Running & Maintaining a Site

Uptime Monitoring

Definition

Uptime monitoring is a check run from somewhere other than your own network, usually every one to five minutes, against a specific URL. The checker records the HTTP status code, how long the response took, and often whether an expected string appears in the body. If the check fails a set number of times in a row — two or three, to avoid crying wolf over one dropped packet — the service raises an alert by email, SMS, push notification or a webhook into a chat channel. Better checkers run from several regions, so a failure in one city can be told apart from a failure everywhere. A free tier from any of the common providers covers one or two sites at five-minute intervals, which is enough for most published documents and small sites.

Why It Matters

The gap between an outage starting and someone noticing is usually the largest part of the total downtime, and it is the part you can fix cheaply. A five-minute check turns a discovery time of hours into minutes. Monitoring is also the only evidence you will have if you later want to claim against an uptime guarantee, because a provider will ask when the failure started and ended and will not take your word for it. And it catches the quiet failures that nobody reports: a certificate that lapsed overnight, a custom domain whose DNS record was edited by mistake, a redirect loop that returns 200 to you because your browser cached the old page.

How It Works

Point the check at a real page rather than the domain root alone, and enable keyword matching so that a page which loads but has lost its content still counts as a failure. Set the interval to what the outage would cost — one minute for anything with a deadline, five minutes for everything else — and require two consecutive failures before an alert fires. Add a separate certificate check that warns you fourteen and seven days before expiry, since that is a different failure with a different fix. Route alerts to at least two destinations, because a single email address is a single point of failure, and send a copy to whatever feeds your status page. Finally, check the monitor itself: most services send a monthly report, and a monitor that has quietly been paused is worse than none, because it feels like coverage.

Real-World Example

A studio publishes a client portal of galleries at ridley-studio.99helpers.site behind a password. A free monitor checks the login page every five minutes from three regions and looks for the word Sign in inside the response. When a DNS change at the registrar drops the custom domain one evening, the alert arrives at 19:04 and the record is corrected by 19:20 — sixteen minutes of downtime instead of a client email the following morning. The same monitor's certificate check had already flagged, a fortnight earlier, that the old domain was set to expire.

Common Mistakes

  • Monitoring only from one location, so a regional network problem looks like a total outage and a real total outage looks the same as both
  • Alerting on a single failed check — transient packet loss will page you at three in the morning and you will start ignoring the alerts
  • Checking that the server returns a response without checking the content, which misses a page that loads empty or shows a stale cached version
  • Sending every alert to one inbox with no escalation, so an outage that starts on a Friday evening is found on Monday

Related Terms

Put a file online in seconds

Drop in a document, an image, a page or a whole static website and share the link — free, with no build step and no server to set up.

Host a file free →