kibana-alerting-rules（kibana 警报规则）

Kibana Alerting Rules

Core Concepts

A rule has three parts: conditions (what to detect), schedule (how often to check), and actions (what happens when conditions are met). When conditions are met, the rule creates alerts, which trigger actions via connectors.

Authentication

All alerting API calls require either API key auth or Basic auth. Every mutating request must include the kbn-xsrf header.

kbn-xsrf: true

Required Privileges

• all privileges for the appropriate Kibana feature (e.g., Stack Rules, Observability, Security)
• read privileges for Actions and Connectors (to attach actions to rules)

API Reference

Base path: <kibana_url>/api/alerting (or /s/<space_id>/api/alerting for non-default spaces).

Creating a Rule

Required Fields

Optional Fields

Example: Create an Elasticsearch Query Rule

curl -X POST "https://my-kibana:5601/api/alerting/rule/my-rule-id" \
  -H "kbn-xsrf: true" \
  -H "Content-Type: application/json" \
  -H "Authorization: ApiKey <your-api-key>" \
  -d '{
    "name": "High error rate",
    "rule_type_id": ".es-query",
    "consumer": "stackAlerts",
    "schedule": { "interval": "5m" },
    "params": {
      "index": ["logs-*"],
      "timeField": "@timestamp",
      "esQuery": "{\"query\":{\"match\":{\"log.level\":\"error\"}}}",
      "threshold": [100],
      "thresholdComparator": ">",
      "timeWindowSize": 5,
      "timeWindowUnit": "m",
      "size": 100
    },
    "actions": [
      {
        "id": "my-slack-connector-id",
        "group": "query matched",
        "params": {
          "message": "Alert: {{rule.name}} - {{context.hits}} hits detected"
        },
        "frequency": {
          "summary": false,
          "notify_when": "onActionGroupChange"
        }
      }
    ],
    "tags": ["production", "errors"]
  }'

The same structure applies to other rule types — set the appropriate rule_type_id (e.g., .index-threshold, .es-query) and provide the matching params object. Use GET /api/alerting/rule_types to discover params schemas.

Updating a Rule

PUT /api/alerting/rule/{id} — send the complete rule body. rule_type_id and consumer are immutable after creation. Returns 409 Conflict if another user updated the rule concurrently; re-fetch and retry.

Finding Rules

curl -X GET "https://my-kibana:5601/api/alerting/rules/_find?per_page=20&page=1&search=cpu&sort_field=name&sort_order=asc" \
  -H "Authorization: ApiKey <your-api-key>"

Query parameters: per_page, page, search, default_search_operator, search_fields, sort_field, sort_order, has_reference, fields, filter, filter_consumers.

Use the filter parameter with KQL syntax for advanced queries:

filter=alert.attributes.tags:"production"

Lifecycle Operations

# Enable
curl -X POST ".../api/alerting/rule/{id}/_enable" -H "kbn-xsrf: true"

# Disable
curl -X POST ".../api/alerting/rule/{id}/_disable" -H "kbn-xsrf: true"

# Mute all alerts
curl -X POST ".../api/alerting/rule/{id}/_mute_all" -H "kbn-xsrf: true"

# Mute specific alert
curl -X POST ".../api/alerting/rule/{rule_id}/alert/{alert_id}/_mute" -H "kbn-xsrf: true"

# Delete
curl -X DELETE ".../api/alerting/rule/{id}" -H "kbn-xsrf: true"

Terraform Provider

Use the elasticstack provider resource elasticstack_kibana_alerting_rule.

