Why can metrics show a problem when I cannot find a matching trace?

Trace sampling can remove request-level span evidence. Metrics generated independently or before trace sampling can still show aggregate symptoms. Metrics derived only from sampled spans represent sampled traffic.

Why is a trace incomplete?

Spans can arrive at different times, be sampled at different boundaries, fail export, or lose propagated context. Confirm span completion, compatible propagation, exporter health, and the selected time range.

Why does one service appear under several names?

The service name changed between instances or deployments. Use one stable service name and represent environment, version, and instance as separate resource attributes.

Can sampled-out or expired traces be recovered?

No. Changing sampling or retention affects future data and does not restore traces that were not retained or have expired.

Which languages are supported?

Manual OpenTelemetry SDK guidance is available for C++, .NET, Erlang, Go, Java, JavaScript, PHP, Python, Ruby, Rust, and Swift. Verify the exact runtime and package versions before rollout.

Which propagation format should I use?

Prefer W3C Trace Context for new integrations. Use B3, Jaeger, or another legacy format only where an existing boundary requires it, and configure both sides deliberately.

Why are trace metrics empty?

Confirm recent trace ingestion, the selected time range and filters, and whether span-metric generation is enabled for the deployment. Metric data can arrive after trace search results.

Do ingestion limits block trace traffic?

The reviewed client usage limits provide warning and critical notifications rather than a hard ingestion gate. Confirm behavior for your deployed version.

What should I provide to support?

Provide the client, UTC time range, service and operation, a trace ID when available, component versions, deployment type, reproduction steps, and sanitized errors. Never include credentials or unapproved payload data.