Appendix I: Operation Reference

Severity Definitions with Response Time Targets

LevelMeaningFirst ResponseFix Target
SEV1User-facing outage15 minutes4 hours
SEV2Degraded experience for a subset of users30 minutes8 hours
SEV3No user impact, fix required1 business dayNext release
SEV4Cosmetic or internal only1 weekWhen scheduled

The team chooses these targets. Adjust based on user expectations and team size.

Runbook Template

Title: [Incident type name]
Severity: [SEV1/SEV2/SEV3/SEV4]
Symptoms: [What alerts or user reports trigger this runbook]

Steps:
1. Verify the symptom
2. Check the affected components
3. Apply the known fix
4. Verify the fix
5. Communicate status

Escalation: [Who to contact if the steps do not resolve the incident]

Write a runbook for every incident type that repeats twice.

SLO Design Patterns

MetricTypical SLOMeasurement Window
Service uptime99.9%30 days rolling
p95 latency500ms5 minutes averaged over 30 days
p99 latency1s5 minutes averaged over 30 days
Error rate<1% of requests10 minutes averaged over 30 days

Do not use the same SLO for every service. A batch job has different targets than an API.

On-call Rotation Template

ElementRecommendation
Rotation length1 week
Team size per rotationMinimum 2 (primary + secondary)
Secondary escalationNames a person, not a team
Handover15-minute sync at rotation change. Review open incidents and known issues.

Adjust based on team size and incident frequency.

Production Readiness Checklist

Before a service receives production traffic:

  • Logs are collected and searchable
  • Metrics are collected and visible on the team dashboard
  • Alerts exist for likely failure modes
  • Deploy process is documented and repeatable
  • Rollback process is documented and tested
  • Data backup strategy is defined
  • Secrets and credentials are not in the code repository
  • Dependencies are documented (databases, APIs, third-party services)
  • Contact for each dependency is known
  • Runbook exists for known failure modes

The team adds items to this list as the system grows.

Cost Review Template

Review these once per cycle:

ItemQuestion
Total infrastructure costIs the trend flat or growing?
Cost per userIs it going up or down?
Cost per serviceDoes any service consume more than 20% of the budget?
Largest cost driverWhat is the single biggest line item?

Act when cost grows faster than user growth or one service exceeds 20% of the budget.