terraform {
  required_providers {
    elasticstack = {
      source  = "elastic/elasticstack"
    }
  }
}

provider "elasticstack" {
  kibana {
    endpoints = ["https://my-kibana:5601"]
    api_key   = var.kibana_api_key
  }
}

resource "elasticstack_kibana_alerting_rule" "cpu_alert" {
  name         = "CPU usage critical"
  consumer     = "stackAlerts"
  rule_type_id = ".index-threshold"
  interval     = "1m"
  enabled      = true

  params = jsonencode({
    index              = ["metrics-*"]
    timeField          = "@timestamp"
    aggType            = "avg"
    aggField           = "system.cpu.total.pct"
    groupBy            = "top"
    termField          = "host.name"
    termSize           = 10
    threshold          = [0.9]
    thresholdComparator = ">"
    timeWindowSize     = 5
    timeWindowUnit     = "m"
  })

  tags = ["infrastructure", "production"]
}

Key Terraform notes:

• params must be passed as a JSON-encoded string via jsonencode()
• Use elasticstack_kibana_action_connector data source or resource to reference connector IDs in actions
• Import existing rules: terraform import elasticstack_kibana_alerting_rule.my_rule <space_id>/<rule_id> (use default for the default space)

Triggering Kibana Workflows from Rules

Preview feature — available from Elastic Stack 9.3 and Elastic Cloud Serverless. APIs may change.

Attach a workflow as a rule action using the workflow ID as the connector ID. Set params: {} — alert context flows automatically through the event object inside the workflow.

curl -X PUT "https://my-kibana:5601/api/alerting/rule/my-rule-id" \
  -H "kbn-xsrf: true" \
  -H "Content-Type: application/json" \
  -H "Authorization: ApiKey <your-api-key>" \
  -d '{
    "name": "High error rate",
    "schedule": { "interval": "5m" },
    "params": { ... },
    "actions": [
      {
        "id": "<workflow-id>",
        "group": "query matched",
        "params": {},
        "frequency": { "summary": false, "notify_when": "onActionGroupChange" }
      }
    ]
  }'

In the UI: Stack Management > Rules > Actions > Workflows. Only enabled: true workflows appear in the picker.

For workflow YAML structure, {{event}} context fields, step types, and patterns, refer to the kibana-connectors skill if available.

Connectors and Actions in Rules

Each action references a connector by ID, an action group, action params (using Mustache templates), and a per-action frequency object. Key fields:

• group — which trigger state fires this action (e.g., "query matched", "Recovered"). Discover valid groups via GET /api/alerting/rule_types.
• frequency.summary — true for a digest of all alerts; false for per-alert.
• frequency.notify_when — onActionGroupChange | onActiveAlert | onThrottleInterval.
• frequency.throttle — minimum repeat interval (e.g., "10m"); only applies with onThrottleInterval.

For full reference on action structure, Mustache variables ({{rule.name}}, {{context.*}}, {{alerts.new.count}}), Mustache lambdas (EvalMath, FormatDate, ParseHjson), recovery actions, and multi-channel patterns, refer to the kibana-connectors skill if available.

Best Practices

1. Set action frequency per action, not per rule. The notify_when field at the rule level is deprecated in favor of per-action frequency objects. If you set it at the rule level and later edit the rule in the Kibana UI, it is automatically converted to action-level values.
2. Use alert summaries to reduce notification noise. Instead of sending one notification per alert, configure actions to send periodic summaries at a custom interval. Use "summary": true and set a throttle interval. This is especially valuable for rules that monitor many hosts or documents.
3. Choose the right action frequency for each channel. Use onActionGroupChange for paging/ticketing systems (fire once, resolve once). Use onActiveAlert for audit logging to an Index connector. Use onThrottleInterval with a throttle like "30m" for dashboards or lower-priority notifications.
4. Always add a recovery action. Rules without a recovery action leave incidents open in PagerDuty, Jira, and ServiceNow indefinitely. Use the connector's native close/resolve event action (e.g., eventAction: "resolve" for PagerDuty) in the Recovered action group.
5. Set a reasonable check interval. The minimum recommended interval is 1m. Very short intervals across many rules clog Task Manager throughput and increase schedule drift. The server setting xpack.alerting.rules.minimumScheduleInterval.value enforces this.
6. Use alert_delay to suppress transient spikes. Setting {"active": 3} means the alert only fires after 3 consecutive runs match the condition, filtering out brief anomalies.
7. Enable flapping detection. Alerts that rapidly switch between active and recovered are marked as "flapping" and notifications are suppressed. This is on by default but can be tuned per-rule with the flapping object.
8. Use server.publicBaseUrl for deep links. Set server.publicBaseUrl in kibana.yml so that {{rule.url}} and {{kibanaBaseUrl}} variables resolve to valid URLs in notifications.
9. Tag rules consistently. Use tags like production, staging, team-platform for filtering and organization in the Find API and UI.
10. Use Kibana Spaces to isolate rules by team or environment. Prefix API paths with /s/<space_id>/ for non-default spaces. Connectors are also space-scoped, so create matching connectors in each space.

