我要评论背景
go 必须要 test 才能跑benchmark，导致一些情况下想要在main函数中测试benchmark会麻烦一些，因此我实现了一个简单的且没有开销的benchmark函数，方便使用！其次也方便大家学习下如何实现一个零开销的benchmark框架!
benchamrk 实现
对于有timeout的benchamrk，每次都去比较 time.before(timeout) 开销非常的大，而且 benchmark 的对外函数也不能是一个空函数一定要带 count ，因为函数调用大概会劣化 ns 级别！
所以一般的benchmark算法都是梯度benchmark，即 1，10，100，1000，10000，100000，1000000 ... 的数量级去benchmark，好处就是避免了大量的time.since 计算开销，因为time.since单次在 30ns 左右（linux环境下），开销非常大的！
package pprof
import (
"fmt"
"sync"
"sync/atomic"
"time"
)
func parallelbenchmark(name string, thread int, duration time.duration, execute func(count int)) {
wg := sync.waitgroup{}
wg.add(thread)
totalcount := uint64(0)
totalspend := uint64(0)
for i := 0; i < thread; i++ {
go func() {
defer wg.done()
spend, count := benchmark(duration, execute)
atomic.adduint64(&totalspend, uint64(spend))
atomic.adduint64(&totalcount, uint64(count))
}()
}
wg.wait()
fmt.printf("name=%s thread=%d duration=%s total=%d avg=%s\n", name, thread, duration, totalcount, avg(time.duration(totalspend), int(totalcount)))
}
func avg(spend time.duration, count int) string {
avg := float64(spend) / float64(count)
if avg > 100 {
return time.duration(avg).string()
}
return fmt.sprintf("%.4fns", avg)
}
func benchmark(duration time.duration, bench func(count int)) (time.duration, int) {
const maxtotalcount = 1000000000 // 10e
count := 1
totalspend := time.duration(0)
totalcount := 0
for {
start := time.now()
bench(count)
spend := time.since(start)
totalspend = totalspend + spend
totalcount = totalcount + count
if totalcount >= maxtotalcount {
break
}
subspend := duration - totalspend
if subspend <= 0 {
break
}
count = totalcount*10 - totalcount
if subcount := int(float64(subspend) / (float64(totalspend) / float64(totalcount))); count > subcount {
count = subcount
}
}
return totalspend, totalcount
}
profile 实现
package pprof
import (
"net/http"
_ "net/http/pprof"
"os"
"runtime"
"runtime/pprof"
)
// initpprof
// go initpprof()
func initpprof() {
err := http.listenandserve(":12345", http.defaultservemux)
if err != nil {
panic(err)
}
}
func startcpuprofile(filename string) (stop func()) {
f, err := os.create(filename)
if err != nil {
panic(err)
}
if err := pprof.startcpuprofile(f); err != nil {
if err := f.close(); err != nil {
panic(err)
}
panic(err)
}
return func() {
pprof.stopcpuprofile()
if err := f.close(); err != nil {
panic(err)
}
}
}
func startmemprofile(filename string) (stop func()) {
f, err := os.create(filename)
if err != nil {
panic(err)
}
return func() {
defer func() {
if err := f.close(); err != nil {
panic(err)
}
}()
runtime.gc() // get up-to-date statistics
if err := pprof.writeheapprofile(f); err != nil {
panic(err)
}
}
}
例子
package main
import (
"github.com/anthony-dong/golang/pkg/pprof"
"sync"
"time"
)
func main() {
// 记录 cup pprof
//stop := pprof.startcpuprofile("cpu.out")
//defer stop()
// 并发测试 sync map的性能
mm := sync.map{}
pprof.parallelbenchmark("test1", 64, time.second, func(count int) {
for i := 0; i < count; i++ {
mm.store(i%10000, 1)
}
})
// name=test1 thread=32 duration=1s total=6708009 avg=4.772µs
// name=test1 thread=64 duration=1s total=6883456 avg=9.3µs
}到此这篇关于go语言如何实现benchmark函数的文章就介绍到这了,更多相关go benchmark内容请搜索代码网以前的文章或继续浏览下面的相关文章希望大家以后多多支持代码网！
发表评论