expected number of pprof 'total samples'

[Xiangdong Ji]

Hi

I wonder if the expected number of pprof 'total samples' could be calculated by:

                duration * sample_rate (100HZ) * CPU usage (by 'top')

say if my program's CPU usage reported by 'top' is 200%, and sampling period is 30 seconds, could I expect the total samples of pprof is approximately:

                  30 * 100 * 2  => 6K ?

The formula works well on some physical machines/docker containers, but not all of them, sometimes the samples made by pprof is only 50~60% of the expected number, that ratio is stable on a specific machine/container.  Curious to know what could be the rationale behind it?

Thanks.