Definition
Metrics summarize numerical observations over time, such as request counts, latency distributions, and resource usage.
Question it answers
Which service or cohort changed, and how large is the impact?
Role in an investigation
Start an investigation with the affected population and time range. Compare request rate, errors, and latency before selecting a representative trace. Keep the aggregation dimensions visible so a fleet-wide average does not conceal a failing instance.
Correlation fields
| Field or dimension | Purpose |
|---|---|
| service and environment dimensions | Match the scope of the chart to the scope of other telemetry. |
| time window and aggregation | Compare equivalent intervals and understand whether values are rates, totals, or distributions. |
| exemplar trace reference | Link selected observations to recorded requests without creating per-trace metric series. |
Narrow a latency regression
Checkout p95 latency rises while request volume stays steady. Splitting by version isolates the newer cohort; an exemplar supplies one request to inspect. Check more requests before treating that example as representative of the full regression.
Limitations
- Aggregates omit per-request details and can hide differences between cohorts.
- Percentiles generally cannot be averaged across instances to obtain a fleet percentile.
- Exemplar availability depends on collection, sampling, storage, and retention.
Supported by
Documented examples, not an exhaustive compatibility list. Features require suitable instrumentation and configuration; availability can depend on the runtime, backend, and subscription.
- Grafana with Prometheus and Tempo — Configured exemplars provide selected metric-to-trace links.
Investigation patterns
- Metrics to Traces
- Deployments to Error Rate
- Alerts to Incidents
- Services to Dependencies
- Infrastructure to Applications
- Trace to Profiles
- Service to Database
- Queue Producer to Consumer
Related concepts
- Metric Correlation
- Trace Correlation
- Event Correlation
- Alert Correlation
- Deployment Correlation
- Root Cause Analysis
Related guides
- What is Observability?
- What is MTTR?
- Grafana: Metrics to Traces with Exemplars
- Deployments: Compare Error Rate by Version
- Troubleshoot: Exemplars Missing in Grafana
FAQ
Does every metric point have a trace?
No. Metrics may describe many operations or infrastructure state. Exemplars link selected observations only.