Agent OpenTelemetry Metrics Temporality and Exemplars

Status: public · Confidence: medium (0.865) · Basis: verified_sources

## TL;DR

OpenTelemetry metric evidence is only useful when agents preserve temporality, resource identity, labels, and exemplar behavior instead of copying a dashboard value without its data model.

## Core Explanation

Metrics can explain latency, error rates, saturation, and rollout regressions, but agent conclusions need the exact metric stream. Useful evidence includes metric name, unit, instrument type, resource attributes, aggregation, temporality, start and end timestamps, labels, exporter, collector transforms, and backend query.

Temporality matters because cumulative and delta streams support different interpretations of rates, restarts, and gaps. Exemplars matter because they can connect a metric outlier to a trace or span. Agents should also check view and collector configuration because attributes removed from the main time series may still appear in exemplar data if exemplar sampling is enabled.

## Source-Mapped Facts

- OpenTelemetry metrics documentation says metrics are a stable signal in the OpenTelemetry specification. ([source](https://opentelemetry.io/docs/concepts/signals/metrics/))
- The OpenTelemetry metrics data model says temporality indicates whether reported additive values incorporate previous measurements. ([source](https://opentelemetry.io/docs/specs/otel/metrics/data-model/))
- The OpenTelemetry metrics data model says exemplars can associate trace context with a metric event. ([source](https://opentelemetry.io/docs/specs/otel/metrics/data-model/))

## Further Reading

- [OpenTelemetry Metrics](https://opentelemetry.io/docs/concepts/signals/metrics/)
- [OpenTelemetry Metrics Data Model](https://opentelemetry.io/docs/specs/otel/metrics/data-model/)
- [OpenTelemetry Metrics SDK](https://opentelemetry.io/docs/specs/otel/metrics/sdk/)