Three moving parts: a probe network, a confirmation layer and a notifier. Nothing exotic — but the details are where most monitoring goes wrong.
Twenty-four regions on five providers, deliberately spread so that one provider outage cannot blind the whole fleet. Probes run on bare metal where we can get it.
| Continent | Regions | Providers |
|---|---|---|
| Europe | 9 | Hetzner, OVH, Scaleway |
| North America | 6 | Vultr, DigitalOcean |
| Asia | 5 | Vultr, Linode |
| South America | 2 | Vultr |
| Africa and Oceania | 2 | OVH, Vultr |
When a probe fails, the check is immediately re-run from three other regions chosen at random. An incident opens only if at least two of them agree. This kills roughly 94% of what would otherwise be false pages — mostly transient routing problems near a single probe.
Every individual probe result, kept for 30 days on all plans.
Kept for 13 months, which covers year-over-year comparisons.
Kept indefinitely, including the full timeline and who acknowledged what.
UptimeGlass started in 2020 as an internal tool at a Dutch hosting company and was spun out a year later. Four engineers, profitable since 2022, no outside investment. We run our own status page on a completely separate provider, because monitoring that shares infrastructure with the thing it monitors is a nice way to learn about correlated failure.