What is uptime monitoring?

Learn what uptime checks observe, when probe failures become alerts, and how to configure HTTP assertions, locations, and reporting windows.

Best practices
Illustration of website and service availability monitoring

A homepage can return 200 OK while the API behind it is failing. It can also load from one region and time out from another. Uptime monitoring runs scheduled checks from known locations and records whether a website or service behaves as expected. Alert rules decide when that evidence should reach a responder.

How uptime checks work

A probe sends a request to a configured URL and checks the result.

  • Status: Match the endpoint contract. A page may need 200 OK, while a successful operation without a response body can return 204 No Content. Decide whether to follow redirects. A connection timeout or failed TLS handshake may occur before there is an HTTP status to record.
  • Timing: Record elapsed time. Where the probe exposes separate phases, DNS lookup, connection setup, TLS, and time to first byte can narrow the investigation.
  • Timeout: Define how long the probe will wait. Choose a value that fits the endpoint and the supported range of the monitoring tool.

One failed check provides evidence about that attempt. The configured rules determine whether to retry, wait for consecutive failures, compare locations, or open an incident. Notification and on-call routing are separate settings.

Check content as well as the status

An application can serve the same 200 status for a working page and a database error page. To tell them apart, check for content in the response body. Use GET when the assertion needs a body. A HEAD response has none.

  • Look for a stable marker associated with the sign-up entry point.
  • Reject error text such as “database connection failed.”
  • Check for an element marker such as <div id="pricing"> when it reliably appears in a healthy response.

Combine status and content checks when the endpoint has a stable success marker. A generic header can also appear on an error page, and localized or redesigned text can cause false alarms. Use a browser probe for JavaScript rendering and interaction.

From failed check to recovery

  1. Record the failure. An abnormal event is generated when the configured detection conditions are met.
  2. Notify responders. Send through configured channels. The on-call and notification rules determine whether an unacknowledged issue escalates, when it escalates, and who receives it.
  3. Acknowledge and investigate. A responder takes ownership according to the team's incident process.
  4. Confirm recovery. Apply the recovery condition and update or close the incident according to the team's workflow. The first failed and recovered samples bound what the probe observed.

Give the responder the first observed failure time, failure type, affected locations, and any useful timing data or log links. A status mismatch, TLS failure, and missing content call for different first checks.

Which endpoints to monitor

Start with the paths people rely on, then choose a check that suits each one.

  • Homepage and major landing pages.
  • Login and registration endpoints.
  • Core APIs such as order creation or payment callbacks. Use a test endpoint or isolated environment for periodic requests that would otherwise create real transactions.
  • Health endpoints such as /healthz. Confirm which dependencies they actually check.
  • Critical CDN assets, including JavaScript and stylesheets.

Calculate availability with a defined denominator

Time-based availability is eligible time minus downtime, divided by eligible time. With all 30 days in scope, 99.9% allows 43.2 minutes of downtime. At 99.99%, the allowance is 4.32 minutes.

Probe success rate counts the share of evaluated samples that passed for the selected locations and time range. Report missing samples and maintenance exclusions separately. A short outage can fall between checks. A service availability SLI can also use the share of successful real requests. These measurements describe different populations, so a passing probe percentage cannot be converted directly into time lost by users.

Verify a success and a failure after setup

With Guance HTTP synthetic tests, set the expected status and response content for the endpoint. Choose All in the default mode when both conditions must pass. Select locations and the task schedule. The shortest preset documented for HTTP tasks is one minute, available on Commercial Plan and above. The Test button checks request configuration. Inspect scheduled results to verify reachability from the selected nodes.

Configure the evaluation interval, lookback, thresholds, and recovery conditions in Synthetic Testing Anomaly Detection, then associate an alert policy. The task and detector have separate schedules. A consecutive trigger rule counts detection evaluations. Run a controlled success, failure, and recovery to confirm the expected events and notifications.

When reporting availability, retain the probe interval, nodes, reporting period, and maintenance policy. For SLA verification, align the evidence with the contract and actual service impact.

Frequently asked questions

Does a faster interval always help?

It can reveal a failed sample sooner. If an outage starts just after a successful check on a fixed one-minute schedule and the rule needs two consecutive scheduled failures, the second failed check starts about two minutes later. Request timeouts, rule evaluation, and notification delivery can add time. Set the schedule against the detection time the service can tolerate, platform options, load, and cost.

Why check from more than one region?

Comparing locations helps separate a path-specific failure from a wider outage. Set any cross-region threshold and time window around where users connect. A region can warrant its own alert when users there are affected and other regions are healthy.

How can I check an internal service?

Place a Guance self-hosted node on a network that can reach the target. Its DataKit synthetic testing collector needs internal tasks enabled and access to the configured task service and data upload destination. Select that node and verify the scheduled results in the workspace.

References. RFC 9110 on HEAD · Google SRE on service level indicators · Guance availability monitoring (Simplified Chinese)

Get a tailored plan

Contact us

Join the community

Scan with WeChat
to join the community

Try Guance

Start online and pay only for what you use.

Get started

Choose a Guance plan

Code hosting