Skip to main content
Telemetry is in early access. To turn it on for your workspace, book a call or ask in Discord.
Explore → Traces has two pages: Overview for a service’s health, and Search for individual traces.

​
Traces overview

The overview summarizes requests. A request is a span that enters a service: a server or consumer span, or the root span of a trace. Pick a Service or keep All services. When spans carry the deployment.environment resource attribute, an Environment list appears too. Choose a range of 15m, 1h (the default), 6h, 24h, or 7d. Turn on Live to reload every 10 seconds. Three tiles chart the range:
  • Requests per minute, with errors marked.
  • Error rate as a share of requests.
  • p95 latency. Bars turn orange from 500 ms and red from 1 s.
Below them:
  • Operations: the top request span names, with Req/min, Errors, and p95. Click one to search its traces.
  • Dependencies: outgoing client calls by p95 latency. Each is named by peer.service, server.address, net.peer.name, or db.system, whichever the span has, and otherwise by its span name.
  • Top errors: failing spans grouped by status message, with how many traces they hit and when they were first seen. Click one to search those traces, or View all in Search.
Traces → Search lists traces that started in the time range, newest first. It uses the same time picker as Logs. The search bar works like the log search. Press / to focus it. Fields include:
  • service.name, status (Error, OK, Unset), and operation (the span name). Plain text filters to that exact span name.
  • duration with >= and a value such as 500ms, 1s, or 1m, up to 60 minutes.
  • trace.id. Paste a trace ID to open that trace directly.
  • Any resource or span attribute seen in the range.
The sidebar has Service, Status, Span name, and Trace duration (Over 100 ms to Over 5 s), and every attribute key below them. A trace matches when any of its spans matches the filters. Its row describes the whole trace: The histogram above counts matching spans by status. The list shows 50 traces per page.

​
Trace inspector

Click a trace to open it in a panel beside the list. Use the arrows to move to the previous or next trace, or Open full page for a page of its own. The header shows the request: method, route, and HTTP status code when the root span has them. Below it are the root service, duration, and counts of spans, services, and errors, with the Trace ID to copy.

​
Waterfall

The Waterfall tab draws each span on a timeline, nested under its parent and colored by service.
  • Find a span or attribute matches span names, services, span IDs, and attribute keys and values. Matching spans stay visible with their parents.
  • Show only errors hides spans that didn’t fail.
  • Collapse all and Expand all fold the tree. Click a span’s arrow to fold one branch.
The inspector opens on the first failing span, or the root span if none failed.

​
Selected span

Click a span to see its name, service, kind, duration, start offset, and Span ID. A failing span shows its error message and stack trace. Attributes are grouped:
  • HTTP: http.*, url.*, and server.* attributes.
  • Request context: the span’s other attributes.
  • Resource attributes: the sending process. Collapsed by default.
Hover an attribute for Filter, Exclude, and Copy. Filter and Exclude add it to your trace search. Logs in this span lists the logs the span wrote. Open in Logs opens them in Logs.

​
Logs and services

  • Logs lists every log in the trace. Click a message to select the span that wrote it.
  • Services lists each service with its spans, errors, and total Time in spans.

​
Query traces from an agent

The MCP server has query_traces and get_trace. The management API has POST /telemetry/traces/search and GET /telemetry/traces/{traceId}.