goroutine - Tony Bai

标签 goroutine 下的文章

Go程序调试、分析与优化

八月 25, 2015
18 条评论

Brad Fitzpatrick在YAPC Asia 2015（Yet Another Perl Conference）上做了一次技术分享，题为："Go Debugging, Profiling, and Optimization"。个人感觉这篇分享中价值最大的是BradFitz现场演示的一个有关如何对Go程序进行调试、分析和优化的 Demo，Brad将demo上传到了他个人在github.com的repo中，但不知为何，repo中的代码似乎与repo里talk.md中的说明不甚一致(btw，我并没有看video)。于是打算在这里按照Brad的思路重新走一遍demo的演示流程(所有演示代码在这里可以下载到)。

一、实验环境

$uname -a
Linux pc-tony 3.13.0-61-generic #100~precise1-Ubuntu SMP Wed Jul 29 12:06:40 UTC 2015 x86_64 x86_64 x86_64 GNU/Linux

注意:在Darwin或Windows下，profile的结果可能与这里有很大不同(甚至完全不一样的输出和瓶颈热点)。

$go version
go version go1.5 linux/amd64

$ go env
GOARCH="amd64"
GOBIN="/home1/tonybai/.bin/go15/bin"
GOEXE=""
GOHOSTARCH="amd64"
GOHOSTOS="linux"
GOOS="linux"
GOPATH="/home1/tonybai/proj/GoProjects"
GORACE=""
GOROOT="/home1/tonybai/.bin/go15"
GOTOOLDIR="/home1/tonybai/.bin/go15/pkg/tool/linux_amd64"
GO15VENDOREXPERIMENT="1"
CC="gcc"
GOGCCFLAGS="-fPIC -m64 -pthread -fmessage-length=0"
CXX="g++"
CGO_ENABLED="1"

代码基于Brad的github.com/bradfitz/talk-yapc-asia-2015。

二、待优化程序(step0)

待优化程序，也就是原始程序，我们放在step0中：

//go-debug-profile-optimization/step0/demo.go

package main

import (
    "fmt"
    "log"
    "net/http"
    "regexp"
)

var visitors int

func handleHi(w http.ResponseWriter, r *http.Request) {
    if match, _ := regexp.MatchString(`^\w*$`, r.FormValue("color")); !match {
        http.Error(w, "Optional color is invalid", http.StatusBadRequest)
        return
    }
    visitors++
    w.Header().Set("Content-Type", "text/html; charset=utf-8")
    w.Write([]byte("<h1 style='color: " + r.FormValue("color") +
        "'>Welcome!</h1>You are visitor number " + fmt.Sprint(visitors) + "!"))
}

func main() {
    log.Printf("Starting on port 8080")
    http.HandleFunc("/hi", handleHi)
    log.Fatal(http.ListenAndServe("127.0.0.1:8080", nil))
}

$go run demo.go
2015/08/25 09:42:35 Starting on port 8080

在浏览器输入：http://localhost:8080/hi

一切顺利的话，页面会显示：

Welcome!

You are visitor number 1!

三、添加测试代码

按照talk.md中的说明，brad repo中demo中根本没有测试代码(commit 2427d0faa12ed1fb05f1e6a1e69307c11259c2b2)。

于是我根据作者的意图，新增了demo_test.go，采用TestHandleHi_Recorder和TestHandleHi_TestServer对HandleHi进行测试：

//go-debug-profile-optimization/step0/demo_test.go
package main

import (
    "bufio"
    "net/http"
    "net/http/httptest"
    "strings"
    "testing"
)

func TestHandleHi_Recorder(t *testing.T) {
    rw := httptest.NewRecorder()
    handleHi(rw, req(t, "GET / HTTP/1.0\r\n\r\n"))
    if !strings.Contains(rw.Body.String(), "visitor number") {
        t.Errorf("Unexpected output: %s", rw.Body)
    }
}

func req(t *testing.T, v string) *http.Request {
    req, err := http.ReadRequest(bufio.NewReader(strings.NewReader(v)))
    if err != nil {
        t.Fatal(err)
    }
    return req
}

func TestHandleHi_TestServer(t *testing.T) {
    ts := httptest.NewServer(http.HandlerFunc(handleHi))
    defer ts.Close()
    res, err := http.Get(ts.URL)
    if err != nil {
        t.Error(err)
        return
    }
    if g, w := res.Header.Get("Content-Type"), "text/html; charset=utf-8"; g != w {
        t.Errorf("Content-Type = %q; want %q", g, w)
    }
    slurp, err := ioutil.ReadAll(res.Body)
    defer res.Body.Close()
    if err != nil {
        t.Error(err)
        return
    }
    t.Logf("Got: %s", slurp)
}

