1.安装易用的pprof

golang自带的prof包是runtime/pprof,这个是低级别的,需要你手动做一些设置等等周边工作,不利于我们快速上手,利用pprof帮助我们解决实际的问题。这里推荐davecheney封装的pprof,它可以1行代码,让你用上pprof,专心解决自己的代码问题,下载:

1
go get github.com/pkg/profile
2.安装graphviz

pprof生成的prof文件时二进制的,需要把这个二进制的文件转换为我们人类可读的,graphviz可以帮助我们把二进制的prof文件转换为图像。Mac安装:

1
brew install graphviz
3.修改main函数

只需要为hi.go增加这一行,defer profile.Start().Stop(),程序运行时,默认就会记录cpu数据

4.编译运行你的函数
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
package main

import (
"fmt"
"github.com/pkg/profile"
)

func main() {
defer profile.Start().Stop()
sl:=makeSlice()
fmt.Printf("sum=%d\n",sumSlice(sl))
}

func makeSlice() []int {
sl := make([]int, 10000000)
for idx := range sl {
sl[idx] = idx
}
return sl
}

func sumSlice(sl []int) int {
sum := 0
for _, x := range sl {
sum += x
}
return sum
}

运行代码

1
2
go build hi.go
./hi

应当看到类似的结果,它输出了生成的cpu.pprof的路径:

1
2
3
2019/02/26 23:24:22 profile: cpu profiling enabled, /var/folders/zy/d8yskqjj09q21y7jnc0vbsqh0000gn/T/profile152894194/cpu.pprof  
sum=49999995000000
2019/02/26 23:24:22 profile: cpu profiling disabled, /var/folders/zy/d8yskqjj09q21y7jnc0vbsqh0000gn/T/profile152894194/cpu.pprof
5.可视化prof

可视化有多种方式,可以转换为text、pdf、svg等等。text命令是

1
go tool pprof --text /path/to/yourbinary /var/path/to/cpu.pprof
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
go tool pprof --text ./hi /var/folders/zy/d8yskqjj09q21y7jnc0vbsqh0000gn/T/profile626723859/cpu.pprof
File: hi
Type: cpu
Time: Feb 26, 2019 at 11:29pm (CST)
Duration: 203.53ms, Total samples = 100ms (49.13%)
Showing nodes accounting for 100ms, 100% of 100ms total
flat flat% sum% cum cum%
60ms 60.00% 60.00% 60ms 60.00% runtime.usleep
30ms 30.00% 90.00% 30ms 30.00% main.makeSlice
10ms 10.00% 100% 10ms 10.00% runtime.nanotime
0 0% 100% 30ms 30.00% main.main
0 0% 100% 10ms 10.00% runtime.gcBgMarkWorker
0 0% 100% 70ms 70.00% runtime.gcBgMarkWorker.func2
0 0% 100% 70ms 70.00% runtime.gcDrain
0 0% 100% 30ms 30.00% runtime.main
0 0% 100% 70ms 70.00% runtime.markroot
0 0% 100% 70ms 70.00% runtime.markroot.func1
0 0% 100% 60ms 60.00% runtime.mstart
0 0% 100% 60ms 60.00% runtime.osyield
0 0% 100% 70ms 70.00% runtime.scang
0 0% 100% 70ms 70.00% runtime.systemstack

生成pdf图

1
2
go tool pprof --pdf ./hi /var/folders/zy/d8yskqjj09q21y7jnc0vbsqh0000gn/T/profile626723859/cpu.pprof > cpu.pdf

6.获取内存ppfo

如果你掌握了cpu pprof,mem pprof轻而易举就能拿下,只需要改1行代码:

1
defer profile.Start(profile.MemProfile).Stop()