On Sun, Jul 17, 2016 at 11:00 AM, ajay aggarwal <ajaysa...@gmail.com> wrote: > > I am trying to analyze why I am getting low throughput for my app. Below is > the 30 second CPU profile output. I would expect more time spent in net/http > serve. Not sure if runtime.mcall and runtime.schedule taking almost 1/4th of > the time is concerning. When I did a "web mcall" it didn't show who is > calling mcall. The web graph started with mcall. Any help will be much > appreciated. Thanks! > > (pprof) top 5 -cum > > 0.09s of 32s total ( 0.28%) > > Dropped 449 nodes (cum <= 0.16s) > > Showing top 5 nodes out of 249 (cum >= 7.49s) > > flat flat% sum% cum cum% > > 0 0% 0% 22.17s 69.28% runtime.goexit > > 0.01s 0.031% 0.031% 8.73s 27.28% net/http.(*conn).serve > > 0 0% 0.031% 7.77s 24.28% runtime.mcall > > 0.07s 0.22% 0.25% 7.62s 23.81% runtime.schedule > > 0.01s 0.031% 0.28% 7.49s 23.41% net/http.serverHandler.ServeHTTP
This is misleading. runtime.mcall and runtime.schedule tend to appear in many backtraces because of how the scheduler works, and that causes them to appear high up in the cumulative profile. It doesn't mean anything, it just means that the goroutine scheduler makes much of the work appear to be called from mcall and schedule. In a cumulative profile you need to look farther down into your program's code to see where the work is happening. Ian -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
