Why architecture matters here

Exemplars fail on missing SDK support, poor sampling, and cardinality explosion. Architecture matters because governance + retention alignment matter as much as instrumentation.

Advertisement

The architecture: every piece explained

The top strip is the mechanism. Metric point counter / histogram. Exemplar attach trace + attrs. Storage TSDB + trace store. Query UI click → trace.

The middle row is control. Sampling policy. Cardinality. Retention align. Alerts + exemplars root-cause.

The lower rows are ops. SDK support OTel Prom. Metrics coverage. Ops — governance + PII.

Metrics exemplars — trace-to-metric link + storage + query + samplingclick a metric spike, land in a traceMetric pointcounter / histogramExemplar attachsampled trace + attrsStorageTSDB + trace storeQuery UIclick exemplar → traceSampling policywhich points get exemplarCardinalitykeep boundedRetentionalign with tracesAlerts + exemplarsroot-cause in alertsSDK supportOTel PromMetricsexemplar coverageOps — governance + PII + link testingsampleboundalignsurfaceinstrumentmeasuremeasuregoverngovern
Metrics exemplars linking metrics to traces.
Advertisement

End-to-end flow

End-to-end: latency histogram accumulates. Exemplars attached to top 1% of buckets. Alert fires on p99 spike; alert body contains exemplar trace ID; oncall opens trace immediately. Root cause visible.