Feedback on the proposal

[jasnell]

First off, a thank you @littledan for pointing me at this proposal. Overall I think it's a great start and a great addition to the API surface here. It should be fairly straightforward to implement support for this in Node.js and definitely think it's something we would do.

A couple of comments on the API... May have more later but these are what jumps out immediately.

1. An alternative to specifying a sample interval would be to specify a sample rate ... for instance, something like await performance.profile({ hz: 20 }) (sample at approximately 20 samples per second).

2. The ProfilerTrace object can potentially be a bit heavy in terms of memory load. It would be very interesting to support a streaming mode as an alternative that just outputs a raw stream of profile data rather than building up the index. Imagine an API like...

const profiler = await performance.profile({ sampleInterval: 10 });
for (let i = 0; i < 1000000; i++) {
     doWork();
}
profiler.stop();

sendTrace(await profiler.trace());  // returns the parsed `ProfilerTrace`

// alternatively

const profiler = await performance.profile({ sampleInterval: 10 });
profiler.reader().pipeTo(writable);  // Consumes the profile as a raw stream of events

for (let i = 0; i < 1000000; i++) {
     doWork();
}

profiler.stop();

The stream approach would generate more raw unprocessed data but would allow it to be piped out and processed later much more efficiently.

[nicjansma]

For (2), a stream is interesting because it could provide a workaround for cases where the Profiler is active and the page is unloading. In that case, the website may wish to stop-and-send the profile, but the Promise from Profiler.stop() may prevent that from happening (since the page will move-on with unloading before the Promise resolves).

If there was a stream available, the website code could maintain the profile stream on its own, and be able to beacon the profile at unload without waiting for a Promise.

I think the challenge is that there are multiple parts to "stream" though, looking at the current output of a Profiler.stop(), where you get frames, resources, samples and stacks top-level objects. Each of those would need to be "streamed" in a way, or the stream would need to be notified if there were any new entries in any of the above.