Configure alerts

Grafana Cloud Knowledge Graph transforms alerts into structured insights that explain system behavior. These insights help you understand failures, configuration changes, resource saturation, anomalies, and error events across your environment.

This topic introduces how alert mapping works and how to configure alerts so they correctly populate the knowledge graph. Subsequent topics provide detailed guidance for failure alerts and amend alerts.

How alert mapping works

The knowledge graph analyzes alerts and metadata to determine:

Which entity the alert belongs to, such as service, deployment, node, instance, database, and so on
What insight category the alert represents
How the alert affects entity health, RCA workbench timelines, and change history

To enable this, custom alert rules must include specific asserts labels that the knowledge graph uses to ingest and interpret signals.

Insight categories

The knowledge graph classifies alerts into one of five categories:

Category	Meaning	When to Use
Failure	The system is in an incorrect or invalid state.	Misconfiguration, missing replicas, incorrect leader, invalid resource state
Saturation	A resource is reaching or exceeding capacity.	CPU/memory pressure, DB connections near max
Error	A fault or runtime error has occurred.	Job failures, 5xx HTTP errors, crash loops
Anomaly	The system deviates from its normal behavior.	Latency spikes, traffic anomalies
Amend	A configuration or metadata change took place.	Version bumps, scaling events, flag toggles

Categories determine whether an alert contributes to entity health (failure, saturation, error) or contextual timelines (amends).

Required labels for alert ingestion

All alerts mapped into the knowledge graph must include the following labels:

Label	Purpose
`asserts_alert_category`	Classifies the alert into an insight category (failure, error, amend, saturation, anomaly)
`asserts_entity_type`	Identifies the type of entity receiving the insight
`asserts_severity`	Indicates the impact level (info, warning, critical)

Without these labels, the alert can’t be ingested into the knowledge graph.

Recommended labels

Label	Purpose
`asserts_env`	Enables accurate entity resolution across environments
`asserts_site`	Identifies region or cluster alignment

Choose the correct type of custom alert

Alert category determines how the knowledge graph interprets a signal:

If the system is configured incorrectly, use failure alerts
If something in the environment changes configuration or metadata, use amend alerts
If the alert measures resource pressure, use saturation alerts
If the alert detects runtime faults, use error alerts
If the alert identifies abnormal behavior, use anomaly alerts

Failure and amend alerts are the most commonly authored types.

For more information about amend alerts, refer to Amend
For more information about failure alerts, refer to Failure
If you want to integrate existing Grafana Cloud alerts into the knowledge graph, refer to Existing alerts