$ go test -v
=== RUN   TestHandleHi_Recorder
— PASS: TestHandleHi_Recorder (0.00s)
=== RUN   TestHandleHi_TestServer
— PASS: TestHandleHi_TestServer (0.00s)
    demo_test.go:45: Got: <h1 style='color: '>Welcome!</h1>You are visitor number 2!
PASS
ok     _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step0    0.007s

测试通过！

至此，step0使命结束。

四、Race Detector(竞态分析）

并发设计使得程序可以更好更有效的利用现代处理器的多核心。但并发设计很容易引入竞态，导致严重bug。Go程序中竞态就是当多个goroutine并发访问某共享数据且未使用同步机制时，且至少一个goroutine进行了写操作。不过go工具自带race分析功能。在分析优化step0中demo代码前，我们先要保证demo代码中不存在竞态。

工具的使用方法就是在go test后加上-race标志，在step0目录下：

$ go test -v -race
=== RUN   TestHandleHi_Recorder
— PASS: TestHandleHi_Recorder (0.00s)
=== RUN   TestHandleHi_TestServer
— PASS: TestHandleHi_TestServer (0.00s)
    demo_test.go:45: Got: <h1 style='color: '>Welcome!</h1>You are visitor number 2!
PASS
ok     _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step0    1.012s

-race通过做运行时分析做竞态分析，虽然不存在误报，但却存在实际有竞态，但工具没发现的情况。接下来我们改造一下测试代码，让test并发起来：

向step1(copy自step0)中demo_test.go中添加一个test method:

//go-debug-profile-optimization/step1/demo_test.go
… …
func TestHandleHi_TestServer_Parallel(t *testing.T) {
    ts := httptest.NewServer(http.HandlerFunc(handleHi))
    defer ts.Close()
    var wg sync.WaitGroup
    for i := 0; i < 2; i++ {
        wg.Add(1)
        go func() {
            defer wg.Done()
            res, err := http.Get(ts.URL)
            if err != nil {
                t.Error(err)
                return
            }
            if g, w := res.Header.Get("Content-Type"), "text/html; charset=utf-8"; g != w {
                t.Errorf("Content-Type = %q; want %q", g, w)
            }
            slurp, err := ioutil.ReadAll(res.Body)
            defer res.Body.Close()
            if err != nil {
                t.Error(err)
                return
            }
            t.Logf("Got: %s", slurp)
        }()
    }
    wg.Wait()
}
… …

执行竞态test：

$ go test -v -race
=== RUN   TestHandleHi_Recorder
— PASS: TestHandleHi_Recorder (0.00s)
=== RUN   TestHandleHi_TestServer
— PASS: TestHandleHi_TestServer (0.00s)
    demo_test.go:46: Got: <h1 style='color: '>Welcome!</h1>You are visitor number 2!
=== RUN   TestHandleHi_TestServer_Parallel
==================
WARNING: DATA RACE
Read by goroutine 22:
_/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step1.handleHi()
      /home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step1/demo.go:17 +0xf5
net/http.HandlerFunc.ServeHTTP()
      /tmp/workdir/go/src/net/http/server.go:1422 +0×47
net/http/httptest.(*waitGroupHandler).ServeHTTP()
      /tmp/workdir/go/src/net/http/httptest/server.go:200 +0xfe
net/http.serverHandler.ServeHTTP()
      /tmp/workdir/go/src/net/http/server.go:1862 +0×206
net/http.(*conn).serve()
      /tmp/workdir/go/src/net/http/server.go:1361 +0x117c

Previous write by goroutine 25:
_/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step1.handleHi()
      /home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step1/demo.go:17 +0×111
net/http.HandlerFunc.ServeHTTP()
      /tmp/workdir/go/src/net/http/server.go:1422 +0×47
net/http/httptest.(*waitGroupHandler).ServeHTTP()
      /tmp/workdir/go/src/net/http/httptest/server.go:200 +0xfe
net/http.serverHandler.ServeHTTP()
      /tmp/workdir/go/src/net/http/server.go:1862 +0×206
net/http.(*conn).serve()
      /tmp/workdir/go/src/net/http/server.go:1361 +0x117c

Goroutine 22 (running) created at:
net/http.(*Server).Serve()
/tmp/workdir/go/src/net/http/server.go:1910 +0×464

