go 可以使用 benchmark 來做一些效能測試，需要調整程式的效能時就會需要用到．
準備要測試的程式，是一個用來產生測試資料的一段程式．所以要測試的是下面三種寫法哪個效能會比較好．
GenTestData1 沒有使用 goroutines．
GenTestData2 是使用了 goroutines，但產生亂數那一行並沒有加到 goroutines 去跑．
GenTestData2 是將 for 迴圈裡面所有的工作都放到 goroutines 去跑．
三個 function 都會先產生檔案，並把測試資料寫入各自的檔案．
package test
import (
"encoding/csv"
"os"
"sync"
ran "util/gendata"
)
func GenTestData1(dataCount int) {
csvFile, _ := os.Create("./test1.csv")
w := csv.NewWriter(csvFile)
for i := 0; i < dataCount; i++ {
func(i int) {
startTime := ran.GenStartTime()
dataLine := []string{ran.GenId(), ran.GenFullURL(), startTime, ran.GenEndTime(startTime), ran.GenDownVolumn(), ran.GenUpVolumn()}
w.Write(dataLine)
}(i)
}
w.Flush()
}
func GenTestData2(dataCount int) {
csvFile, _ := os.Create("./test2.csv")
w := csv.NewWriter(csvFile)
var wg sync.WaitGroup
var m sync.Mutex
wg.Add(dataCount)
for i := 0; i < dataCount; i++ {
startTime := ran.GenStartTime()
dataLine := []string{ran.GenId(), ran.GenFullURL(), startTime, ran.GenEndTime(startTime), ran.GenDownVolumn(), ran.GenUpVolumn()}
go func(i int) {
defer wg.Done()
m.Lock()
w.Write(dataLine)
m.Unlock()
}(i)
}
wg.Wait()
w.Flush()
}
func GenTestData3(dataCount int) {
csvFile, _ := os.Create("./test3.csv")
w := csv.NewWriter(csvFile)
var wg sync.WaitGroup
var m sync.Mutex
wg.Add(dataCount)
for i := 0; i < dataCount; i++ {
go func(i int) {
defer wg.Done()
startTime := ran.GenStartTime()
dataLine := []string{ran.GenId(), ran.GenFullURL(), startTime, ran.GenEndTime(startTime), ran.GenDownVolumn(), ran.GenUpVolumn()}
m.Lock()
w.Write(dataLine)
m.Unlock()
}(i)
}
wg.Wait()
w.Flush()
}
接著跟 testing 一樣寫一隻 example_test.go，如果是要單元測試的話 function 名稱是 Test 開頭．
但要效能測試的 function 名稱是要 Benchmark 開頭．根據要比較的三個 function 各自寫一個測試 function．
package test
import (
"testing"
)
func BenchmarkGenTestData1(b *testing.B) {
for i := 0; i < b.N; i++ {
GenTestData1(10000)
}
}
func BenchmarkGenTestData2(b *testing.B) {
for i := 0; i < b.N; i++ {
GenTestData2(10000)
}
}
func BenchmarkGenTestData3(b *testing.B) {
for i := 0; i < b.N; i++ {
GenTestData3(10000)
}
}
下指令執行 benchmark 跑效能測試，在 go test 後加上 -bench=.，最後一個 . 是代表當前 package．
> go test -bench=. .
goos: darwin
goarch: amd64
pkg: practice/test
BenchmarkGenTestData1-8 2 815488165 ns/op
BenchmarkGenTestData2-8 2 782245901 ns/op
BenchmarkGenTestData3-8 5 279053275 ns/op
PASS
ok practice/test 8.018s
BenchmarkGenTestData1-8 的 -8 代表目前 CPU 的核心數．
2 跟 5 是指 1 秒鐘可以跑 2 次及 5 次．而每一次需要約 77779208 ns．
根據上面的數字來看 GenTestData3 的程式寫法效能是最好的．因為它 1 秒鐘可以跑 5 次，另外兩個都只能跑 2 次．而 GenTestData1 與 GenTestData2 的執行時間差不多，代表效能瓶頸會在產生測試資料的部分
startTime := ran.GenStartTime()
ådataLine := []string{ran.GenId(), ran.GenFullURL(), startTime, ran.GenEndTime(startTime), ran.GenDownVolumn(), ran.GenUpVolumn()}
產生 cpu、memory 及 blocking 的效能分析檔案
go test -v -bench=. -cpuprofile=cpu.out -memprofile=mem.out -blockprofile=block.out .
