# Response time alerts

Spectate records the response time of every check. With response time alerts on, a monitor also fires an alert when it answers, but too slowly. There are two thresholds: a hard limit in milliseconds, and a rise in percent compared with the monitor's own recent response times.

## Turn on response time alerts

1. Open the monitor.
2. Select the **Settings** tab.
3. Under **Alerting**, turn on **Slow responses**.
4. Set **Hard threshold**, **Deviation**, **Baseline window** and **Compared against**.
5. Select **Save changes**.

Response time alerts are off for a new monitor. In the API, the switch is `performance_check_enabled`.

## Hard threshold

**Hard threshold** is a response time in milliseconds, 1 to 30,000. The dashboard suggests 5,000.

Spectate compares each successful check with the hard threshold. If the response time is above it, Spectate fires the alert "Response time exceeded threshold for NAME" with the severity `high`. The alert shows the response time and the threshold.

The alert resolves at the first successful check that is at or below the threshold. Failed checks do not count. A failed check is a [down](/monitors/locations-and-checks) problem, not a response time problem.

## Deviation from the baseline

The deviation alert finds a monitor that gets slower than usual, also when it stays below the hard threshold. Every 5 minutes, Spectate compares two median response times:

- The median of the last **Baseline window**, for example the last 15 minutes.
- The median of the **Compared against** period before that window, for example the 2 hours before.

If the recent median is higher than the earlier median by **Deviation** percent or more, Spectate fires the alert "Response time deviated for NAME". The severity is `medium`. The alert resolves when a later comparison finds a rise below **Deviation**.

For example, the earlier median is 200 ms and **Deviation** is 100%. The alert fires when the recent median reaches 400 ms.

| Setting | Values | Default in the dashboard | API field |
|---|---|---|---|
| **Deviation** | 1 to 500 percent | 100 | `performance_deviation_percentage` |
| **Baseline window** | 15 minutes to 2 hours, in steps of 15 minutes | 15 minutes | `performance_deviation_timeframe`, in minutes |
| **Compared against** | 1 to 24 hours, in whole hours | 2 hours | `performance_deviation_before_timeframe`, in minutes |

**Baseline window** must be shorter than **Compared against**. Spectate needs checks in both periods. Each period must hold enough checks for a useful median. For example, with an interval of 5 minutes, a 15-minute window holds 3 checks.

## How the two thresholds work together

- A monitor has one response time alert at a time. While one fires, Spectate does not fire a second.
- If you set only **Hard threshold**, a check at or below it resolves the alert.
- If you leave a threshold empty in the API, Spectate does not use it. `performance_hard_threshold: null` turns off the hard threshold. `performance_deviation_percentage: null` turns off the deviation check.
- Spectate does not compare response times of a paused monitor.

Response time alerts follow the same [escalation paths](/alerting/escalation-paths) as the other alerts of the monitor. Because they have a lower severity than a down alert, a path can [branch on severity](/alerting/escalation-paths) and send them to a quieter channel.

## See the response times

The **Performance** tab of a monitor shows the response times for a period you select. It shows **Average**, **p50**, **p95**, **Apdex** and the number of **Samples**. For HTTP and HTTPS monitors, it also splits each request into phases: DNS lookup, connection, TLS handshake and time to first byte.

All response times come from the monitor's primary region. A monitor in Europe that checks a server in Asia has a longer response time than a monitor in Asia. See [Locations and checks](/monitors/locations-and-checks).