1. 性能分析工具

graphviz windows安装

graphviz-install-5.0.0-win64.zip 安装完成后，需要把文件夹中的文件拷贝到自己本地的GOPATH目录下。（我的：D:\GoPackage） go get github.com/uber/go-torch

image.png
image.png

go tool pprof .\cpu.prof

  1. Type: cpu
  2. Time: Aug 10, 2022 at 7:48pm (CST)
  3. Duration: 1.32s, Total samples = 750ms (56.65%)
  4. Entering interactive mode (type "help" for commands, "o" for options)

go tool pprof prof .\cpu.prof

  1. prof: open prof: The system cannot find the file specified.
  2. Fetched 1 source profiles out of 2
  3. Type: cpu
  4. Time: Aug 10, 2022 at 7:48pm (CST)
  5. Duration: 1.32s, Total samples = 750ms (56.65%)
  6. Entering interactive mode (type "help" for commands, "o" for options)

top

  1. # flat 函数执行时间及所占比例
  2. Showing top 10 nodes out of 13
  3.       flat  flat%   sum%        cum   cum%
  4.      410ms 54.67% 54.67%      490ms 65.33%  math/rand.(*Rand).Int31n
  5.      200ms 26.67% 81.33%      720ms 96.00%  main.fillMatrix
  6.       60ms  8.00% 89.33%       60ms  8.00%  math/rand.(*rngSource).Uint64 (inline)
  7.       30ms  4.00% 93.33%      520ms 69.33%  math/rand.(*Rand).Intn
  8.       20ms  2.67% 96.00%       20ms  2.67%  main.calculate (inline)
  9.       10ms  1.33% 97.33%       80ms 10.67%  math/rand.(*Rand).Int31 (inline)
  10.       10ms  1.33% 98.67%       70ms  9.33%  math/rand.(*rngSource).Int63
  11.       10ms  1.33%   100%       10ms  1.33%  runtime/pprof.(*profMap).lookup
  12.          0     0%   100%      740ms 98.67%  main.main
  13.          0     0%   100%       70ms  9.33%  math/rand.(*Rand).Int63 (inline)

list fillMatrix

  1. Total: 750ms                                                                                                         
  2. ROUTINE ======================== main.fillMatrix in D:\JetbrainsCode\Learn_Go\go-learning\src\ch40\tools\file\prof.go
  3.      200ms      720ms (flat, cum) 96.00% of Total                                                                    
  4.          .          .     15:)                                                                                       
  5.          .          .     16:                                                                                        
  6.          .          .     17:func fillMatrix(m *[row][col]int) {                                                     
  7.          .          .     18:   s := rand.New(rand.NewSource(time.Now().UnixNano()))                                 
  8.          .          .     19:                                                                                        
  9.       10ms       10ms     20:   for i := 0; i < row; i++ {                                                           
  10.       20ms       20ms     21:           for j := 0; j < col; j++ {
  11.      170ms      690ms     22:                   m[i][j] = s.Intn(100000)
  12.          .          .     23:           }
  13.          .          .     24:   }
  14.          .          .     25:}
  15.          .          .     26:
  16.          .          .     27:func calculate(m *[row][col]int) {

svg 生成svg图片

profile001.svg

go tool pprof prof .\mem.prof top list main.main

  1. Total: 4.22MB
  2. ROUTINE ======================== main.main in D:\JetbrainsCode\Learn_Go\go-learning\src\ch40\tools\file\prof.go
  3.          0     1.72MB (flat, cum) 40.76% of Total
  4.          .          .     39:   if err != nil {
  5.          .          .     40:           log.Fatal("could not create CPU profile: ", err)
  6.          .          .     41:   }
  7.          .          .     42:
  8.          .          .     43:   // 获取系统信息
  9.          .     1.72MB     44:   if err := pprof.StartCPUProfile(f); err != nil { //监控cpu
  10.          .          .     45:           log.Fatal("could not start CPU profile: ", err)
  11.          .          .     46:   }
  12.          .          .     47:   defer pprof.StopCPUProfile()
  13.          .          .     48:
  14.          .          .     49:   // 主逻辑区，进行一些简单的代码运算

profile002.svg

go-torch 试了一个多小时，还是不行，略过…

image.png
image.png

2. 性能调优示例

go test -bench=”.” -cpuprofile=”cpu.prof”

