write-incident-runbook（编写事件操作手册）

Write Incident Runbook

Create actionable runbooks that guide responders through incident diagnosis and resolution.

When to Use

• Documenting response procedures for recurring alerts or incidents
• Standardizing incident response across on-call rotation members
• Reducing mean time to resolution (MTTR) with clear diagnostic steps
• Creating training materials for new team members on incident handling
• Establishing escalation paths and communication protocols
• Migrating tribal knowledge to written documentation
• Linking alerts to resolution procedures (alert annotations)

Inputs

• Required: Incident or alert name/description
• Required: Historical incident data and resolution patterns
• Optional: Diagnostic queries (Prometheus, logs, traces)
• Optional: Escalation contacts and communication channels
• Optional: Previous incident post-mortems

Procedure

Step 1: Choose Runbook Template Structure

See Extended Examples for complete template files.

Select an appropriate template based on incident type and complexity.

Basic runbook template structure:

# [Alert/Incident Name] Runbook
## Overview | Severity | Symptoms
## Diagnostic Steps | Resolution Steps
## Escalation | Communication | Prevention | Related

Advanced SRE runbook template (excerpt):

# [Service Name] - [Incident Type] Runbook

## Metadata
- Service, Owner, Severity, On-Call, Last Updated

## Diagnostic Phase
### Quick Health Check (< 5 min): Dashboard, error rate, deployments
### Detailed Investigation (5-20 min): Metrics, logs, traces, failure patterns
# ... (see EXAMPLES.md for complete template)

Key template components:

• Metadata: Service ownership, severity, on-call rotation
• Diagnostic Phase: Quick checks → detailed investigation → failure patterns
• Resolution Phase: Immediate mitigation → root cause fix → verification
• Escalation: Criteria and contact paths
• Communication: Internal/external templates
• Prevention: Short/long-term actions

Expected: Template selected matches incident complexity, sections appropriate for service type.

On failure:

• Start with basic template, iterate based on incident patterns
• Review industry examples (Google SRE books, vendor runbooks)
• Adapt template based on team feedback after first use

Step 2: Document Diagnostic Procedures

See Extended Examples for complete diagnostic queries and decision trees.

Create step-by-step investigation procedures with specific queries.

Six-step diagnostic checklist:

1. Verify Service Health: Health endpoint checks and uptime metrics curl -I https://api.example.com/health # Expected: HTTP 200 OK up{job="api-service"} # Expected: 1 for all instances
2. Check Error Rate: Current error percentage and breakdown by endpoint sum(rate(http_requests_total{status=~"5.."}[5m])) / sum(rate(http_requests_total[5m])) * 100 # Expected: < 1%
3. Analyze Logs: Recent errors and top error messages from Loki {job="api-service"} |= "error" | json | level="error"
4. Check Resource Utilization: CPU, memory, and connection pool status avg(rate(container_cpu_usage_seconds_total{pod=~"api-service.*"}[5m])) * 100 # Expected: < 70%
5. Review Recent Changes: Deployments, git commits, infrastructure changes
6. Examine Dependencies: Downstream service health, database/API latency

Failure pattern decision tree (excerpt):

• Service down? → Check all pods/instances
• Error rate elevated? → Check specific error types (5xx, gateway, database, timeouts)
• When did it start? → After deployment (rollback), gradual (resource leak), sudden (traffic/dependency)

Expected: Diagnostic procedures are specific, include expected vs actual values, guide responder through investigation.

On failure:

• Test queries in actual monitoring system before documenting
• Include screenshots of dashboards for visual reference
• Add "Common mistakes" section for frequently missed steps
• Iterate based on feedback from incident responders

Step 3: Define Resolution Procedures

See Extended Examples for all 5 resolution options with full commands and rollback procedures.

Document step-by-step remediation with rollback options.

Five resolution options (brief summary):

1. Rollback Deployment (fastest): For post-deployment errors kubectl rollout undo deployment/api-service Verify → Monitor → Confirm resolution (error rate < 1%, latency normal, no alerts)
2. Scale Up Resources: For high CPU/memory, connection pool exhaustion kubectl scale deployment/api-service --replicas=$((current * 3/2))
3. Restart Service: For memory leaks, stuck connections, cache corruption kubectl rollout restart deployment/api-service
4. Feature Flag / Circuit Breaker: For specific feature errors or external dependency failures kubectl set env deployment/api-service FEATURE_NAME=false
5. Database Remediation: For database connections, slow queries, pool exhaustion -- Kill long-running queries, restart connection pool, increase pool size

Universal verification checklist:

• Error rate < 1%
• Latency P99 < threshold
• Throughput at baseline
• Resource usage healthy (CPU < 70%, Memory < 80%)
• Dependencies healthy
• User-facing tests pass
• No active alerts

Rollback procedure: If resolution worsens situation → pause/cancel → revert → reassess

Expected: Resolution steps are clear, include verification checks, provide rollback options for each action.

On failure:

• Add more granular steps for complex procedures
• Include screenshots or diagrams for multi-step processes
• Document command outputs (expected vs actual)
• Create separate runbook for complex resolution procedures

Step 4: Establish Escalation Paths

See Extended Examples for full escalation levels and contact directory template.

Define when and how to escalate incidents.

