pprof
pprof 是用于可视化和分析性能分析数据的工具;
- runtime/pprof：采集程序（非 Server）的运行数据进行分析
- net/http/pprof：采集 HTTP Server 的运行时数据进行分析
启用方式
在main函数之前使用启动, DoProfile(6060)
import (
"net/http"
_ "net/http/pprof"
"strconv"
)
type ProfileServer struct {
}
func (this *ProfileServer) DoProfile(port int) {
go func() {
err := http.ListenAndServe(":"+strconv.FormatInt(int64(port), 10), nil)
if err != nil {
log.Errorf("Failed to do profile on port: %d", port)
} else {
log.Infof("pprof start successfully on port %d", port)
}
}()
}
分析
curl 'http://127.0.0.1:6060/debug/pprof/profile' -o profile.20240223
curl 'http://127.0.0.1:6060/debug/pprof/heap' -o heap.20240223
curl 'http://127.0.0.1:6060/debug/pprof/block' -o block.20240223
curl 'http://127.0.0.1:6060/debug/pprof/mutex' -o mutex.20240223
curl 'http://127.0.0.1:6060/debug/pprof/trace' -o trace.20240223
输出对应的pprof文件之后, 对文件进行分析;
- Flat：函数自身运行耗时
- Flat%：函数自身耗时比例
- Sum%：指的就是每一行的flat%与上面所有行的flat%总和
- Cum：当前函数加上它所有调用栈的运行总耗时
- Cum%：当前函数加上它所有调用栈的运行总耗时比例
使用命令行分析
输入 help 查看帮助命令, 常有 top 和 list
top 找到占用资源最高的函数, list methodName 找到源码所在;
go tool pprof profile.230904
File: xx_cc
Type: cpu
Time: Feb 23, 2024 at 10:53am (CST)
Duration: 30s, Total samples = 230ms ( 0.77%)
Entering interactive mode (type "help" for commands, "o" for options)
(pprof) top 20
Showing nodes accounting for 300ms, 88.24% of 340ms total
Showing top 20 nodes out of 212
flat flat% sum% cum cum%
70ms 20.59% 20.59% 70ms 20.59% runtime.step
40ms 11.76% 32.35% 40ms 11.76% runtime/internal/syscall.Syscall6
20ms 5.88% 38.24% 20ms 5.88% runtime.unlock2
10ms 2.94% 41.18% 10ms 2.94% encoding/json.simpleLetterEqualFold
10ms 2.94% 44.12% 10ms 2.94% fmt.(*pp).doPrintf
10ms 2.94% 47.06% 10ms 2.94% net/http.readRequest
10ms 2.94% 50.00% 10ms 2.94% runtime.(*moduledata).textAddr
10ms 2.94% 52.94% 10ms 2.94% runtime.casgstatus
10ms 2.94% 55.88% 10ms 2.94% runtime.elideWrapperCalling
10ms 2.94% 58.82% 10ms 2.94% runtime.epollwait
10ms 2.94% 61.76% 10ms 2.94% runtime.findfunc
10ms 2.94% 64.71% 10ms 2.94% runtime.getStackMap
10ms 2.94% 67.65% 30ms 8.82% runtime.gwrite
10ms 2.94% 70.59% 10ms 2.94% runtime.heapBitsSetType
10ms 2.94% 73.53% 10ms 2.94% runtime.lock2
10ms 2.94% 76.47% 20ms 5.88% runtime.mallocgc
10ms 2.94% 79.41% 10ms 2.94% runtime.memmove
10ms 2.94% 82.35% 90ms 26.47% runtime.pcvalue
10ms 2.94% 85.29% 20ms 5.88% runtime.printlock
10ms 2.94% 88.24% 10ms 2.94% runtime.scanblock
(pprof) list doPrintf
Total: 340ms
ROUTINE ======================== fmt.(*pp).doPrintf in /usr/lib/golang/src/fmt/print.go
10ms 10ms (flat, cum) 2.94% of Total
. . 1009: p.reordered = false
. . 1010:formatLoop:
. . 1011: for i := 0; i < end; {
. . 1012: p.goodArgNum = true
. . 1013: lasti := i
10ms 10ms 1014: for i < end && format[i] != '%' {
. . 1015: i++
. . 1016: }
. . 1017: if i > lasti {
. . 1018: p.buf.writeString(format[lasti:i])
. . 1019: }
(pprof)
web 界面查看分析结果
需要先安装 graphviz
brew install graphviz # for macos
apt install graphviz # for ubuntu
yum install graphviz # for centos
界面分析更为方便, 还能看 CPU 火焰图, 更容易找到问题所在; 启动web服务查看profile分析结果:
go tool pprof -http=127.0.0.1:8001 profile.230904
Serving web UI on http://127.0.0.1:8001