Investigate requests and traces

Locate the operation

Open Observability → Requests and select a range containing the event. Narrow the list using service, method, status, capture state, or search. Select a request to inspect its request, response, and context tabs.

Compare its HTTP status and measured duration with the request statistics for that selection. A successful HTTP response does not prove all downstream operations succeeded; inspect the trace and logs before concluding that a dependency was healthy.

Error rate counts HTTP 5xx responses or server spans explicitly marked as failed, divided by matching requests. Ordinary 4xx responses do not increase it. The separate 4xx rate counts client responses, with a clickable status-code breakdown and filters for codes such as 401, 403, and 429. Successful shows 2xx/3xx responses that are not marked as failed; 4xx responses are not labeled successful. A 4xx span explicitly marked as failed appears in both rates. All requests and logs remain captured under your existing capture settings.

Understand captured details

Request spans can provide method, route, status, duration, service, and trace context without recording bodies. Payload capture is explicitly opt-in.

• Metadata means the operation is visible without a full eligible payload capture.
• Full identifies captured request/response data; capture limits still apply.
• Redacted identifies evidence containing redacted capture content.

Check whether individual headers or bodies were captured and whether they were truncated. A missing body is not an empty body: capture may be disabled, content may be ineligible, or there may be no recording span. Redacted fields are not recoverable from the inspector.

Follow the trace

Open the linked trace, or find it in Observability → Traces using the event range, search, and errors-only filter. Select a trace to load its spans and inspect the waterfall.

Use parent/child relationships and offsets to see where time was spent. Spans can overlap, so summing every duration does not calculate total request time. Select a span to inspect its reported attributes, resource attributes, events, and links.

Trace-list summary statistics cover the selected period, not necessarily the filtered rows. Changing search or errors-only narrows the list; it does not redefine every aggregate shown above it.

Correlate logs and add missing work

Request details can show correlated logs. For deeper inspection, open Logs and follow reported trace context. Logs emitted outside an active span may have no request association.

Add an application span when a meaningful operation is not already instrumented:

import { withOutraySpan } from "@outray/observability";

const order = await withOutraySpan(
  "load order",
  () => database.orders.findOne({ id: orderId }),
  { attributes: { "db.system.name": "postgresql" } },
);

database and orderId above are your application objects. The helper joins the active trace, records thrown errors, and closes the span. Keep span names stable; put only safe, bounded context in attributes.

Ask Agent about the request

Use Explain request to open an Agent investigation attached to the operation. Agent can inspect safe request and trace metadata, correlated log hints, and comparisons. It does not replay the request, read secret values, or change configuration.

If a trace or log is absent, check instrumentation order and context propagation, then consider export delay and the selected range. Missing evidence and a failed data query are different conditions; neither justifies inventing the omitted work.