Uptime Monitoring That Tells You Something Useful
Nearly every managed site has uptime monitoring. Far fewer have monitoring that would have caught last quarter's actual incident, because the outages that hurt a business are rarely "the whole server is down."
What basic monitoring misses
A homepage check returning 200 tells you the web server answered. It doesn't tell you:
- The checkout is throwing an error on step two.
- The contact form submits but the mail never leaves.
- The database is up but responding in nine seconds.
- Search returns nothing for every query.
- The SSL certificate expires on Thursday.
- The site is up but is now serving someone else's content because a DNS record changed.
- It's fine from Virginia and unreachable from Europe.
Every one of those is a real incident we've seen, and every one of them passes a homepage ping.
What to monitor instead
The pages that make money. Homepage, top service pages, checkout, booking. Not just one URL.
A real transaction, end to end. A synthetic check that fills the form and confirms a success state catches the failures customers actually experience. If you monitor one thing beyond the homepage, monitor this.
Content, not just status. Assert that the response contains an expected string. A 200 that returns a "database connection error" page still counts as up to a naive check.
Response time, with a threshold. A site taking eight seconds is functionally down for a good share of visitors. Alert on slow, not only on dead.
Certificate and domain expiry. Both cause total outages, both are entirely calendar-driven, both are preventable with a thirty-day warning.
From more than one place. Regional network problems and geographic blocks look like uptime to a single-location checker.
Making alerts worth reading
The failure mode of good monitoring is too many alerts. An operator who has learned to ignore notifications is worse off than one who has none, because everybody believes it's covered.
- Confirm before alerting. Two consecutive failures from two locations, not one blip.
- Route by severity. Site down goes to a phone. Certificate expiring in 30 days goes to a ticket.
- Alert a person, not a mailbox. With a rota, so it's always somebody's turn.
- Review the noise monthly. Any alert that fires regularly and is regularly ignored gets fixed or deleted. There is no third option.
The number that matters
Not uptime percentage — everyone reports 99.9% and the definition varies. What matters is time to detection: from the moment it broke to the moment a human knew.
If your customers are still telling you before your monitoring does, no dashboard number is worth anything.