When to escalate immediately:

• Customer-facing outage > 15 minutes
• SLO error budget > 10% depleted
• Data loss/corruption or security breach suspected
• Unable to identify root cause within 20 minutes
• Mitigation attempts fail or worsen situation

Five escalation levels:

1. Primary On-Call (5 min response): Deploy fixes, rollback, scale (up to 30 min solo)
2. Secondary On-Call (auto after 15 min): Additional investigation support
3. Team Lead (architectural decisions): Database changes, vendor escalation, incidents > 1 hour
4. Incident Commander (cross-team coord): Multiple teams, customer comms, incidents > 2 hours
5. Executive (C-level): Major impact (>50% users), SLA breach, media/PR, outages > 4 hours

Escalation process:

1. Notify target with: current status, impact, actions taken, help needed, dashboard link
2. Handoff if needed: share timeline, actions, access, remain available
3. Don't go silent: update every 15 min, ask questions, provide feedback

Contact directory: Maintain table with role, Slack, phone, PagerDuty for:

• Platform/Database/Security/Network teams
• Incident Commander
• External vendors (AWS, database vendor, CDN provider)

Expected: Clear criteria for escalation, contact information readily accessible, escalation paths aligned with organizational structure.

On failure:

• Validate contact information is current (test quarterly)
• Add decision tree for when to escalate
• Include examples of escalation messages
• Document response time expectations for each level

Step 5: Create Communication Templates

See Extended Examples for all internal and external templates with full formatting.

Provide pre-written messages for incident updates.

Internal templates (Slack #incident-response):

1. Initial Declaration: 🚨 INCIDENT: [Title] | Severity: [Critical/High/Medium] Impact: [users/services] | Owner: @username | Dashboard: [link] Quick Summary: [1-2 sentences] | Next update: 15 min
2. Progress Update (every 15-30 min): 📊 UPDATE #N | Status: [Investigating/Mitigating/Monitoring] Actions: [what we tried and outcomes] Theory: [what we think is happening] Next: [planned actions]
3. Mitigation Complete: ✅ MITIGATION | Metrics: Error [before→after], Latency [before→after] Root Cause: [brief or "investigating"] | Monitoring 30min before resolved
4. Resolution: 🎉 RESOLVED | Duration: [time] | Root Cause + Impact + Follow-up actions
5. False Alarm: No impact, no follow-up needed

External templates (status page):

• Initial: Investigating, started time, next update in 15 min
• Progress: Identified cause (customer-friendly), implementing fix, estimated resolution
• Resolution: Resolved time, root cause (simple), duration, prevention measures

Customer email template: Timeline, impact description, resolution, prevention, compensation (if applicable)

Expected: Templates save time during incidents, ensure consistent communication, reduce cognitive load on responders.

On failure:

• Customize templates to match company communication style
• Pre-fill templates with common incident types
• Create Slack workflow/bot to populate templates automatically
• Review templates during incident retrospectives

Step 6: Link Runbook to Monitoring

See Extended Examples for complete Prometheus alert configuration and Grafana dashboard JSON.

Integrate runbook with alerts and dashboards.

Add runbook links to Prometheus alerts:

- alert: HighErrorRate
  annotations:
    runbook_url: "https://wiki.example.com/runbooks/high-error-rate"
    dashboard_url: "https://grafana.example.com/d/service-overview"
    incident_channel: "#incident-platform"

Embed quick diagnostic links in runbook:

• Service Overview Dashboard
• Error Rate Last 1h (Prometheus direct link)
• Recent Error Logs (Loki/Grafana Explore)
• Recent Deployments (GitHub/CI)
• PagerDuty Incidents

Create Grafana dashboard panel with runbook links (markdown panel listing all incident runbooks with on-call and escalation info)

Expected: Responders can access runbooks directly from alerts or dashboards, diagnostic queries pre-filled, one-click access to relevant tools.

On failure:

• Verify runbook URLs are accessible without VPN/login
• Use URL shorteners for complex Grafana/Prometheus links
• Test links quarterly to ensure they don't break
• Create browser bookmarks for frequently used runbooks

Validation

• Runbook follows consistent template structure
• Diagnostic procedures include specific queries and expected values
• Resolution steps are actionable with clear commands
• Escalation criteria and contacts are current
• Communication templates provided for internal and external audiences
• Runbook linked from monitoring alerts and dashboards
• Runbook tested during incident simulation or actual incident
• Feedback from responders incorporated into runbook
• Revision history tracked with dates and authors
• Runbook accessible without authentication (or cached offline)

Common Pitfalls

• Too generic: Runbooks with vague steps like "check the logs" without specific queries are not actionable. Be specific.
• Outdated information: Runbooks referencing old systems or commands become useless. Review quarterly.
• No verification steps: Resolution without verification leads to false positives. Always include "how to confirm it's fixed."
• Missing rollback procedures: Every action should have a rollback plan. Don't trap responders in worse state.
• Assuming knowledge: Runbooks for experts only exclude junior engineers. Write for the least experienced person on rotation.
• No ownership: Runbooks without owners become stale. Assign team/person responsible for updates.
• Hidden behind auth: Runbooks inaccessible during VPN/SSO issues are useless during crisis. Cache copies or use public wiki.

Related Skills

• configure-alerting-rules - Link runbooks to alert annotations for immediate access during incidents
• build-grafana-dashboards - Embed runbook links in dashboards and diagnostic panels
• setup-prometheus-monitoring - Include diagnostic queries from Prometheus in runbook procedures
• define-slo-sli-sla - Reference SLO impact in incident severity classification