testing: show rusage statistics for benchmarks

[LMMilewski]

When optimizing CPU usage, it would be useful to get CPU usage information from Go benchmarks.

Could we add, where supported, getrusage calls before and after running each benchmark, in addition to wall-time?

I have no opinion about the API for requesting those statistics. Perhaps adding a -test.stats or -test.resources flag would work?
For example: -test.stats=utime,stime,maxrss.

This doesn't have to be rusage specific. For instance, it could be extended with statistics from the perf_event API at some point.

I could send a CL if this is a welcome change.

[davecheney]

I think the execution tracer is a better, and pre existing, solution to this requirement.

…

On 18 Apr 2018, at 08:51, LMMilewski ***@***.***> wrote: When optimizing CPU usage, it would be useful to get CPU usage information from Go benchmarks. Could we add, where supported, getrusage calls before and after running each benchmark, in addition to wall-time? I have no opinion about the API for requesting those statistics. Perhaps adding a -test.stats or -test.resources flag would work? For example: -test.stats=utime,stime,maxrss. This doesn't have to be rusage specific. For instance, it could be extended with statistics from the perf_event API at some point. I could send a CL if this is a welcome change. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

[josharian]

See also #21295

[LMMilewski]

@davecheney could you clarify your recommendation?

I'm familiar with the execution tracer. I use it when it's a good match for what I want to do.
However, I'm not sure how I could easily use it to:

• communicate performance improvements to others (e.g. in a CL description, on a performance dashboard, etc.), or
• quickly compare multiple optimization ideas to see which one yields the best results, or
• do multiple runs and process results with benchstat

To clarify, I'm proposing adding, to benchmark output, columns with rusage diffs for requested resources. That is, we would have, for example, user time ("utime/op") column next to "ns/op".