Goroutine 25 (running) created at:
net/http.(*Server).Serve()
      /tmp/workdir/go/src/net/http/server.go:1910 +0×464
==================
— PASS: TestHandleHi_TestServer_Parallel (0.00s)
    demo_test.go:71: Got: <h1 style='color: '>Welcome!</h1>You are visitor number 3!
    demo_test.go:71: Got: <h1 style='color: '>Welcome!</h1>You are visitor number 4!
PASS
Found 1 data race(s)
exit status 66
FAIL    _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step1    1.023s

工具发现demo.go第17行：
visitors++
是一处潜在的竞态条件。

visitors被多个goroutine访问但未采用同步机制。

既然发现了竞态条件，我们就需要fix it。有多种fix方法可选：

1、使用channel
2、使用Mutex
3、使用atomic

Brad使用了atomic：

//go-debug-profile-optimization/step1/demo.go
… …
var visitors int64 // must be accessed atomically

func handleHi(w http.ResponseWriter, r *http.Request) {
    if match, _ := regexp.MatchString(`^\w*$`, r.FormValue("color")); !match {
        http.Error(w, "Optional color is invalid", http.StatusBadRequest)
        return
    }
    visitNum := atomic.AddInt64(&visitors, 1)
    w.Header().Set("Content-Type", "text/html; charset=utf-8")
    w.Write([]byte("<h1 style='color: " + r.FormValue("color") +
        "'>Welcome!</h1>You are visitor number " + fmt.Sprint(visitNum) + "!"))
}
… …

再做一次测试：

竞态条件被消除了！

至此，step1结束了使命！

五、CPU Profiling

要做CPU Profilling，我们需要benchmark数据，Go test提供benchmark test功能，我们只要写对应的Benchmark测试方法即可：

//go-debug-profile-optimization/step2/demo_test.go
… …
func BenchmarkHi(b *testing.B) {
b.ReportAllocs()

    req, err := http.ReadRequest(bufio.NewReader(strings.NewReader("GET / HTTP/1.0\r\n\r\n")))
    if err != nil {
        b.Fatal(err)
    }

    for i := 0; i < b.N; i++ {
        rw := httptest.NewRecorder()
        handleHi(rw, req)
    }
}
… …

$ go test -v -run=^$ -bench=.
PASS
BenchmarkHi-4 100000 14808 ns/op 4961 B/op 81 allocs/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step2 1.648s

开始CPU Profiling：

$ go test -v -run=^$ -bench=^BenchmarkHi$ -benchtime=2s -cpuprofile=prof.cpu
PASS
BenchmarkHi-4 200000 14679 ns/op 4961 B/op 81 allocs/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step2 3.096s

执行完benchmark test后，step2目录下出现两个新文件prof.cpu和step2.test，这两个文件将作为后续go tool pprof的输入：
$ls
demo.go demo_test.go prof.cpu step2.test*

使用go profile viewer工具：

$ go tool pprof step2.test prof.cpu
Entering interactive mode (type "help" for commands)
(pprof) top
1830ms of 3560ms total (51.40%)
Dropped 53 nodes (cum <= 17.80ms)
Showing top 10 nodes out of 133 (cum >= 1290ms)
      flat flat%   sum%        cum   cum%
     480ms 13.48% 13.48%      980ms 27.53% runtime.growslice
     360ms 10.11% 23.60%      700ms 19.66% runtime.mallocgc
     170ms 4.78% 28.37%      170ms 4.78% runtime.heapBitsSetType
     170ms 4.78% 33.15%      200ms 5.62% runtime.scanblock
     120ms 3.37% 36.52%     1100ms 30.90% regexp.makeOnePass.func2
     120ms 3.37% 39.89%      550ms 15.45% runtime.newarray
     110ms 3.09% 42.98%      300ms 8.43% runtime.makeslice
     110ms 3.09% 46.07%      220ms 6.18% runtime.mapassign1
     100ms 2.81% 48.88%      100ms 2.81% runtime.futex
      90ms 2.53% 51.40%     1290ms 36.24% regexp.makeOnePass

