Uptime Monitoring Guide: How to Set It Up Without False Alarms

Learn what uptime monitoring is, what to check, how to set intervals and failure confirmation, and how to cut false alarms. A practical guide for small teams.

Uptime monitoring means a service outside your infrastructure requests your website, API or port on a schedule, and alerts you when it stops answering correctly. It tells you your site is down before your customers do. The hard part is not detecting an outage. It is setting up checks so that the alerts you get are real, and so that you can trust the pager at 3 a.m.

This guide covers what to monitor, how to choose intervals and locations, how to read availability percentages, and how to keep false alarms down. It also covers the extras that cause quiet outages: cron jobs, certificates and domains.

Uptime Beacon
Keep False Alarms Off Your Pager
Uptime Beacon checks endpoints, ports and hosts from five regions, opening incidents only after repeated failures are confirmed from another location on paid plans.
Explore uptime monitoring

TL;DR

  • Monitor from the outside, from more than one region, and open an incident only after several failures in a row confirmed from a second location.
  • Check the thing users depend on: a real endpoint that returns the right status, not just a server that answers pings.
  • Turn availability targets into minutes. A 99.9% monthly target leaves about 43 minutes of downtime.
  • Watch cron jobs with heartbeats, and watch certificate and domain expiry, because those failures are silent until they are not.
  • Keep your status page independent of your own stack, and if you handle EU personal data, check where your monitoring results are stored.

Table of Contents

What uptime monitoring is

Uptime is the share of time a service is available and working. Uptime monitoring measures it by running automated checks from outside your network. A monitor sends a request, compares the response with what you expect, records the result and the response time, and triggers an alert when the expectation fails.

The outside view matters. Your own dashboards run inside the same infrastructure that might be broken. If a load balancer, DNS record or upstream provider fails, internal metrics can look healthy while real visitors see an error page. An external probe sees what a visitor sees.

For most teams, website uptime monitoring is the first monitoring they set up, and it stays the most valuable. It answers one question reliably: can a user reach this right now?

How uptime monitoring works

Every uptime check follows the same loop.

  1. Probe. A monitoring node in a given region sends a request: an HTTP(S) request, a TCP connection to a port, or a ping to a host.
  2. Evaluate. The result is judged against rules: status code, response within a timeout, optional content match, or whether the port accepted a connection.
  3. Record. Status and response time are stored, which gives you history for incident reviews and reports.
  4. Decide. The system decides whether this failure is an incident. This step is where good tools and noisy tools differ.
  5. Notify. If it is an incident, alerts go to email, Slack or a webhook. When the check passes again, the incident resolves and the recovery is logged.

Check types

Check type What it confirms Typical use
HTTP(S) The URL returns the expected status within the timeout Websites, web apps, API endpoints
TCP port The port accepts connections Databases, mail servers, custom services
Host The machine is reachable Servers and network devices
Heartbeat A job checked in on schedule Cron jobs, backups, queue workers
Certificate and domain expiry Time left before expiry TLS certificates, domain registrations

What you should monitor

Do not start with a list of servers. Start with what your users and customers rely on, then work backward.

  • The public front door. Your homepage or marketing site. Easy to check, and the one customers notice first.
  • The critical user paths. Login, checkout, the app’s main page. A site can return 200 on the homepage while sign-in is broken.
  • Your API. Monitor a dedicated health endpoint plus one or two real read endpoints. We go deeper in API uptime monitoring: what to check and how to alert and in API health checks: endpoint design and monitoring.
  • Dependencies you own. Database and cache ports, the mail server, anything a single failure would take your service down with.
  • Background work. Scheduled jobs and workers that do not serve requests but must run. See the cron section below.
  • Things that expire. Certificates and domains.

A useful rule: if a failure would make a customer contact you, there should be a monitor for it. If a failure would not matter to anyone, it probably should not page anyone.

Do not stop at “the server answers”

A ping or a bare TCP connection proves the machine is on. It does not prove the application works. An HTTP check that expects a specific status code, and optionally a piece of content, comes much closer to proving the user experience. For APIs, point the check at an endpoint that touches the database, so a dead connection pool shows up as a failure.

Metrics that matter, and how to read availability numbers

Uptime percentage is the headline number, but it is only one signal. Google’s SRE guidance names four golden signals for monitoring a system: latency, traffic, errors and saturation, as described in the SRE book chapter on monitoring distributed systems. External uptime checks give you the first and third directly: how long a response took and whether it errored. Traffic and saturation usually come from your own metrics. Use both views.

