LEARN. PRACTICE. GROW.Concept and mechanism
A distributed operation requires connecting requests, messages, and effects. Traces show relationships and duration when context propagates correctly; logs explain local events and metrics show frequency and trends. No single signal describes the whole system. Missing spans can result from instrumentation, propagation, or sampling without proving a call never occurred. Retain appropriate technical references and keep credentials or personal data out of baggage. That information can cross boundaries and reach systems that should not receive it. Diagnosis also needs business-operation state and careful interpretation of clocks and timestamps.
Guided application
In a fictional scenario, all containers are running but reconciliation requests accumulate without completion. Inspect the complete path, starting with an affected operation and dependencies showing the longest waits. In Kubernetes, readiness indicates ability to receive traffic; failure is not by itself a restart instruction. Liveness can trigger restart and should represent a problem recoverable that way. When the same external database fails, restarting every service can add load without addressing the cause. A startup probe allows initialization before other probes run. Use these differences to choose signals and actions while retaining recovery criteria observable by the business.
Running containers and HTTP200 on a health endpoint do not establish completed reconciliations.
Common pitfalls
Missing span as nonexistent request; baggage as a vault; equivalent readiness and liveness.
Related topics: Boundaries and contracts · Data and distributed transactions · Resilience and load
Correlate health, waiting, and final effects with the minimum necessary data.
Reference: Context propagation · Microservice architecture patterns and scoped platform examples; primary guidance consulted 2026-09-30