How often should an uptime monitor check? 1 minute or 5
A check is a sample, and between two checks the monitor knows nothing. Here is what an interval costs you in minutes, and how to pick one.
The problem
Every uptime monitor asks the same question when you add a site: how often should it be checked? Every 30 seconds sounds safest. Every 5 minutes sounds like plenty. The honest answer depends on a little arithmetic and on what an outage costs you, and the arithmetic surprises most people.
A check is a sample. Between two checks, the monitor knows nothing. Everything below follows from that.
The arithmetic
Say a site is checked every 5 minutes, and it goes down at a random moment. The next check could be a second away or almost 5 minutes away. On average, it is half an interval away: 2 minutes 30 seconds before anything notices.
Most monitors then want a second failure before they alert, because a single failed check is often a blip. If the second check waits for the next interval, that adds a whole interval more. So the average time to an alert is one and a half intervals, and the worst case is two:
| Interval | Checks a day | First failed check, on average | Alert after a second check, on average | Alert, at worst |
|---|---|---|---|---|
| 30 seconds | 2,880 | 15 seconds | 45 seconds | 1 minute |
| 1 minute | 1,440 | 30 seconds | 1 minute 30 seconds | 2 minutes |
| 5 minutes | 288 | 2 minutes 30 seconds | 7 minutes 30 seconds | 10 minutes |
| 15 minutes | 96 | 7 minutes 30 seconds | 22 minutes 30 seconds | 30 minutes |
Those times are before anyone reads the alert. Add the minutes it takes a person to see it, open a laptop and start looking.
Short outages can be missed entirely
The second surprise. An outage shorter than the interval is only seen if a check happens to land inside it. A 2 minute outage on a 5 minute interval is caught by a check two times in five, and missed three times in five. With a second check required before an alert, it is never confirmed at all, because two checks 5 minutes apart cannot both land inside 2 minutes.
So the interval is also the shortest outage you can count on hearing about. If your site falls over for three minutes every night during a backup, a 5 minute monitor will mostly tell you it is fine. Your uptime figure will be just as wrong, because it is built from the same samples.
What decides it
- What an hour of downtime costs. A shop that takes an order a minute loses money for every minute nobody knows. A blog loses a few readers who come back tomorrow. The more a minute costs, the more each minute of earlier warning is worth.
- How fast someone can act. If the alert goes to an inbox read at nine in the morning, a 30 second interval buys nothing at three in the morning. The interval should match how quickly a person, or an automatic restart, will actually respond.
- What you promise. If a contract or your status page promises 99.9 percent a month, that is about 43 minutes of downtime allowed. A monitor that takes ten minutes to notice uses a quarter of it before anyone starts.
- How many things you watch. Every monitor at a short interval is more checks, more data and, with some tools, a higher bill. Spend the short intervals where they matter and let the rest run slower.
Shorter is not always better
A shorter interval has costs of its own.
Load. One check a minute is 1,440 requests a day, per location. For a home page served from a cache that is nothing. For a page that runs a heavy database query on every request, check a lighter page or a health endpoint instead, so the monitor never becomes part of the problem.
More chances for a blip. A check every 30 seconds sees every brief network hiccup, every slow garbage collection pause and every deploy restart. Without confirmation, each of those is an alert, and alerts that are not real teach people to ignore alerts.
The fix is confirmation, not a longer interval. A short interval paired with a quick re-check and a second location keeps the speed and drops the noise. A failed check tried again a few seconds later, from the same place, catches most blips. Requiring a second location to agree catches the rest, because a broken network route between one checker and your server does not break every route. That is how you get an alert in about a minute without waking someone for a hiccup.
A rule of thumb
| What it is | Interval | Why |
|---|---|---|
| Checkout, sign-in, a payment API | 30 seconds to 1 minute, from two locations | Every minute down is lost money or locked out customers. |
| The main site and the public API | 1 minute | You hear in about two minutes, before most customers write in. |
| Docs, blog, marketing pages | 5 minutes | A short outage costs little, and nobody needs waking for it. |
| Internal tools and staging | 5 minutes | Someone will notice during working hours anyway. |
| Nightly jobs and backups | No interval: a heartbeat | A job that runs once a night cannot be checked from outside. It has to report in. |
For the last row, the guide to a cron job that stopped running shows the heartbeat pattern.
When in doubt, start at 1 minute with confirmation from two locations, and change it only when you have a reason: an outage you heard about too late, or alerts nobody acted on.
How Upglory does it
The fastest interval a monitor can use depends on the plan: 5 minutes on Free, 1 minute on Solo, 30 seconds on Team. Free is 0 EUR, Solo is 15 EUR a month and Team is 49 EUR a month.
A failed check is tried again from the same location 3 seconds later, inside the same check, and only the second result is recorded. A monitor checked from two locations is down when both have failed twice in a row. One location failing alone shows it as degraded and sends no alert. Free is the exception: a free workspace alerts from its one location, after two failures in a row there.
On Team, a 30 second monitor checked from two locations is checked every 30 seconds in all: the two locations take turns, each checking once a minute. When one of them fails, the other is asked to check at once instead of waiting for its turn.
The pricing page has every plan side by side, and the docs cover setting up a monitor.