Go语言性能调试方法
Go 语言内置了性能调优的工具
一、准备工作
- 安装 graphviz
brew install graphviz
- 将 $ GOPATH/bin 加入 $PATH
- Mac OS：在 .bash_profile 中修改路径
- 安装 go-torch（火炬图）
- go get github.com/uber/go-torch
- 下载并复制 flamegraph.pl (https://github.com/brendangregg/FlameGraph) 至 $GOPATH/bin 路径下
- 将 $GOPATH/bin 加入 $PATH
二、性能调优步骤和方法
示例代码：https://github.com/shiiiiyd/csgo/blob/main/tools/file/prof.go
1、build 程序
首先使用 go build prof.go 命令构建prof.go 文件，然后运行构建好的prof，使用 ls 命令可以查看到当前目录下会多出几个文件，包含：cpu.prof、goroutine.prof、mem.prof文件
2、使用pprof命令查看性能
查看CPU使用情况：
csgo/tools/file on main [!+?] via 🐹 v1.17.12 took 2s
❯ ls
cpu.prof goroutine.prof mem.prof prof prof.go
csgo/tools/file on main [!+?] via 🐹 v1.17.12
❯ go tool pprof prof cpu.prof ## 进入pprof命令
File: prof
Type: cpu
Time: Jul 28, 2022 at 12:27pm (CST)
Duration: 907.52ms, Total samples = 760ms (83.74%)
Entering interactive mode (type "help" for commands, "o" for options)
(pprof)
(pprof) top ## top 命令查看 cpu 性能
Showing nodes accounting for 760ms, 100% of 760ms total
Showing top 10 nodes out of 21
flat flat% sum% cum cum%
720ms 94.74% 94.74% 730ms 96.05% main.fillMatrix
20ms 2.63% 97.37% 20ms 2.63% main.calculate (inline)
10ms 1.32% 98.68% 10ms 1.32% math/rand.(*rngSource).Int63
10ms 1.32% 100% 10ms 1.32% runtime.(*spanSet).pop
0 0% 100% 760ms 100% main.main
0 0% 100% 10ms 1.32% math/rand.(*Rand).Int31 (inline)
0 0% 100% 10ms 1.32% math/rand.(*Rand).Int31n
0 0% 100% 10ms 1.32% math/rand.(*Rand).Int63 (inline)
0 0% 100% 10ms 1.32% math/rand.(*Rand).Intn
0 0% 100% 10ms 1.32% os.Create (inline)
(pprof)
(pprof)
(pprof) list fillMatrix ## 查看fillMatrix具体耗费时间的代码片段
Total: 760ms
ROUTINE ======================== main.fillMatrix in /Users/shiyd/go/src/github.com/csgo/tools/file/prof.go
720ms 730ms (flat, cum) 96.05% of Total
. . 19:func fillMatrix(m *[row][col]int) {
. . 20: s := rand.New(rand.NewSource(time.Now().UnixNano()))
. . 21:
. . 22: for i := 0; i < row; i++ {
. . 23: for j := 0; j < col; j++ {
720ms 730ms 24: m[i][j] = s.Intn(100000)
. . 25: }
. . 26: }
. . 27:}
. . 28:
. . 29:func calculate(m *[row][col]int) {
(pprof)
(pprof)
生成SVG图
## svg 命令是显示 graphviz 输出的svg文件，该文件是可以查看代码耗费的时间
(pprof) svg
Generating report in profile001.svg
(pprof)exit
查看内存占用（mem）情况：
csgo/tools/file on main [!+?] via 🐹 v1.17.12
❯ go tool pprof prof mem.prof
File: prof
Type: inuse_space
Time: Jul 28, 2022 at 12:27pm (CST)
Entering interactive mode (type "help" for commands, "o" for options)
(pprof)
(pprof) top
Showing nodes accounting for 3233.64kB, 100% of 3233.64kB total
Showing top 10 nodes out of 19
flat flat% sum% cum cum%
1184.27kB 36.62% 36.62% 1184.27kB 36.62% runtime/pprof.StartCPUProfile
1025.12kB 31.70% 68.33% 1025.12kB 31.70% runtime.allocm
512.20kB 15.84% 84.17% 512.20kB 15.84% runtime.malg
512.04kB 15.83% 100% 512.04kB 15.83% runtime.bgscavenge
0 0% 100% 1184.27kB 36.62% main.main
0 0% 100% 1184.27kB 36.62% runtime.main
0 0% 100% 512.56kB 15.85% runtime.mcall
0 0% 100% 512.56kB 15.85% runtime.mstart
0 0% 100% 512.56kB 15.85% runtime.mstart0
0 0% 100% 512.56kB 15.85% runtime.mstart1
(pprof)
(pprof) list main.main
Total: 3.16MB
ROUTINE ======================== main.main in /Users/shiyd/go/src/github.com/csgo/tools/file/prof.go
0 1.16MB (flat, cum) 36.62% of Total
. . 41: if err != nil {
. . 42: log.Fatal("could not create CPU profile: ", err)
. . 43: }
. . 44:
. . 45: // 获取系统信息
. 1.16MB 46: if err := pprof.StartCPUProfile(f); err != nil { //监控cpu
. . 47: log.Fatal("could not start CPU profile: ", err)
. . 48: }
. . 49: defer pprof.StopCPUProfile()
. . 50:
. . 51: // 主逻辑区，进行一些简单的代码运算
(pprof)
(pprof)
3、使用 flame graphs 查看
go-torch 命令
使用go-torch cpu.prof 命令生成火焰图
❯ go-torch cpu.prof
INFO[10:03:05] Run pprof command: go tool pprof -raw -seconds 30 cpu.prof
INFO[10:03:05] Writing svg to torch.svg
三、通过HTTP方式输出Profile
上面的方式适用于非运行时的服务，如果是运行时的服务则可以通过HTTP访问的方式输出Profile 文件查看。
-
简单，适用于持续性运行的应用
-
在应用程序中导入 import _ "net/http/pprof"，并启动 http server 即可
- http调用：
import _ "net/http/pprof" - Gin 调用：
import "github.com/DeanThompson/ginpprof"r := gin.Default() ginpprof.Wrap(r) - 示例：go-torch -u http://localhost:8080/ -t 30
- http调用：
-
测试命令：
-
http://<host>:<port>/debug/pprof/-
示例：http://127.0.0.1:8081/debug/pprof
又可能不会出现包含block、mutex、heap的界面，需要重启一下服务
-
-
go tool pprof http://<host>:<port>/debug/pprof/profile?seconds=10(默认为30秒)- 示例：
go tool pprof http://127.0.0.1:8080/debug/pprof/profile -t 10
- 示例：
-
go-torch -seconds 10 http://<host>:<port>/debug/pprof/profile
-
参考
[1]：https://pkg.go.dev/github.com/uber/go-torch@v0.0.0-20181107071353-86f327cc820e#section-readme
[2]：https://time.geekbang.org/course/detail/100024001-91253
[3]：https://ruanyifeng.com/blog/2017/09/flame-graph.html
[4]：https://blog.csdn.net/qq_17612199/article/details/80353355
© shiyd.RSS