Detect data anomalies with the intelligent detector - Application Real-Time Monitoring Service

Managed Service for Prometheus provides an intelligent detector to identify unusual fluctuations in your KPI time series data. This enables anomaly detection, helping you make informed decisions for alerting, automated mitigation, and root cause analysis. This topic describes how to use the intelligent detector to detect data anomalies.

Limitations

This feature is not supported on v2 instances.

Detect anomalies in a Prometheus instance

Log on to the ARMS console.
In the left navigation pane, choose Managed Service for Prometheus > Instances.
For the target instance, click Grafana Workspace.
In the navigation pane on the left, click the icon (Explore). Then, select the target data source from the drop-down list next to Explore in the upper-left corner.
In the expression field next to Metrics browser, enter the following sample PromQL statement. This statement uses the intelligent detector to detect unusual fluctuations in a metric.
```
anomaly_detect(arms_cms_collector_duration_seconds[180m], 3)
```
Note
- arms_cms_collector_duration_seconds: The target metric. Replace this with your actual metric name.
- The input for the anomaly_detect() function must be a range vector. Therefore, you must append a time selector, such as [180m], after the metric name. In this example, [180m] is the time range and 3 is the sensitivity parameter. If you use an aggregate function like sum() before calling anomaly_detect(), you must change the time selector to [180m:] to ensure the result is a range vector. For example: anomaly_detect(sum(node_memory_free_bytes)[180m:], 3).