Turn percentages into minutes

Percentages hide how little room you have. The same source notes that a 99.9% monthly objective allows roughly 43 minutes of downtime. Here is the arithmetic for a 30-day month:

Availability target Allowed downtime per 30-day month
99% about 7 hours 12 minutes
99.9% about 43 minutes
99.95% about 21.6 minutes
99.99% about 4.3 minutes

Two practical lessons follow.

First, a single 20-minute outage consumes about half of a 99.9% month. Know your budget before you promise a number to customers.

Second, the tighter the target, the less a human can do about it. At 99.99%, you have about four minutes, which is less than the time it takes to wake up, find a laptop and log in. Targets like that depend on automation and redundancy, not on a faster pager response.

Turn percentages into minutes — overview diagram

Measure what the contract says

If you promise uptime to clients, decide in advance how it is measured: which endpoints, from where, and whether planned maintenance counts. Monitoring history that records every check gives you evidence in either direction. Maintenance windows keep planned work from being counted as incidents.

How to set up uptime monitoring

A setup that works for a solo developer works for a team of twenty if you follow the same order.

1. List the endpoints that matter

Write down five to ten. Start with the homepage, the login or main app URL, the API health endpoint and any key dependency port. Resist monitoring everything on day one.

2. Choose the check interval

The interval is the gap between checks. A shorter interval finds outages sooner, but costs more and creates more chances for a one-off failure. As a starting point:

  • Every 5 minutes is fine for marketing sites, internal tools and low-stakes services.
  • Every 1 minute suits customer-facing apps and APIs with uptime commitments.

The interval is only half of detection time. If you require three failures in a row before alerting, a one-minute interval means about three minutes before an incident opens. Plan your budget with that delay included.

3. Use more than one location

A single probe location cannot tell the difference between “your service is down” and “the route between that probe and your service is broken.” Multiple regions solve this and also show partial outages, such as a CDN problem that only affects Asia Pacific. Pick regions near your users plus at least one far away.

4. Require confirmation before alerting

This is the single biggest lever against false alarms. Instead of alerting on the first failed request, open an incident only after several consecutive failures, and confirm from a second location. A packet loss event between one probe and your server then fails the first check and passes the confirmation, and nobody gets paged.

Failure retries and second-location alert confirmation

5. Route alerts to the right place

Send alerts to a channel someone actually watches. Email suits low-urgency monitors. Slack suits team awareness. A webhook lets you connect paging or automation tools. Not every monitor deserves the same route: a staging site and a payment API should not wake the same person.

6. Publish a status page

A status page lets customers see what is happening without opening support tickets. It must not depend on the system it reports on. If it runs on the same stack, or behind the same dashboard, it goes down with everything else. For ideas on what to copy and what to skip, see status page examples.

Best practices that cut false alarms

Alert fatigue is the real cost of bad monitoring. An engineer who gets three meaningless pages learns to ignore the fourth, which is the real one.

  • Confirm from a second region. A failure seen from one place only is a network question, not an incident.
  • Require consecutive failures. One timeout is noise. Several in a row is a pattern.
  • Set sensible timeouts. A timeout shorter than your normal slow response creates failures out of healthy traffic. Look at your typical response time first.
  • Alert on symptoms, not every cause. Page when users are affected. Send lower-level warnings, such as disk filling, to a channel that does not wake anyone.
  • Give each alert an owner. An alert with no clear responder is background noise.
  • Use maintenance windows. Planned deploys and migrations should not open incidents.
  • Review noisy monitors monthly. If a monitor fired and nobody acted, change it or delete it.
  • Keep incident history. Failures, region by region, with recovery times, are what you need for the postmortem and for client reports.

For how this plays out for engineers carrying the pager, read SaaS uptime monitoring that pages only for real failures.

Cron jobs, certificates and domains

Some of the most damaging outages come from things that are not web requests at all.

Cron jobs and workers

A scheduled job that stops running makes no noise. The backup does not run, the invoices do not go out, and nobody knows until they are needed. The fix is heartbeat monitoring: the job pings a unique URL when it finishes, and the monitor alerts if the ping does not arrive within the expected window. Set the grace period slightly longer than the job’s longest normal run. More detail is in our guide to cron job monitoring.

Certificate expiry

An expired TLS certificate makes browsers show a full-page warning, and API clients refuse to connect. It is entirely preventable. Watch every certificate you rely on, and alert weeks ahead, not days, so that renewal problems surface while there is time to fix them. Remember that automatic renewal can fail without telling anyone.

Domain expiry

A lapsed domain takes the website, email and everything else with it. Registrations are often held by a former employee or an agency, and renewal notices go to inboxes nobody reads. Monitor expiry on every domain, and start the countdown early. We explain why in domain expiry monitoring for teams.

