Definition
Traces describe recorded distributed operations as spans with execution relationships, duration, and attributes.
Question it answers
Where did a recorded request spend time, and which dependency failed?
Role in an investigation
Use the span hierarchy to choose a service or dependency to investigate. Join logs with trace and span IDs, and compare similar requests. Parent-child relationships describe execution; they do not establish that the longest span caused the incident.
Correlation fields
| Field or dimension | Purpose |
|---|---|
| trace_id and span_id | Identify the operation and individual span. |
| parent span or span links | Describe execution relationships, including asynchronous work when instrumentation supplies them. |
| service and span attributes | Scope the operation, endpoint, dependency, and runtime. |
Follow checkout into inventory
A checkout request contains an inventory client span that is slower than normal. The inventory server span and its logs help distinguish processing time from retries or transport delay. Missing server instrumentation leaves that distinction uncertain.
Limitations
- Sampled traces are a subset of traffic and may overrepresent particular request types.
- Missing instrumentation or propagation creates gaps.
- Span duration includes waiting as well as work; long duration does not imply high CPU usage.
Supported by
Documented examples, not an exhaustive compatibility list. Features require suitable instrumentation and configuration; availability can depend on the runtime, backend, and subscription.
- OpenTelemetry — Propagated trace context relates spans across instrumented boundaries.
- Grafana Tempo — A configured processor derives observed service edges from recorded spans.
Investigation patterns
- Trace to Logs
- Metrics to Traces
- Deployments to Error Rate
- Services to Dependencies
- Infrastructure to Applications
- Trace to Profiles
- Service to Database
- User Session to Backend Trace
- Queue Producer to Consumer
Related concepts
- Correlation IDs
- Log Correlation
- Metric Correlation
- Trace Correlation
- Deployment Correlation
- Context Propagation
- Root Cause Analysis
Related guides
- What is Observability?
- What is MTTR?
- OpenTelemetry: Logs to Traces in Python
- Grafana: Metrics to Traces with Exemplars
- Deployments: Compare Error Rate by Version
- OpenTelemetry: Queue Context, Retries and Batches
- Troubleshoot: Log Trace ID Not Found
- Troubleshoot: Exemplars Missing in Grafana
FAQ
Is an incomplete trace useless?
No. Recorded spans can narrow the investigation, provided you keep missing boundaries and sampling limits explicit.