signal-choice-advisor

OpenTelemetry signal selection depends on your data shape: use metrics for counts and durations, spans for distributed tracing workflows, and structured logs for discrete point-in-time events with trace context attached.

Use a Histogram to measure request duration when you need aggregated latency distributions and percentiles. Use a Span when you need to trace the causal path and execution context of individual requests across distributed service boundaries.

Metric cardinality refers to the number of unique label combinations in a metric. Prevent high-cardinality mistakes by avoiding high-cardinality identifiers like user IDs or raw IP addresses in metric dimensions, using resource attributes instead.

Yes, you should use OpenTelemetry semantic conventions before creating custom metric names. Standard conventions ensure interoperability and reduce redundancy, providing a verified workflow to model duration, counts, and events semantically correct.

Yes, you can attach trace context to structured logs in OpenTelemetry. This combines discrete point-in-time event recording with distributed tracing, allowing you to correlate specific log entries with their originating spans for better debugging.

The best way to model point-in-time events is using structured logs with trace context, while aggregated counts require UpDownCounter or Counter instruments. Gauges capture instantaneous values, preventing the need for high-cardinality metric dimensions.