Reading a monitor's page

The cards

Uptime 24h / 7d / 30d — the share of checks that passed in that window. Checks during a maintenance window are excluded, so planned work does not dent the figure.

A new monitor shows 100% because nothing has failed yet, not because it has been perfect for a month. Check the number of checks beside it.

Avg response — the mean of the successful checks in the window shown. Failures are excluded; a timeout would otherwise drag the average somewhere meaningless.

Checks — how many ran in the window. Useful mostly as a sanity check: far fewer than you expect means the monitor is paused, or was created recently.

The response time chart

One line, over the window you pick with the 1h / 6h / 24h / 7d buttons.

What to look for is shape rather than value. A flat line that steps up and stays up is a change — a deploy, a new region, a database that lost its index. A sawtooth is usually a cache filling and expiring. Spikes at the same time each day are someone else's batch job.

Check results

One bar per check, in order, so a run of failures is visible at a glance next to the timing above.

This is the fastest way to tell a genuine outage from flapping: a solid red block is an outage, scattered red is a flaky target, and a red block that starts exactly on the hour is probably something scheduled.

Configuration

The right-hand card repeats the monitor's settings, plus when it was last checked and when it was created. "Last checked" is worth a look if something seems stuck — if it is older than the interval, the monitor is paused.

Host check monitors

A monitor created by a server check shows something different: the measured value over time, with current, peak, average and trend, and a projection to 100% for a disk that is filling. Uptime and response time are not shown, because the agent pushes the reading — there is no request to time, and a disk at 94% is "up" right until it is not. See server monitoring.