Support to manually aggregate reports? (e.g. after joining parallel jobs?)

[EricCousineau-TRI]

Problem

I'd like to aggregate reports across multiple processes, e.g. using multiprocessing.Pool.
(In my case, it's a manual worker dispatch based on mp.Process - see here for an old port of the code).

Motivating Example

In my current code, I do something like this:

def worker(values):
    for value in values:
        out, timing = do_work(value)
        yield out, timing

def main():
    values = range(n)  # etc.
    results = parallel_work(worker, values)
    outs, timings = zip(*results)
    # Print aggregation of "timings" reports.

More concretely, here's the example code (doesn't have all deps, but it communicates the intent):
https://github.com/EricCousineau-TRI/repro/blob/54494a5c5154f19e693e4862fbaa79cddcd78d6f/drake_stuff/multibody_plant_prototypes/generate_poses_sink_clutter.py#L507-L516

Request

Is there an easy way to aggregate results themselves using public API?

Currently, it looks like aggregation is done internally:

stopwatch/stopwatch.py

Lines 303 to 306 in 94f59aa

	agg_report = AggregatedReport(self._reported_values, tr_data)
	# Stash information internally
	self._last_trace_report = self._reported_traces
	self._last_aggregated_report = agg_report

[nipunn1313]

You can print them out with get_last_aggregated_report (or format_last_report for pretty)

get_last_trace_report should help you out for non-aggregated ones (to get information for every span).

Stopwatch (at least as currently designed) may not be great for understanding how the parallel tasks perform together. However, it may be useful for understanding how the time is being spent within a task.

[EricCousineau-TRI]

Thanks! In a week or so I'll see if I can try to write a unittest here to show that workflow (of using get_last_trace_report, and then adding multiple sets of reports together)