  1. goos: windows
  2. goarch: amd64                                
  3. pkg: go-learning/src/ch41                    
  4. cpu: Intel(R) Core(TM) i5-8265U CPU @ 1.60GHz
  5. BenchmarkProcessRequest-8         183476              6594 ns/op
  6. BenchmarkProcessRequestOld-8       59528             19808 ns/op
  7. PASS
  8. ok      go-learning/src/ch41    2.912s

go tool pprof .\cpu.prof 

  1. Type: cpu                                      
  2. Time: Aug 11, 2022 at 11:13am (CST)            
  3. Duration: 2.85s, Total samples = 2.15s (75.50%)
  4. Entering interactive mode (type "help" for commands, "o" for options)

top

  1. Showing nodes accounting for 900ms, 41.86% of 2150ms total
  2. Dropped 67 nodes (cum <= 10.75ms)
  3. Showing top 10 nodes out of 131
  4.       flat  flat%   sum%        cum   cum%
  5.      130ms  6.05%  6.05%      240ms 11.16%  github.com/mailru/easyjson/jlexer.(*Lexer).FetchToken
  6.      110ms  5.12% 11.16%      350ms 16.28%  runtime.concatstrings
  7.      110ms  5.12% 16.28%      190ms  8.84%  strconv.ParseInt
  8.      100ms  4.65% 20.93%      100ms  4.65%  runtime.memmove
  9.       90ms  4.19% 25.12%      280ms 13.02%  runtime.mallocgc
  10.       80ms  3.72% 28.84%      640ms 29.77%  go-learning/src/ch41.easyjson6a975c40DecodeGoLearningSrcCh411
  11.       80ms  3.72% 32.56%       80ms  3.72%  strconv.ParseUint
  12.       70ms  3.26% 35.81%       70ms  3.26%  runtime.stdcall2
  13.       70ms  3.26% 39.07%       70ms  3.26%  runtime.stdcall3
  14.       60ms  2.79% 41.86%      120ms  5.58%  encoding/json.checkValid

top -cum

  1. Showing nodes accounting for 0.09s, 4.19% of 2.15s total
  2. Dropped 67 nodes (cum <= 0.01s)
  3. Showing top 10 nodes out of 131
  4.       flat  flat%   sum%        cum   cum%
  5.          0     0%     0%      1.59s 73.95%  testing.(*B).launch
  6.          0     0%     0%      1.59s 73.95%  testing.(*B).runN
  7.          0     0%     0%      0.88s 40.93%  go-learning/src/ch41.BenchmarkProcessRequestOld
  8.          0     0%     0%      0.88s 40.93%  go-learning/src/ch41.processRequestOld
  9.          0     0%     0%      0.71s 33.02%  go-learning/src/ch41.BenchmarkProcessRequest
  10.      0.01s  0.47%  0.47%      0.71s 33.02%  go-learning/src/ch41.processRequest
  11.          0     0%  0.47%      0.64s 29.77%  go-learning/src/ch41.(*Request).UnmarshalJSON (partial-inline)
  12.      0.08s  3.72%  4.19%      0.64s 29.77%  go-learning/src/ch41.easyjson6a975c40DecodeGoLearningSrcCh411
  13.          0     0%  4.19%      0.51s 23.72%  runtime.systemstack
  14.          0     0%  4.19%      0.38s 17.67%  encoding/json.Unmarshal

