10. Observability and Reliability

A production system must prove what it is doing. OpenTelemetry defines traces, metrics, and logs as major telemetry signals. SLOs connect these signals to a target level of service reliability. Health checks should distinguish whether a process is alive from whether it is ready to receive traffic.

Evidence: S22, S23, S26

Checklist

• Critical user journeys have SLIs and SLOs.

• Availability and latency are measured from a user-relevant point.

• Logs include enough context to investigate failures without exposing sensitive data.

• Metrics cover traffic, errors, latency, saturation, and business-critical signals.

• Distributed traces propagate across service boundaries where useful.

• Correlation IDs or trace context are consistent.

• Alerts are actionable and tied to user impact or real operational risk.

• Readiness and liveness checks have different semantics where the platform supports them.

• Health checks do not create cascading failure under load.

• Backup policy is defined for durable data.

• Restore is tested.

• RPO/RTO are technically achievable by the selected design.

• Dependency health and external limits are observable.

• Dashboards show both current health and capacity trend.

Evidence: S22, S23, S24, S26