リポジトリの原文
見出し、例、コード、表、リンク、参照画像を含む原文を表示しています。
Grafana OnCall & IRM
OnCall docs: https://grafana.com/docs/oncall/latest/ IRM docs: https://grafana.com/docs/grafana-cloud/alerting-and-irm/
Grafana OnCall OSS is in maintenance mode (archived March 2026). Cloud users → IRM. Concepts (chains, schedules, integrations) are identical.
Prerequisites
- Grafana Cloud stack with IRM/OnCall enabled
- API token (
Authorization: <token>) - Slack workspace + admin to install the OnCall app (for ChatOps)
Core concepts
| Concept | Description |
|---|---|
| Integration | Webhook URL accepting alerts; one per source |
| Route | Jinja2 condition that maps to an escalation chain (first True wins) |
| Escalation Chain | Wait / notify schedule / notify team / webhook / auto-resolve steps |
| Schedule | Calendar-based rotation (web / iCal / Terraform) |
| Alert Group | Related alerts collapsed by a Grouping ID template |
| Notification Policy | Per-user channels — Slack, mobile push, SMS, phone, email |
Flow: alert → integration → routing template → escalation chain → notifications → ack / resolve.
Common Workflows
1. Wire Alertmanager → IRM and verify routing
yaml
# 1. In IRM: New integration → Alertmanager. Copy the webhook URL.
# 2. alertmanager.yml (see references/integrations.md for full block):
receivers:
- name: grafana-oncall
webhook_configs:
- url: https://<stack>.grafana.net/integrations/v1/alertmanager/<id>/
send_resolved: true
max_alerts: 100bash
# 3. Test the routing template BEFORE going live (UI: Integration → Route → "Preview")
# Paste a sample payload (use a real Alertmanager test webhook). Expect:
# Routing result: True
# Escalation chain selected: <expected chain>
# If False — your Jinja expression is wrong; fix and re-preview.
# 4. End-to-end test — fire amtool (or any test webhook) at the URL
amtool alert add foo severity=critical team=platform --alertmanager.url http://localhost:9093
# 5. Verify in IRM → Alert Groups (should appear within ~5s) — confirm:
# - Correct route fired
# - Correct schedule/user notified
# - Slack message appeared in the configured channelRouting template syntax + Jinja helpers: `references/templates-schedules.md`. Other integrations (Grafana Alerting, generic webhook, Slack): `references/integrations.md`.
2. Build an escalation chain + verify
1. Notify "Primary On-Call" (Important Notifications)
2. Wait 5 min
3. Notify "Primary On-Call" (Default Notifications)
4. Wait 10 min
5. Notify team "Platform"
6. Trigger outgoing webhook (PagerDuty / ticket)bash
# Verify: create a test alert (UI → Integration → "Send demo alert"),
# then watch the alert-group timeline tick through steps 1 → 6.
# Use IRM → Escalation Chains → "Test" if available, otherwise the demo alert is the canonical check.3. Create a schedule from iCal + verify "who is on-call right now"
bash
# 1. Create the schedule
curl -X POST https://<stack>.grafana.net/api/v1/schedules/ \
-H "Authorization: <token>" -H "Content-Type: application/json" \
-d '{"name":"Platform On-Call",
"ical_url_primary":"https://calendar.example.com/platform.ics",
"slack":{"channel_id":"C123456ABC","user_group_id":"S123456ABC"}}'
# 2. Verify the schedule was created
curl -s https://<stack>.grafana.net/api/v1/schedules/ \
-H "Authorization: <token>" | jq '.results[] | select(.name=="Platform On-Call")'
# 3. Verify who is on-call right now
curl -s https://<stack>.grafana.net/api/v1/schedules/<schedule_id>/next_shifts/ \
-H "Authorization: <token>" | jq '.results[0]'
# Expect a shift starting now (or recently) with the right user_id.Terraform variant + shift block: `references/templates-schedules.md`.
Best practices
- Keep chains ≤4 levels with a definitive final step (webhook to PagerDuty or auto-resolve)
- Always set
send_resolved: truein Alertmanager so OnCall auto-resolves - Use
max_alerts: 100in Alertmanager webhook config - Combine Slack + mobile push for delivery reliability
- Assign integrations/schedules to teams for RBAC
