如何理解 golang pprof 的 cum ?感觉解释不通啊

2019-01-15 11:51:05 +08:00
 hheedat

https://www.reddit.com/r/golang/comments/7ony5f/what_is_the_meaning_of_flat_and_cum_in_golang/

我看了一些资料,都是如这个帖子里面所说,cum ( cumulative )是本函数以及本函数调用的子函数们的时间和。 如帖子里面的调用方式我写了一个 demo,结果符合预期


func ctA() int64 {
	return ctB()
}

func ctB() int64 {
	var x int64 = 0
	for i := 0; i < 1e9; i++ {
		x = x + 1
	}
	return x + ctC() + ctD()
}

func ctC() int64 {
	var x int64 = 0
	for i := 0; i < 1e8; i++ {
		x = x + 1
	}
	return x
}

func ctD() int64 {
	var x int64 = 0
	for i := 0; i < 5e8; i++ {
		x = x + 1
	}
	return x
}
(pprof) top
Showing nodes accounting for 48.68s, 99.43% of 48.96s total
Dropped 11 nodes (cum <= 0.24s)
      flat  flat%   sum%        cum   cum%
    30.31s 61.91% 61.91%     48.68s 99.43%  main.ctB
    15.35s 31.35% 93.26%     15.35s 31.35%  main.ctD
     3.02s  6.17% 99.43%      3.02s  6.17%  main.ctC
         0     0% 99.43%     48.68s 99.43%  main.ctA
         0     0% 99.43%     48.68s 99.43%  main.main
         0     0% 99.43%     48.68s 99.43%  runtime.main
         0     0% 99.43%      0.27s  0.55%  runtime.mstart
         0     0% 99.43%      0.27s  0.55%  runtime.mstart1
         0     0% 99.43%      0.27s  0.55%  runtime.sysmon

但是稍微复杂一些的例子,里面的 main.main,cum 的值就很小了(如下),这就说不通了啊,main 不是一直在函数的调用栈上么?理论上应该是最长的呀。

(pprof) top30
Showing nodes accounting for 15.82s, 95.36% of 16.59s total
Dropped 92 nodes (cum <= 0.08s)
Showing top 30 nodes out of 67
      flat  flat%   sum%        cum   cum%
     6.69s 40.33% 40.33%      6.69s 40.33%  runtime.pthread_cond_wait
     4.61s 27.79% 68.11%      4.61s 27.79%  runtime.pthread_cond_signal
     1.35s  8.14% 76.25%      1.35s  8.14%  runtime.memmove
     0.84s  5.06% 81.31%      0.84s  5.06%  runtime.usleep
     0.51s  3.07% 84.39%      0.52s  3.13%  runtime.findnull
     0.49s  2.95% 87.34%      0.49s  2.95%  runtime.nanotime
     0.44s  2.65% 89.99%      0.44s  2.65%  runtime.(*semaRoot).dequeue
     0.34s  2.05% 92.04%      0.34s  2.05%  runtime.memclrNoHeapPointers
     0.20s  1.21% 93.25%      0.33s  1.99%  runtime.scanobject
     0.09s  0.54% 93.79%      0.09s  0.54%  runtime.pthread_mutex_lock
     0.05s   0.3% 94.09%      0.13s  0.78%  runtime.gentraceback
     0.03s  0.18% 94.27%      0.35s  2.11%  sync.(*Mutex).Lock
     0.03s  0.18% 94.45%      0.38s  2.29%  sync.(*RWMutex).Lock
     0.02s  0.12% 94.58%      0.14s  0.84%  bufio.(*Scanner).Scan
     0.02s  0.12% 94.70%      0.75s  4.52%  runtime.scang
     0.02s  0.12% 94.82%      8.40s 50.63%  runtime.schedule
     0.01s  0.06% 94.88%      0.19s  1.15%  main.Loop
     0.01s  0.06% 94.94%      0.14s  0.84%  runtime.callers
     0.01s  0.06% 95.00%      7.07s 42.62%  runtime.findrunnable
     0.01s  0.06% 95.06%      1.20s  7.23%  runtime.gcDrain
     0.01s  0.06% 95.12%      0.11s  0.66%  runtime.mallocgc
     0.01s  0.06% 95.18%      3.04s 18.32%  runtime.newproc.func1
     0.01s  0.06% 95.24%      3.03s 18.26%  runtime.newproc1
     0.01s  0.06% 95.30%      0.56s  3.38%  runtime.osyield
     0.01s  0.06% 95.36%      4.92s 29.66%  runtime.systemstack
         0     0% 95.36%      0.18s  1.08%  main.(*TrieNode).addWord
         0     0% 95.36%      0.52s  3.13%  main.BuildTire
         0     0% 95.36%      2.45s 14.77%  main.Loop.func1
         0     0% 95.36%      0.53s  3.19%  main.main
         0     0% 95.36%      0.10s   0.6%  runtime.(*gcWork).balance

如何理解这件事情呢?

