How often should an uptime monitor check? Choosing a check interval

Every 10 seconds sounds better than every 5 minutes, but the right interval depends on what you promise, what a check costs the target, and how you handle a single failure. A practical way to choose.

Check interval is the first knob everyone turns and the one most often set by gut feeling. The trade is simple: a shorter interval detects an outage sooner and measures availability more precisely, at the cost of more load on the target and more chances for a single blip to page someone. Here is how to reason about it.

Start from the promise

If you commit to 99.9% monthly availability, your error budget is about 43 minutes. A 5-minute interval cannot see an outage shorter than 5 minutes, and on average it notices one 2.5 minutes late; over a month of small incidents that is a meaningful slice of the budget spent before anyone knows. The interval should be small relative to the shortest outage that matters to the promise — as a rule of thumb, a tenth of the time you are willing to be down before reacting.

Then add the threshold

Detection time is not the interval; it is *interval × consecutive failures required*. A 30-second interval with a failure threshold of 2 pages after about a minute and never pages on a single dropped packet. A 10-second interval with a threshold of 1 pages faster and pages on every hiccup. Most teams are happier with the first: choose the interval for measurement precision and the threshold for noise tolerance.

What a check costs the target

• A HEAD or a small GET to a health endpoint every 10 seconds from three regions is 18 requests a minute — nothing for any real service.
• A multi-step journey that signs in, searches and creates an order is different: it consumes rate limits, writes rows and shows up in analytics. Run those every few minutes, from fewer regions, against a test account.
• A full SSL/TLS security assessment probes protocols and ciphers; once an hour is plenty. Certificates change daily at most.
• DNS and domain checks are cheap and change rarely; a few minutes is fine, and the value is in change detection rather than speed.

A reasonable default set