Server Uptime Monitoring: What to Check, How Often and Who Gets the Alert

Your server can be down for an hour before you notice, and your customers will notice first. Uptime monitoring exists so that you hear about a problem within minutes, not from an angry email. This guide explains how it works and how to set it up so that alerts are fast and trustworthy.

What "99.9% uptime" really means

Uptime is the share of time your server is reachable. The percentages sound close, but the downtime they allow is very different. Over a 30-day month:

• 99% allows about 7 hours and 12 minutes of downtime.
• 99.9% allows about 43 minutes.
• 99.99% allows about 4 minutes and 20 seconds.

When a provider quotes an uptime number, ask what is counted. Planned maintenance, network faults and your own misconfiguration are often excluded.

Two ways to watch a server

From the outside. A monitoring service connects to your server (or your website) every minute from another place and reports if it does not answer. This sees what a visitor sees, but it cannot see inside the machine, and a firewall can block it.

From the inside. A tiny program on the server sends a short "I am alive" signal every minute. If the signals stop, the server (or its network) is in trouble. This is cheap, works behind any firewall and uses almost no CPU, but it only tells you the signal stopped, not why.

The best setups use both: an outside check for the website and an inside heartbeat for the machine.

A simple heartbeat you can build yourself

You can build a basic version with a cron job on the server and any service that expects a regular call:

* * * * * curl -fsS -m 10 --retry 2 https://your-monitor.example/ping/your-id > /dev/null

This runs once a minute, waits at most ten seconds and retries twice if the network blinks. The monitor raises an alert when it has not heard from you for a few minutes.

How often should you check?

One minute is the usual choice for production servers. Checking every five minutes means a typical outage can run for five minutes before you are told. Checking more often than once a minute rarely helps and creates more false alarms.

Avoiding false alarms

An alert that cries wolf gets ignored. Good monitoring waits for more than one missed check before it calls a server down. Two or three missed minutes is a sensible rule: a single lost packet does not wake anyone up, but a real outage is reported within a few minutes. It should also send a second message when the server comes back, with how long it was down.

What else to monitor besides "is it up"

• Disk space. A full disk stops databases and mail.
• Memory and load. Steady growth often means a leak or an attack.
• Certificate expiry. An expired SSL certificate looks like an outage to visitors.
• Backups. Check that the last backup finished, not just that the job exists.
• Key services. A server can answer pings while the web server or database is dead.

Who should get the alert?

Send alerts to a person who can act, and to a second person or a shared mailbox in case the first is asleep. An alert that goes to an address nobody reads is the same as no monitoring.

Let someone else watch it

If you do not want to maintain monitoring yourself, our managed services include 24/7 monitoring and alerts, with the response time set by your plan. You can also start small: even a one-minute heartbeat and an email on failure puts you ahead of most servers.