goos: darwin
goarch: amd64
pkg: practice/test
BenchmarkGenTestData1-8 2 740294511 ns/op
BenchmarkGenTestData2-8 2 794994639 ns/op
BenchmarkGenTestData3-8 5 269349969 ns/op
PASS
ok practice/test 8.155s
產生的檔案如下，test.test，檔名 test 是測試的 package 名稱．
> ll
total 14592
-rwxrwxrwx 1 daniel staff 2.0K Oct 27 23:21 block.out
-rwxrwxrwx 1 daniel staff 17K Oct 27 23:21 cpu.out
-rwxrwxrwx 1 daniel staff 1.4K Oct 27 22:10 example.go
-rwxrwxrwx 1 daniel staff 331B Oct 27 22:00 example_test.go
-rwxrwxrwx 1 daniel staff 3.7K Oct 27 23:21 mem.out
-rwxrwxrwx 1 daniel staff 3.5M Oct 27 23:21 test.test
-rwxrwxrwx 1 daniel staff 995K Oct 27 23:21 test1.csv
-rwxrwxrwx 1 daniel staff 995K Oct 27 23:21 test2.csv
-rwxrwxrwx 1 daniel staff 995K Oct 27 23:21 test3.csv
使用 go tool pprof 來看分析檔案，會根據消耗最多的來排序．cpu 的分析檔案
> go tool pprof -text test.test cpu.out
File: test.test
Type: cpu
Time: Oct 27, 2018 at 11:21pm (CST)
Duration: 8.01s, Total samples = 19.90s (248.58%)
Showing nodes accounting for 18.86s, 94.77% of 19.90s total
Dropped 141 nodes (cum <= 0.10s)
flat flat% sum% cum cum%
6.60s 33.17% 33.17% 6.60s 33.17% runtime.usleep
2.86s 14.37% 47.54% 2.86s 14.37% runtime.pthread_cond_signal
2.69s 13.52% 61.06% 2.69s 13.52% runtime.pthread_cond_wait
1.91s 9.60% 70.65% 1.91s 9.60% syscall.Syscall
0.85s 4.27% 74.92% 0.85s 4.27% math/rand.seedrand (inline)
0.64s 3.22% 78.14% 1.49s 7.49% math/rand.(*rngSource).Seed
0.62s 3.12% 81.26% 0.62s 3.12% runtime.pthread_cond_timedwait_relative_np
0.55s 2.76% 84.02% 0.55s 2.76% runtime.memclrNoHeapPointers
0.28s 1.41% 85.43% 0.28s 1.41% runtime.kevent
0.27s 1.36% 86.78% 0.32s 1.61% runtime.scanobject
...
memoery 的分析檔案
> go tool pprof -text test.test mem.out
File: test.test
Type: alloc_space
Time: Oct 27, 2018 at 11:21pm (CST)
Showing nodes accounting for 5190.38MB, 99.11% of 5236.92MB total
Dropped 46 nodes (cum <= 26.18MB)
flat flat% sum% cum cum%
5166.88MB 98.66% 98.66% 5166.88MB 98.66% math/rand.NewSource
8.50MB 0.16% 98.82% 4310.96MB 82.32% util/gendata.stringWithCharset
6MB 0.11% 98.94% 883.42MB 16.87% util/gendata.GenId
4.50MB 0.086% 99.03% 925.15MB 17.67% practice/test.GenTestData2
4.50MB 0.086% 99.11% 4315.46MB 82.40% util/gendata.GenFullURL
0 0% 99.11% 934.74MB 17.85% practice/test.BenchmarkGenTestData1
0 0% 99.11% 925.15MB 17.67% practice/test.BenchmarkGenTestData2
...
blocking 的分析檔案
> go tool pprof -text test.test block.out
File: test.test
Type: delay
Time: Oct 27, 2018 at 11:21pm (CST)
Showing nodes accounting for 49.21mins, 99.67% of 49.37mins total
Dropped 36 nodes (cum <= 0.25mins)
flat flat% sum% cum cum%
49.21mins 99.67% 99.67% 49.21mins 99.67% sync.(*Mutex).Lock
0 0% 99.67% 0.50mins 1.01% fmt.Sprintf
0 0% 99.67% 0.41mins 0.83% fmt.newPrinter
0 0% 99.67% 0.50mins 1.01% github.com/Pallinder/go-randomdata.StringNumber
0 0% 99.67% 0.50mins 1.01% github.com/Pallinder/go-randomdata.StringNumberExt
0 0% 99.67% 49.21mins 99.67% practice/test.GenTestData3.func1
0 0% 99.67% 0.41mins 0.83% sync.(*Pool).Get
0 0% 99.67% 0.38mins 0.76% sync.(*Pool).getSlow
0 0% 99.67% 0.45mins 0.91% util/gendata.GenDownVolumn
iThome鐵人賽
看影片追技術