Observability

Health signals tell you whether a component can serve traffic. Traces help explain a request’s path. Audit records explain governed decisions and administrative actions. Use each signal for its intended question.

Readiness before throughput

Monitor control-plane and data-plane readiness, dependency failures, projection freshness, and managed replica acknowledgement where applicable. A high request success rate on one replica can hide a second replica that has not accepted a new configuration.

Trace a request

Use customer-controlled OpenTelemetry to follow latency and failures through the configured runtime. Correlate a slow call with routing, credential resolution, and upstream response behavior. Keep secrets and raw tool payloads out of routine telemetry.

Verify audit delivery

Inspect a known test request in the audit search window. If you configure external export, verify that same event at the destination. Configuration readiness and observed delivery are separate facts.

During an incident

Record the time window, affected scope, safe reason codes, and configuration versions. After a correction, compare the same signals and run the original test again. See troubleshooting for a symptom table.

Implementation reference

Read the source contract.