3419 次点击
所在节点    Go 编程语言
4 条回复
ensonmj
2019-01-15 13:32:13 +08:00
说明你的主要任务都是在其他 goroutine 里面跑的,main 只是等待?
reus
2019-01-15 13:44:47 +08:00
runtime 的一些 goroutine 在 main 执行前就会开启了
用户代码也可能在 main 前执行一些东西,例如 init 函数,或者其他初始化
main 并不是 go 程序的入口
janxin
2019-01-15 13:59:51 +08:00
实际上你写程序的 main 是跟 C 程序中的 Entry Point 的理解不一样,即便是 VC 之类的程序也是在 main 之前有一段小的初始化代码在执行,完成之后才会跳到 main 中。你可以提供一下 demo 代码,没看到不知道你测试了什么呀...
hheedat
2019-01-15 15:59:46 +08:00
@ensonmj 真相了,我把带 goroutine 的那个 LOOP 方法去掉,就符合预期了

@janxin 下面是我的代码
```
package main

import (
"flag"
"os"
"runtime/pprof"
"log"
"fmt"
"runtime"
"bufio"
"sync"
)

var cpuprofile = flag.String("cpuprofile", "", "write cpu profile to `file`")
var memprofile = flag.String("memprofile", "", "write memory profile to `file`")
var blockprofile = flag.String("blockprofile", "", "write block profile to `file`")
var mutexprofile = flag.String("mutexprofile", "", "write mutex profile to `file`")

func main() {
flag.Parse()
if *cpuprofile != "" {
f, err := os.Create(*cpuprofile)
if err != nil {
log.Fatal("could not create CPU profile: ", err)
}
if err := pprof.StartCPUProfile(f); err != nil {
log.Fatal("could not start CPU profile: ", err)
}
defer pprof.StopCPUProfile()
}
if *blockprofile != "" {
runtime.SetBlockProfileRate(10 * 1000)
defer writeProfTo("block", *blockprofile)
}
if *mutexprofile != "" {
runtime.SetMutexProfileFraction(2)
defer writeProfTo("mutex", *mutexprofile)
}

m := make(map[int]*TrieNode, 4)
for i := 0; i < 3; i++ {
m[i] = BuildTire()
//loop.Loop()
}
//for i := 1; i < 10; i++ {
// ctA()
//}

if *memprofile != "" {
f, err := os.Create(*memprofile)
if err != nil {
log.Fatal("could not create memory profile: ", err)
}
//runtime.GC() // get up-to-date statistics
if err := pprof.WriteHeapProfile(f); err != nil {
log.Fatal("could not write memory profile: ", err)
}
f.Close()
}
}

func ctA() int64 {
return ctB()
}

func ctB() int64 {
var x int64 = 0
for i := 0; i < 1e9; i++ {
x = x + 1
}
return x + ctC() + ctD()
}

func ctC() int64 {
var x int64 = 0
for i := 0; i < 1e8; i++ {
x = x + 1
}
return x
}

func ctD() int64 {
var x int64 = 0
for i := 0; i < 5e8; i++ {
x = x + 1
}
return x
}

func writeProfTo(name, fn string) {
p := pprof.Lookup(name)
if p == nil {
fmt.Errorf("%s prof not found", name)
return
}
f, err := os.Create(fn)
if err != nil {
fmt.Errorf("%v", err.Error())
return
}
defer f.Close()
err = p.WriteTo(f, 0)
if err != nil {
fmt.Errorf("%v", err.Error())
return
}
}

type TrieNode struct {
Child *map[string]TrieNode
Exist bool
}

var mySlice []int
var sliceLock sync.RWMutex

func BuildTire() *TrieNode {
root := TrieNode{nil, false}

inFile, _ := os.Open("/Users/admin/stopwords.txt")
defer inFile.Close()
scanner := bufio.NewScanner(inFile)
scanner.Split(bufio.ScanLines)

for scanner.Scan() {
lineStr := scanner.Text()
root.addWord(lineStr)
Loop()
}

return &root
}

func Loop() {
var wg sync.WaitGroup
for i := 0; i < 10; i++ {
wg.Add(1)
go func() {
for i := 0; i < 10; i++ {
sliceLock.Lock()
mySlice = append(mySlice, i)
sliceLock.Unlock()
}
wg.Done()
}()
}
wg.Wait()
}

func GetMySliceLen() int {
return len(mySlice)
}

func (n *TrieNode) addWord(words string) {

runes := []rune(words)
keyStr := string(runes[0])

var exist bool
var restStr string
if len(runes) == 1 {
exist = true
restStr = ""
} else {
exist = false
restStr = string(runes[1:])
}

if n.Child == nil {
tm := make(map[string]TrieNode)
n.Child = &tm
}

tmpMap := *n.Child

if _, ok := tmpMap[keyStr]; !ok {
tmpMap[keyStr] = TrieNode{nil, exist}
} else {
if exist {
//tmpMap[keyStr].Exist = exist
tm := tmpMap[keyStr]
tm.Exist = exist
tmpMap[keyStr] = tm
}
}

n.Child = &tmpMap

if len(restStr) > 0 {
tm := tmpMap[keyStr]
tm.addWord(restStr)
tmpMap[keyStr] = tm
}
}

```

这是一个专为移动设备优化的页面(即为了让你能够在 Google 搜索结果里秒开这个页面),如果你希望参与 V2EX 社区的讨论,你可以继续到 V2EX 上打开本讨论主题的完整版本。

https://www.v2ex.com/t/527178

V2EX 是创意工作者们的社区,是一个分享自己正在做的有趣事物、交流想法,可以遇见新朋友甚至新机会的地方。

V2EX is a community of developers, designers and creative people.

© 2021 V2EX