cpu-profile-analysis

Parse the JSON to build a node map from the nodes array, then walk samples and timeDeltas to compute self-time per function. Sort functions by self-time descending to find the top cost centers, filtering out idle, program, and garbage collector frames.

A .cpuprofile contains only V8 sampling profiler data (nodes, samples, timeDeltas). A Trace-*.json uses the Chrome Trace Event Format and additionally includes layout/paint events, user timing marks, GC events, input events, and multi-process data from Renderer, Browser, and GPU processes.

V8 cannot create strings larger than roughly 512MB, so readFileSync with utf8 encoding fails on huge profiles. Read the file as a raw Buffer instead, scan for JSON key boundaries by bytes, and parse small sub-buffer sections individually.

Filter traceEvents for RunTask entries with phase X on the CrRendererMain thread where dur exceeds 50000 microseconds. Sort by duration descending to surface the tasks that blocked rendering the longest.

Yes. Trace files embed the full CPU profile as Profile and ProfileChunk events in the disabled-by-default-v8.cpu_profiler category. Merge the nodes, samples, and timeDeltas from all chunks to reconstruct a complete cpuprofile-equivalent structure.

Sampling profilers only capture functions on the stack at each tick, so exact names are unreliable, and bundled code may have mangled names. Use sets of related marker functions and cross-reference callFrame.lineNumber with source maps.