Severity Classification

Immediate Response (First 5 Minutes)

1. Acknowledge the alert — post in incident channel: "Investigating [alert name]"
2. Assess scope: How many users affected? Which endpoints?
3. Check recent changes: git log --since="30 minutes ago" — was there a recent deploy?
4. Check the dashboard: error rate, p99 latency, connection pool utilization, cache hit rate

Investigation Decision Tree

Is error rate > 50%?
  YES → Check: recent deploy? → YES: rollback
        Check: DB connectivity? → NO: escalate infra
        Check: dependency outage? → YES: enable circuit breaker
  NO  → Check: latency elevated?
    YES → Check: DB query count spiking? → YES: N+1 regression
          Check: connection pool exhausted? → YES: see pool runbook
          Check: cache hit rate dropped? → YES: cache cold start
    NO  → False alarm or intermittent; investigate traces

Rollback Procedure

# Kubernetes
kubectl rollout undo deployment/my-service
kubectl rollout status deployment/my-service

# Verify
curl -w "@curl-format.txt" https://api.example.com/health

Escalation

Escalate to on-call lead if:

• P0 incident not resolved in 15 minutes
• Multiple services affected
• Data integrity concern (writes may be corrupted)
• Security concern (potential breach)

Post-Incident

Within 48 hours:

1. Write blameless post-mortem
2. Add monitoring that would have caught this sooner
3. Document in ../../case-studies/

Related Modules

• ../../bsps/10-production-systems/02-incident-response.md — theory
• ../../bsps/10-production-systems/01-observability.md — what to look at