of this list. Uptime Kuma does the uptime part well and is free; Zabbix does all of it if you bring the time to build it. DeviceShelf Server is our take: one Docker container that discovers the network by itself, runs the checks above out of the box, and alerts through whatever channel you already use. Since 1.6.0 it also watches Docker, Proxmox, NAS health and hardware temperatures. There's an honest side-by-side with Kuma, PRTG and Zabbix if you want to pick for yourself.
Whichever tool you choose: put it on hardware you own, give it its own notification path, and let it be the one thing in your stack that has never heard of us-east-1.