Uptime monitoring

On this page 8

Uptime monitoring is the practice of repeatedly probing a server, website or service from outside at fixed intervals to confirm that it responds correctly, and of recording each result to measure availability over time. Within the monitoring & logs discipline for servers, it answers the most basic operational question: is the service reachable and working right now, and how often has it failed in the past?

Unlike log analysis, which examines what a system reports about itself, uptime monitoring observes the service from the user’s point of view. A failed probe triggers alerting, and the stored history feeds availability reports and service level agreements.

How uptime monitoring probes a service

An uptime monitor works by sending a synthetic request to a target on a schedule and comparing the response against expected conditions. The checking agent, often called a probe or poller, records the outcome, the latency and any error details for every attempt.

A single probe cycle usually passes through several stages: DNS resolution, a TCP connection, an optional TLS handshake, the request itself and validation of the reply. Failure at any stage marks the check as down, and the stage at which it failed is a valuable diagnostic clue for the engineer who responds.

The following table lists the common check types and what each one actually verifies.

Check typeWhat it verifiesTypical use
ICMP pingHost answers echo requestsBasic network reachability
TCP port checkA port accepts connectionsDatabases, mail, SSH
HTTP(S) checkStatus code and response timeWebsites and APIs
Keyword checkExpected text appears in the bodyDetecting broken pages that return 200
DNS checkRecords resolve to expected valuesDomain and failover setups
Heartbeat checkA job reports in before a deadlineCron jobs and backups
Common uptime check types

The heartbeat check inverts the model: instead of the monitor calling the service, the service or scheduled job calls a unique URL after each successful run, and silence beyond a grace period raises the alarm. This approach suits workloads that have no listening port at all.

Key parameters that shape detection

The usefulness of uptime monitoring depends on a handful of configuration parameters that balance detection speed against false alarms. Setting them deliberately matters more than the choice of tool.

ParameterTypical rangeEffect of a lower valueEffect of a higher value
Check interval30 s to 5 minFaster detection, more loadShort outages may be missed
Timeout5 to 30 sSlow responses count as failuresHangs are detected late
Retry count1 to 3 retriesMore false alarmsSlower confirmation of outages
Probe locations1 to 10+ regionsRegional faults look globalBetter isolation of network issues
Confirmation quorum2 of 3 locationsSingle probe can trigger alertsFewer alerts from probe-side faults
Configuration parameters and their effects

Multi-location confirmation is the most effective defense against noise. If a probe in one region fails while others succeed, the problem is likely a transit or probe-side fault, not an outage of the service itself, and many teams alert only when a quorum of locations agrees.

Availability percentages and allowed downtime

Uptime monitoring produces the availability percentage, calculated as successful checks or available minutes divided by the total measured period. This figure is the basis of most service level objectives and contractual service level agreements.

Each additional “nine” shrinks the permitted downtime by a factor of ten, as the next table shows for a 30-day month and a 365-day year.

AvailabilityDowntime per monthDowntime per year
99%About 7.2 hoursAbout 3.65 days
99.9%About 43 minutesAbout 8.8 hours
99.95%About 22 minutesAbout 4.4 hours
99.99%About 4.3 minutesAbout 53 minutes
99.999%About 26 secondsAbout 5.3 minutes
Availability targets and permitted downtime

These numbers expose a measurement limit: with a one-minute interval, a monitor cannot credibly verify five nines, because a single missed or misread check already represents a large share of the yearly budget. Measurement resolution must match the target being reported.

Uptime monitoring is often confused with neighboring practices, but each looks at a different layer of evidence. The comparison below clarifies where it fits among the tools of the same topic.

PracticeViewpointMain question answered
Uptime monitoringExternal, synthetic requestsIs the service reachable and correct
Health checkInternal endpoint or orchestratorIs this instance fit to receive traffic
Infrastructure monitoringAgents on hostsAre CPU, memory and disks healthy
Log analysisRecords written by the systemWhat happened and why
Real user monitoringActual visitor browsersWhat do real users experience
Uptime monitoring compared with related practices

Platforms such as Nagios and Zabbix combine external checks with agent-based infrastructure metrics, while dedicated uptime services focus on probing from many public locations. In practice, uptime monitoring tells a team that something broke, and logs from an aggregation stack explain why.

Designing checks that reflect real availability

A well-designed check tests the path that users depend on, not merely whether a machine is powered on. A server can answer ping perfectly while the web application returns errors, so application-level HTTP checks with content validation are the better default for public services.

  • Point checks at a lightweight status endpoint that touches critical dependencies such as the database, rather than at a heavy page.
  • Validate status code, response time and a keyword, because error pages frequently return a 200 status.
  • Monitor TLS certificate expiry and domain expiry, which cause complete outages that no server metric predicts.
  • Route alerts through escalation policies with on-call rotation so that a failure reaches a person who can act.
  • Publish results on a status page to reduce support load during incidents.

Limitations and common mistakes

The main limitation is that uptime monitoring samples availability rather than observing it continuously. Problems that last less than one interval, or that affect only some users, can remain invisible.

MistakeConsequenceFix
Monitoring only with pingApplication failures go unnoticedAdd HTTP checks with keyword validation
Single probe locationNetwork blips cause false alertsRequire confirmation from several regions
Monitor inside the same networkOutage takes down the monitor tooRun probes from independent infrastructure
No retry before alertingAlert fatigue and ignored pagesUse one or two retries with short delay
Ignoring latency trendsSlow degradation precedes outagesAlert on response time thresholds
Frequent mistakes and how to fix them

A further warning concerns scheduled maintenance: without maintenance windows, planned work distorts availability statistics and trains responders to ignore alerts, which is the most dangerous failure mode of any monitoring system.

In plain words

Uptime monitoring is like asking a friend to knock on a shop door every minute and write down whether someone answers. If nobody answers several times in a row, the friend calls the owner right away.

Over a month, the notebook shows how often the shop was open when it promised to be. The key idea is simple: check often, from outside, and keep a record, so problems are noticed before customers complain and the owner can prove how reliable the service really is.

A real-world example

A small hosting company runs an online booking platform for about 300 dental clinics and promises 99.9% monthly availability. Its operations team configures HTTPS checks every 60 seconds from four regions against a status endpoint that queries the database, with a 10-second timeout, one retry and alerts only when two of four locations fail.

One evening, the checks begin failing with a keyword mismatch while the servers still answer ping. The on-call engineer receives a page within three minutes, finds that the database connection pool is exhausted, and restarts the affected service. Total downtime is nine minutes, well inside the 43-minute monthly budget, and the stored check history gives the clinics an accurate, verifiable availability report at the end of the month.