list processRequest

  1. Total: 2.15s                                                                                                                         
  2. ROUTINE ======================== go-learning/src/ch41.processRequest in D:\JetbrainsCode\Learn_Go\go-learning\src\ch41\optmization.go
  3.       10ms      710ms (flat, cum) 33.02% of Total                                                                                    
  4.          .          .     21:                                                                                                        
  5.          .          .     22:func processRequest(reqs []string) []string {                                                           
  6.          .          .     23:   reps := []string{}                                                                                   
  7.          .          .     24:   for _, req := range reqs {
  8.          .          .     25:           reqObj := &Request{}
  9.          .      470ms     26:           reqObj.UnmarshalJSON([]byte(req))
  10.          .          .     27:           //      json.Unmarshal([]byte(req), reqObj)
  11.          .          .     28:
  12.          .          .     29:           var buf strings.Builder
  13.       10ms       10ms     30:           for _, e := range reqObj.PayLoad {
  14.          .      100ms     31:                   buf.WriteString(strconv.Itoa(e))
  15.          .       10ms     32:                   buf.WriteString(",")
  16.          .          .     33:           }
  17.          .          .     34:           repObj := &Response{reqObj.TransactionID, buf.String()}
  18.          .      100ms     35:           repJson, err := repObj.MarshalJSON()
  19.          .          .     36:           //repJson, err := json.Marshal(&repObj)
  20.          .          .     37:           if err != nil {
  21.          .          .     38:                   panic(err)
  22.          .          .     39:           }
  23.          .       20ms     40:           reps = append(reps, string(repJson))
  24.          .          .     41:   }
  25.          .          .     42:   return reps
  26.          .          .     43:}
  27.          .          .     44:
  28.          .          .     45:func processRequestOld(reqs []string) []string {
  29.          .      380ms     49:           json.Unmarshal([]byte(req), reqObj)
  30.          .          .     50:           ret := ""
  31.          .          .     51:           for _, e := range reqObj.PayLoad {
  32.          .      370ms     52:                   ret += strconv.Itoa(e) + ","
  33.          .          .     53:           }
  34.          .          .     54:           repObj := &Response{reqObj.TransactionID, ret}
  35.          .      130ms     55:           repJson, err := json.Marshal(&repObj)
  36.          .          .     56:           if err != nil {
  37.          .          .     57:                   panic(err)
  38.          .          .     58:           }
  39.          .          .     59:           reps = append(reps, string(repJson))
  40.          .          .     60:   }

easyjson -all structs.go go test -bench=”.” -memprofile=”mem.prof”

image.png
image.png

Wall Time 挂钟时间， 你的程序运行的绝对时间和某个函数运行的绝对时间，其中有可能有阻塞。调用外部系统，等待外部响应，这些会影响你的绝对时间 CPU Time CPU消耗时间，包括内存分配，GC次数，GC的耗时
Block Time Memory allocation GC times/time spent

image.png
image.png

3. 别让性能被锁住

image.png
image.png
image.png

Disruptor

4. GC友好的代码

image.png
image.png

GODEBUG=gctrace=1 go test -bench=”BenchmarkPassingArrayWithValue” 无法使用 go test -bench=”BenchmarkPassingArrayWithValue” “-trace=trace_val.out” 

  1. goos: windows
  2. goarch: amd64
  3. pkg: go-learning/src/ch43/gc_friendly/passing_ref
  4. cpu: Intel(R) Core(TM) i5-8265U CPU @ 1.60GHz
  5. BenchmarkPassingArrayWithValue-8              85          12794193 ns/op
  6. PASS
  7. ok      go-learning/src/ch43/gc_friendly/passing_ref    1.216s

go test -bench=”BenchmarkPassingArrayWithValue” “-trace=trace_ref.out” 

  1. goos: windows
  2. goarch: amd64
  3. pkg: go-learning/src/ch43/gc_friendly/passing_ref
  4. cpu: Intel(R) Core(TM) i5-8265U CPU @ 1.60GHz
  5. BenchmarkPassingArrayWithValue-8              85          13425085 ns/op
  6. PASS
  7. ok      go-learning/src/ch43/gc_friendly/passing_ref    1.311s

go tool trace .\trace_ref.out

  1. 2022/08/11 12:24:34 Parsing trace...
  2. 2022/08/11 12:24:34 Splitting trace...                                                 
  3. 2022/08/11 12:24:34 Opening browser. Trace viewer is listening on http://127.0.0.1:7042
  6. 此时会跳出一个网页页面

image.png
image.png
image.png
image.png