LEARN. PRACTICE. GROW.Concept and mechanism
An application can wait for network or database work while using little CPU. Correlated traces help relate requests and dependencies when instrumentation and context propagation are configured. Choose a mechanism supported by the language and hosting; current Application Insights guidance uses OpenTelemetry for supported scenarios. Check that signals arrive and represent the intended operation. A dashboard created without validation traffic does not demonstrate that critical dependencies are instrumented.
Guided application
Interpret timing structure before adding durations. Sequential calls accumulate time; overlapping spans should not all be added to the critical path. Distinguish one sample from the distribution of many requests. Standard availability tests observe configured requests and can check response and certificate, but a healthy home page does not prove batch completion. Combine technical signals with business-outcome evidence. When communicating an incident, identify measurement, window, affected population, and limitations while keeping sensitive content out of logs.
Three sequential calls of 120, 180, and 250 ms plus 50 ms locally total 600 ms. That describes this trace, not the global p95.
Common pitfalls
Adding overlapping spans; confusing endpoint availability with process success; concluding low CPU rules out a performance issue.
Related topics: Messages, events, and API contracts · Versions and controlled deployment
Explain what evidence measures and what still needs observation.
Reference: Application Insights OpenTelemetry overview · AZ-204 archived objectives 2026-01-14; retired 2026-07-31