Common Pitfalls

1. Missing kbn-xsrf header. All POST, PUT, DELETE requests require kbn-xsrf: true or any truthy value. Omitting it returns a 400 error.
2. Wrong consumer value. Using an invalid consumer (e.g., observability instead of infrastructure) causes a 400 error. Check the rule type's supported consumers via GET /api/alerting/rule_types.
3. Immutable fields on update. You cannot change rule_type_id or consumer with PUT. You must delete and recreate the rule.
4. Rule-level notify_when and throttle are deprecated. Setting these at the rule level still works but conflicts with action-level frequency settings. Always use frequency inside each action object.
5. Rule ID conflicts. POST to /api/alerting/rule/{id} with an existing ID returns 409. Either omit the ID to auto-generate, or check existence first.
6. API key ownership. Rules run using the API key of the user who created or last updated them. If that user's permissions change or the user is deleted, the rule may fail silently. Use _update_api_key to re-associate.
7. Too many actions per rule. Rules generating thousands of alerts with multiple actions can clog Task Manager. The server setting xpack.alerting.rules.run.actions.max (default varies) limits actions per run. Design rules to use alert summaries or limit term sizes.
8. Long-running rules. Rules that run expensive queries are cancelled after xpack.alerting.rules.run.timeout (default 5m). When cancelled, all alerts and actions from that run are discarded. Optimize queries or increase the timeout for specific rule types.
9. Concurrent update conflicts. PUT returns 409 if the rule was modified by another user since you last read it. Always GET the latest version before updating.
10. Import/export loses secrets. Rules exported via Saved Objects are disabled on import. Connectors lose their secrets and must be re-configured.

Examples

Create a threshold alert: "Alert me when CPU exceeds 90% on any host for 5 minutes." Use rule_type_id: ".index-threshold", aggField: "system.cpu.total.pct", threshold: [0.9], and timeWindowSize: 5. Attach a PagerDuty action on "threshold met" and a matching Recovered action to auto-close incidents.

Find rules by tag: "Show all production alerting rules." GET /api/alerting/rules/_find with filter=alert.attributes.tags:"production" and sort_field=name to page through results.

Pause a rule temporarily: "Disable rule abc123 until next Monday." POST /api/alerting/rule/abc123/_disable. Re-enable with _enable when ready; the rule retains all configuration while disabled.

Guidelines

• Include kbn-xsrf: true on every POST, PUT, and DELETE; omitting it returns 400.
• Set frequency inside each action object — rule-level notify_when and throttle are deprecated.
• rule_type_id and consumer are immutable after creation; delete and recreate the rule to change them.
• Prefix paths with /s/<space_id>/api/alerting/ for non-default Kibana Spaces.
• Always pair an active action with a Recovered action to auto-close PagerDuty, Jira, and ServiceNow incidents.
• Run GET /api/alerting/rule_types first to discover valid consumer values and action group names.
• Use alert_delay to suppress transient spikes; use the flapping object to reduce noise from unstable conditions.

Additional Resources

• Kibana Alerting API Reference
• Alerting Concepts
• Create and Manage Rules (UI)
• Rule Action Variables
• Alerting Production Considerations
• Terraform: elasticstack_kibana_alerting_rule
• Terraform: Managing Kibana Rule and Connector Resources