Problem with etcd profiling

[alexanius]

Hello. I'm trying to get the pprof of the etcd-3.5.11, but it gives strange results. What do I do:

$ ./server --enable-pprof=true

$ ../tools/benchmark/benchmark --conns=1 --clients=1 range "11111111" --consistency=l --total=10000

$ go tool pprof http://localhost:2379/debug/pprof/profile
(pprof) top
Showing nodes accounting for 100ms, 100% of 100ms total
Showing top 10 nodes out of 14
      flat  flat%   sum%        cum   cum%
      50ms 50.00% 50.00%       50ms 50.00%  runtime/internal/syscall.Syscall6
      20ms 20.00% 70.00%       30ms 30.00%  runtime.stealWork
      10ms 10.00% 80.00%       10ms 10.00%  go.etcd.io/etcd/raft/v3.(*node).Ready
      10ms 10.00% 90.00%       10ms 10.00%  runtime.runqsteal
      10ms 10.00%   100%       10ms 10.00%  runtime.write1
         0     0%   100%       10ms 10.00%  go.etcd.io/etcd/server/v3/etcdserver.(*raftNode).start.func1
         0     0%   100%       90ms 90.00%  runtime.findRunnable
         0     0%   100%       90ms 90.00%  runtime.mcall
         0     0%   100%       50ms 50.00%  runtime.netpoll
         0     0%   100%       10ms 10.00%  runtime.netpollBreak

We can see, that the counters do not show what etcd actually did. If I use the perf, I get the following result:

   6,18%  server   server                [.] runtime.mallocgc
   5,02%  server   server                [.] runtime.greyobject
   4,21%  server   server                [.] runtime.stealWork
   3,97%  server   server                [.] runtime.selectgo
   3,46%  server   server                [.] runtime.scanobject
   3,15%  server   server                [.] runtime.step
   2,58%  server   server                [.] runtime.pcvalue
   2,38%  server   server                [.] runtime.findObject
   2,31%  server   [vdso]                [.] __vdso_clock_gettime
   2,12%  server   server                [.] runtime.lock2

What is the problem?