Learning center

Link the browser evidence to the server's logs

A maintainer with access to server logs can explain a failed request in minutes, if they can find it. "Around 3 pm, on the orders page" matches thousands of log lines. A request ID or trace ID from the failing response, and a timestamp in UTC, match one. Collect them from the browser before you write the report.

Look for an identifier in the response#

Open the failing request in the Network panel and read its response headers. Common identifiers include:

HeaderSet by
X-Request-Id, X-Correlation-IdMany frameworks, proxies and API gateways, by convention
traceparentW3C Trace Context, when the server echoes or returns it
X-Amzn-RequestId, X-Amz-Cf-IdAWS services and CloudFront
CF-RayCloudflare
X-Cloud-Trace-ContextGoogle Cloud load balancing

Names vary by stack; the app's own error page or JSON error body may also print an ID. Copy every identifier the response carries, because the CDN's ID and the application's ID lead to different logs.

traceparent follows the W3C Trace Context format: version-traceid-parentid-flags, such as 00-4bf92f3577b34da6a3ce929d0e0e4736-00f067aa0ba902b7-01. The 32-hex-digit trace ID in the middle is the part to search for in a tracing backend.

Record the time in UTC#

The Network panel's request times are relative to the page load. Read an absolute time from the response's Date header, which is in GMT, or note the clock time with its time zone. In the Console, new Date().toISOString() prints the current time in UTC. A report with 2026-11-21T14:05:12Z lets the maintainer narrow the logs to a few seconds.

When the response has no ID#

If nothing in the response identifies the request, give the maintainer enough to filter on:

  • The method and path, such as POST /api/orders
  • The status code and the response's size
  • The time in UTC, to the second
  • The test account's user ID, not its email address, if logs record it

Then suggest the fix that would have made this easy: returning a request ID header on every response, especially errors.

The browser can send one too#

Some applications send their own ID with each request, such as a traceparent header added by a browser tracing library, or an X-Request-Id the client generates. Check the request headers as well as the response headers. A client-sent ID survives even when the response never arrives, which makes it the only link for a request that timed out or failed with no status.

Worked example: an intermittent 502#

This example is fictional. A tester sees 502 Bad Gateway on saving an invoice, once in a while. The developer searched the application's error logs and found nothing.

The tester's report includes the IDs on the 502 response: CF-Ray: 8f1c2a3b4d5e6f70-FRA and no application X-Request-Id. The missing application ID showed that a layer in front of the application wrote the response. It did not show that the application never saw the request, so the developer checked both sides. The CDN's log, found by the ray ID, recorded the origin closing the connection. The load balancer's log at the same UTC second recorded its 30-second idle timeout. The application's access log, searched by path and time, showed the request arriving and finishing successfully after 34 seconds: large invoices took longer to render than the load balancer would wait, so the application logged no error. Without the ray ID and the UTC time, the developer would have kept searching error logs for a failure the application never recorded.

Questions#

Is a request ID safe to share?

Usually yes: it identifies a log entry, not a session. It lets anyone with log access find the request, which may hold personal data, so treat the logs, not the ID, as sensitive.

Should I include IDs from successful requests?

Include one from a successful request made just before the failure when you can. It shows the maintainer which logs and services a normal request passes through, so they can see where the failing one diverged.

What if the server and browser clocks differ?

They often differ by a little. The Date header is the server's clock, so prefer it when lining up with server logs.

Return to reproducing what someone else saw for the other layers a replay needs.

One month of free code review.

New organizations get one calendar month of hosted code review for private and public GitHub repositories. No card or cloud connection required. Eligibility and fair-use terms.

Start free with GitHub

Related reading