vmprof.disable() stops sampling, then does an expensive scan of code objects and debug info to symbolize addresses, then fsync()'s the profile, and only then records the end timestamp. In my experience this results in an end timestamp that's between 80 and 500 milliseconds after the sampling actually stopped, which interferes with efforts to match up vmprof-recorded stacks against other sources of profiling information.
Could the end time recorded in the profile be made to more closely match the time at which sampling last stopped?
vmprof.disable()
stops sampling, then does an expensive scan of code objects and debug info to symbolize addresses, then fsync()'s the profile, and only then records the end timestamp. In my experience this results in an end timestamp that's between 80 and 500 milliseconds after the sampling actually stopped, which interferes with efforts to match up vmprof-recorded stacks against other sources of profiling information.Could the end time recorded in the profile be made to more closely match the time at which sampling last stopped?