Span Metrics

Description

This connector automatically generates Request, Error, and Duration (R.E.D) metrics from incoming trace spans.

Use

Use this connector to:

• Monitor request rates, error rates, and latency across services without manual metric instrumentation

• Build R.E.D dashboards and alerts directly from existing trace data

• Add custom span attributes as metric dimensions for more granular analysis

• Correlate metrics back to traces using exemplars

Stability

• Alpha

Supported Types

Input	Output
Traces	Metrics

Generated Metrics

The connector produces the following metrics (prefixed by the namespace, default traces.span.metrics):

Metric	Type	Description
calls	Sum	Count of spans. Error counts are derived via the status.code attribute.
duration	Histogram	Distribution of span durations
events	Sum	Count of span events (optional, disabled by default)

Every metric includes these default attributes:

Attribute	Description
service.name	The service that produced the span
span.name	The operation name
span.kind	The span kind (client, server, internal, producer, consumer)
status.code	The span status code (Ok, Error, Unset)

Configuration

Field	Description	Default
Metric Namespace	Prefix for generated metric names. Leave empty for the default (traces.span.metrics).	(empty)
Aggregation Temporality	How metrics are aggregated over time: Cumulative or Delta.	Cumulative
Metrics Flush Interval	How often generated metrics are flushed. Examples: 15s, 1m, 60s.	60s

Attributes

Control which span and resource attributes appear as dimensions on the generated metrics.

Field	Description	Default
Additional Attributes	Span or resource attributes to include as metric dimensions on all generated metrics. Map attribute name to a default value (leave value empty if the attribute should be omitted when missing).	(none)
Exclude Default Attributes	Remove specific default attributes (service.name, span.name, span.kind, status.code) from metrics.	(none)
Calls Metric Attributes	Additional attributes for the calls metric only, on top of the common attributes above.	(none)

Duration Histogram

Configure how span duration is measured and bucketed.

Field	Description	Default
Enable Duration Histogram	Generate a duration histogram metric. When disabled, only call count metrics are produced.	true
Histogram Type	Explicit uses fixed bucket boundaries. Exponential uses dynamic boundaries based on scale.	Explicit
Duration Unit	Unit for duration measurements: ms (milliseconds) or s (seconds).	ms
Bucket Boundaries	Custom histogram bucket boundaries. Leave empty for defaults: 2ms, 4ms, 6ms, 8ms, 10ms, 50ms, 100ms, 200ms, 400ms, 800ms, 1s, 1400ms, 2s, 5s, 10s, 15s.	(default buckets)
Exponential Max Size	Maximum number of buckets per positive or negative range (exponential histograms only).	160
Duration Histogram Attributes	Additional attributes for the duration histogram metric only, on top of the common attributes above.	(none)

Advanced Configuration

The following settings are available under the advanced configuration toggle.

Exemplars

Exemplars attach references from metric data points back to the original trace spans that produced them. This enables drill-down from a metric chart directly to the relevant trace.

Field	Description	Default
Enable Exemplars	Attach trace exemplars to histogram metrics.	false
Max Exemplars Per Data Point	Maximum number of exemplars to attach per histogram data point.	5

Events

Generate a counter metric for span events. This is useful for tracking exception counts or custom application events.

Field	Description	Default
Enable Events Metric	Generate a metric counting span events.	false
Event Attributes	Span event attributes to include as dimensions on the events metric. At least one is required when events are enabled. Map attribute name to a default value (leave value empty if the attribute should be omitted when missing).	(none)

Tuning

These settings control performance and caching behavior. For most deployments the defaults are appropriate.

Field	Description	Default
Metrics Expiration	Time after which metrics stop being exported if no new spans arrive. 0 means never expire.	0
Resource Metrics Cache Size	Size of the internal cache holding metrics per resource.	1000
Metric Timestamp Cache Size	Cache size for tracking metric timestamps (used with delta temporality).	1000
Aggregation Cardinality Limit	Maximum unique dimension combinations. 0 means unlimited.	0
Include Instrumentation Scope	Only process spans from these instrumentation scope names. If empty, all scopes are included.	(all)
Resource Metrics Key Attributes	Resource attributes used to determine resource identity for metric grouping. If empty, all resource attributes are used.	(all)

For more information on configuration options, see the OpenTelemetry documentation.

Example Configurations

Basic R.E.D Metrics

Generate request, error, and duration metrics with default settings. No additional configuration is needed — just add the connector between a traces source and a metrics destination.

• Aggregation Temporality: Cumulative

• Metrics Flush Interval: 60s

• Enable Duration Histogram: true

HTTP Latency with Custom Buckets

Track HTTP request latency with custom histogram buckets and the HTTP method as a dimension.

• Additional Attributes: http.request.method (no default value)

• Enable Duration Histogram: true

• Histogram Type: Explicit

• Duration Unit: ms

• Bucket Boundaries: 5ms, 10ms, 25ms, 50ms, 100ms, 250ms, 500ms, 1s, 2s, 5s

Error Tracking with Exemplars

Enable exemplars to correlate error metrics back to the original traces.

• Additional Attributes: http.response.status_code (no default value)

• Enable Duration Histogram: true

• Enable Exemplars: true (advanced)

• Max Exemplars Per Data Point: 5 (advanced)

With exemplars enabled, clicking on a metric data point in a compatible backend (e.g. Grafana with Tempo) will link directly to the trace that produced it.

Exception Counting with Events

Count span events by exception type across services.

• Enable Events Metric: true (advanced)

• Event Attributes: exception.type (no default value) (advanced)

This produces a traces.span.metrics.events counter with exception.type as a dimension, letting you track exception rates by type.

Technical Notes

Flush interval: Metrics are emitted on a timer (default 60 seconds). After adding the connector, wait at least one flush interval before expecting metrics to appear.

Single writer principle: When multiple collectors generate the same cumulative time series, most backends will treat that as a single writer violation. Enable collector.instance.id on the generated metrics to ensure each collector writes uniquely identified series. This requires the feature gate connector.spanmetrics.includeCollectorInstanceID. It is enabled by default in BDOT.

High cardinality warning: The span.name attribute is included by default. Non-normalized span names (e.g. GET /user/12345 instead of GET /user/{id}) can cause metric cardinality to explode. Use a Transform processor upstream to normalize span names, or set an Aggregation Cardinality Limit as a safety net.

Related Resources

• OTel spanmetrics connector docs