Tracing two services without losing the plot

Tracing shines when you can follow a user action across boundaries. We start with manual span IDs in logs, then graduate to propagation headers participants can grep in staging. Malaysian teams often run hybrid stacks; we emphasise vendor-neutral naming so dashboards stay readable after tool churn. Sampling comes up early — nobody benefits from storing every health check. Incident writing is the hidden outcome: if your trace cannot answer the on-call question, you iterate on instrumentation, not dashboards alone.