Traces in Insights

Start from a service, narrow to the spans that describe your problem, then open one request and read it end to end.

Open Insights → Traces.

Pick what a span has to be

The primary-signal selector decides which spans the whole session looks at.

Signal	Scope
Root spans	The root span of each trace
All spans	Raw span data. Expect long query times
Server spans	Server-side segments of traces
Consumer spans	Interactions started by consumer services
Database calls	Database interactions

Root spans is the default and the right starting point for "which requests are slow". Database calls is the shortcut for "is it the database".

Rate, errors, duration

Three panels sit above the results: request rate, error rate and duration. Duration is drawn as a latency heatmap, so a second mode hiding under a p95 line is visible rather than averaged away.

Each panel doubles as a selector. Click a small one to promote it into the big slot.

Narrow by attribute

The attributes rail lists the span attributes present in the current selection. Pick one and the breakdown draws a panel per value, ranked so the highest-rate values come first. Include and Exclude narrow the whole session.

Read one trace

Click a result to open the waterfall.

Critical path highlights the segments that actually determined the trace's wall-clock duration. At each span the last-finishing child is on the path, and the gaps before and after it are the span's own work. A span can be slow and irrelevant. The critical path is how you tell.

Minimap gives you the whole trace at once when it is too long to scroll.

Span detail lists attributes, events and links. Span links are not tree edges, so a linked span is reachable without being a child.

Logs for this span jumps to the log lines recorded while that span ran.

Compare, group, count

Four more views over the same selection:

Comparison splits matching spans into a baseline and a selection, and ranks attributes by how differently their values show up on each side. Use it after a deploy: baseline yesterday, select now, read what changed.

Service graph draws the call graph between services with rate and error rate on the edges.

Service structure returns the matching trees of spans and reassembles them into a forest of service and operation nodes. It answers "what does a request through this service actually touch".

Exceptions pulls exception.type, exception.message and exception.stacktrace from every span with status = error, then groups by message. The count next to each group tells you whether you are looking at the outage or at background noise.

Jump straight to a trace ID

Paste a trace ID into the lookup box. This is the other end of the log-to-trace link, and it is what to do with the trace ID a customer sends you.

Alerting on traces

Trace queries do not offer Create alert. The alert model covers Prometheus and Loki queries. To alert on trace-derived behaviour, alert on the metrics your spans produce. See Alerting.

The same data over HTTP

This page queries the Tempo-compatible endpoint. See Trace query API overview for TraceQL over HTTP.