Problem
The tab-navigation spans (ManualNavigateToInboxTab, ManualNavigateToReports*) contain events that don't measure a navigation. Abandoned navigations are canceled but still sent with status OK and wall-clock durations, so a span canceled after the laptop wakes up records the whole absence — up to 2-hour "navigations" in production. Under 1% of web events but 61% of recorded time (web avg 1250ms with them, ~420ms without). Wide vs narrow layouts (~3x cost difference) and warm re-focuses (~30ms) vs cold mounts (~150ms) were also indistinguishable in the data.
Solution
Added canceledTabNavigationFilter in beforeSend to drop canceled tab-navigation spans (only matches tab-navigation ops with the canceled attribute set by cancelSpan). Added new Inbox span attributes for honest slicing: wide_layout, skeleton_shown, and is_warm. Removed dead render-duration instrumentation from useSidebarOrderedReports.
PR
#95480
Issue Owner
Current Issue Owner: @marufsharifi
Problem
The tab-navigation spans (
ManualNavigateToInboxTab,ManualNavigateToReports*) contain events that don't measure a navigation. Abandoned navigations are canceled but still sent with statusOKand wall-clock durations, so a span canceled after the laptop wakes up records the whole absence — up to 2-hour "navigations" in production. Under 1% of web events but 61% of recorded time (web avg 1250ms with them, ~420ms without). Wide vs narrow layouts (~3x cost difference) and warm re-focuses (~30ms) vs cold mounts (~150ms) were also indistinguishable in the data.Solution
Added
canceledTabNavigationFilterinbeforeSendto drop canceled tab-navigation spans (only matches tab-navigation ops with thecanceledattribute set bycancelSpan). Added new Inbox span attributes for honest slicing:wide_layout,skeleton_shown, andis_warm. Removed dead render-duration instrumentation fromuseSidebarOrderedReports.PR
#95480
Issue Owner
Current Issue Owner: @marufsharifi