Our perspective

We built Uptime Beacon because we spent decades on the receiving end of bad alerts. Between us, the founders have 30+ years running production in banking, fintech, telecom and SaaS, and the pattern was always the same: monitoring that was technically correct and practically useless, because it paged people for network blips.

So our view is simple. A monitor’s job is not to report every failed request. Its job is to tell a human, as rarely as possible, that something real needs attention. That is why an incident opens only after several failures in a row, and on paid plans only after a second location confirms it.

The second thing we believe is that diagnosis is where the time goes. When a monitor fails, the first ten minutes are usually spent asking the same questions: which regions fail, since when, and is it getting worse? An AI assistant can answer those from check history. So we put an MCP server on every plan, Free included. Connect Claude, ChatGPT, Cursor, VS Code, Codex or another MCP client, sign in, and it can read your monitors and incidents and diagnose a failing one region by region. Changes are off by default. If a team turns on “Allow changes”, the assistant can create, pause and resume monitors, acknowledge incidents and start maintenance windows. Nothing can be deleted over MCP. For more, see AI incident response for on-call teams.

Third, a status page that depends on the dashboard fails when you need it most, so ours does not.

— Aaditya Parashar

Data residency and GDPR for EU teams

Monitoring tools store response data, IP-related metadata and incident history, and some teams treat that as personal data or as part of their compliance scope. If you are an EU company, ask where your results are stored and who processes them.

The legal frame is Chapter V of the GDPR. Article 44 says transfers of personal data to a third country may take place only if the Chapter V conditions are met, so the level of protection is not undermined. In practice, check these items with any provider:

  • Where the control plane and database sit, and whether anything leaves the EU or EEA.
  • Whether the data processing agreement sets conditions for transfers outside the EU or EEA, including, unless agreed otherwise, EU-approved Standard Contractual Clauses.
  • Whether you have assessed third-country laws that could interfere with fundamental rights. The EU Cloud Code of Conduct states that cloud customers and providers transferring data outside the EEA remain responsible for this due diligence.

At Uptime Beacon, every result is stored on a control plane and database in the EU, and the service is operated by Wealth Beacon OÜ in Estonia. Your legal team should still review the agreement for your own situation.

Choosing a plan

You can start without a card. The Free plan gives you 5 monitors checked every 5 minutes from EU Central, one alert channel (email or Slack) and one status page, plus the MCP server. That is enough for a side project or a first proof of concept.

When you need more monitors, more regions and confirmation from a second location, paid plans take over:

Plan Price
Free €0 per month
Starter €14.99 per month
Pro €39.99 per month
Business €89.99 per month
Enterprise No published price

Extra capacity is available as add-ons from €5 per month: 10 more monitors on Starter for €5 per month, 50 more on Pro for €10 per month, 100 more on Business for €15 per month, and 10 more certificate or domain expiry watches for €5 per month. Maintenance windows come with Starter and above. See the plans on the Uptime Beacon site, or read how API and cron monitoring works for SaaS teams. If you run a store, the online shop use case is a better starting point.

A first-week checklist

  1. Day 1: add monitors for your homepage, your main app URL and your API health endpoint.
  2. Day 1: connect one alert channel and send yourself a test alert.
  3. Day 2: add heartbeat checks for your backups and your most important cron jobs.
  4. Day 3: add certificate and domain expiry watches for every domain you own.
  5. Day 4: publish a status page and link it from your footer or support page.
  6. Day 5: write down your availability target in minutes, and who owns each alert.
  7. Day 7: review what fired. Fix, tune or remove anything that paged without a real cause.

FAQ

What is the difference between uptime monitoring and server monitoring?

Uptime monitoring checks from the outside whether a service responds correctly, the way a visitor would see it. Server monitoring looks inside a machine at CPU, memory and disk. You want both: uptime checks tell you users are affected, server metrics help explain why.

How often should I check my website?

Every 5 minutes is enough for most sites, and every minute suits customer-facing apps with uptime commitments. Remember that requiring several consecutive failures adds to detection time, so a faster interval shortens the delay before an incident opens.

How do I stop getting false alerts?

Require several failures in a row and confirm from a second region before an incident opens. Also check that your timeout is longer than your normal slow response, and use maintenance windows for planned work. These three changes remove most network-blip pages.

Is 99.9% uptime good enough?

It depends on what you promise and to whom. A 99.9% monthly target allows roughly 43 minutes of downtime, which is fine for many SaaS products and too loose for some. Pick a target you can measure and afford to meet, then monitor against it.

Sources

Put your site on it. Sleep through the blips.

Add your first monitor now. If your site is fine, you will not hear from us.

Monitor it free
Sign up with GoogleOpening Google…
Free plan, no cardSign up with Google or email