Page Extractor observability

One extraction can cross the parent and content processes, run in several stages, and depend on a site that behaves differently in a hidden browser. Page Extractor records each instrumented stage as a Glean event and, while the profiler runs, as a profiler marker, so you can investigate failures and performance as one flow.

Each record includes the stage, process, outcome, strategy, and size information. Nested stages overlap, so do not add their durations together. A shared flow identifier connects work across processes and between profiler and telemetry data.

Investigate an extraction

Start the profiler from the toolbar button or about:profiling.
Run the extraction.
Look for PageExtractor #n tracks in the marker chart.

Markers are blue for success, red for thrown errors, and yellow for handled outcomes such as unavailable or empty content. The marker details name the stage and strategy that produced the outcome. In a hidden browser, pass the callback’s traceId into getText so loading and extraction stay in the same flow.

Instrumenting JavaScript explains the profiler mechanism.

Telemetry

Finished stages record page_extractor.phase events in the page-extractor ping, subject to Glean collection settings. The event holds operational information such as outcome, duration, strategy, and coarse size counts. It holds no page text, URLs, image data, or error messages. The ping has no client identifier and uses OHTTP.

When you analyze the events, pick the stage that represents the behavior you are measuring. Counting every nested stage counts one failure more than once. Group records by flow identifier to follow a request across processes.

The events module and the Glean metrics and pings definitions next to it define the phases and fields. The design history is in bug 2065318 and bug 2058247.