(pprof) top –cum
0.18s of 3.56s total ( 5.06%)
Dropped 53 nodes (cum <= 0.02s)
Showing top 10 nodes out of 133 (cum >= 1.29s)
      flat flat%   sum%        cum   cum%
         0     0%     0%      3.26s 91.57% runtime.goexit
     0.02s 0.56% 0.56%      2.87s 80.62% BenchmarkHi
         0     0% 0.56%      2.87s 80.62% testing.(*B).launch
         0     0% 0.56%      2.87s 80.62% testing.(*B).runN
     0.03s 0.84% 1.40%      2.80s 78.65% step2.handleHi
     0.01s 0.28% 1.69%      2.46s 69.10% regexp.MatchString
         0     0% 1.69%      2.24s 62.92% regexp.Compile
         0     0% 1.69%      2.24s 62.92% regexp.compile
     0.03s 0.84% 2.53%      1.56s 43.82% regexp.compileOnePass
     0.09s 2.53% 5.06%      1.29s 36.24% regexp.makeOnePass

(pprof) list handleHi
Total: 3.56s
ROUTINE ======================== handleHi in go-debug-profile-optimization/step2/demo.go
      30ms      2.80s (flat, cum) 78.65% of Total
         .          .      9:)
         .          .     10:
         .          .     11:var visitors int64 // must be accessed atomically
         .          .     12:
         .          .     13:func handleHi(w http.ResponseWriter, r *http.Request) {
         .      2.47s     14:    if match, _ := regexp.MatchString(`^\w*$`, r.FormValue("color")); !match {
         .          .     15:        http.Error(w, "Optional color is invalid", http.StatusBadRequest)
         .          .     16:        return
         .          .     17:    }
      10ms       20ms     18:    visitNum := atomic.AddInt64(&visitors, 1)
      10ms       90ms     19:    w.Header().Set("Content-Type", "text/html; charset=utf-8")
      10ms       20ms     20:    w.Write([]byte("<h1 style='color: " + r.FormValue("color") +
         .      200ms     21:        "'>Welcome!</h1>You are visitor number " + fmt.Sprint(visitNum) + "!"))
         .          .     22:}
         .          .     23:
         .          .     24:func main() {
         .          .     25:    log.Printf("Starting on port 8080")
         .          .     26:    http.HandleFunc("/hi", handleHi)
(pprof)

从top –cum来看，handleHi消耗cpu较大，而handleHi中，又是MatchString耗时最长。

六、第一次优化

前面已经发现MatchString较为耗时，优化手段：让正则式仅编译一次(step3)：

// go-debug-profile-optimization/step3/demo.go

… …
var visitors int64 // must be accessed atomically

var rxOptionalID = regexp.MustCompile(`^\d*$`)

func handleHi(w http.ResponseWriter, r *http.Request) {
    if !rxOptionalID.MatchString(r.FormValue("color")) {
        http.Error(w, "Optional color is invalid", http.StatusBadRequest)
        return
    }

    visitNum := atomic.AddInt64(&visitors, 1)
    w.Header().Set("Content-Type", "text/html; charset=utf-8")
    w.Write([]byte("<h1 style='color: " + r.FormValue("color") +
        "'>Welcome!</h1>You are visitor number " + fmt.Sprint(visitNum) + "!"))
}
… …

运行一下bench：

$ go test -bench=.
PASS
BenchmarkHi-4 1000000 1678 ns/op 720 B/op 9 allocs/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step3 1.710s

对比之前在step2中运行的bench结果：

$ go test -v -run=^$ -bench=.
PASS
BenchmarkHi-4 100000 14808 ns/op 4961 B/op 81 allocs/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step2 1.648s

耗时相同，但优化后的bench运行了100w次，而之前的Bench运行10w次，相当于性能提高10倍。

再看看cpu prof结果：

$ go test -v -run=^$ -bench=^BenchmarkHi$ -benchtime=3s -cpuprofile=prof.cpu
PASS
BenchmarkHi-4 3000000 1640 ns/op 720 B/op 9 allocs/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step3 6.540s

$ go tool pprof step3.test prof.cpu
Entering interactive mode (type "help" for commands)
(pprof) top –cum 30
2.74s of 8.07s total (33.95%)
Dropped 72 nodes (cum <= 0.04s)
Showing top 30 nodes out of 103 (cum >= 0.56s)
      flat flat%   sum%        cum   cum%
         0     0%     0%      7.17s 88.85% runtime.goexit
     0.05s 0.62% 0.62%      6.21s 76.95% step3.BenchmarkHi
         0     0% 0.62%      6.21s 76.95% testing.(*B).launch
         0     0% 0.62%      6.21s 76.95% testing.(*B).runN
     0.06s 0.74% 1.36%      4.96s 61.46% step3.handleHi
     1.15s 14.25% 15.61%      2.35s 29.12% runtime.mallocgc
     0.02s 0.25% 15.86%      1.63s 20.20% runtime.systemstack
         0     0% 15.86%      1.53s 18.96% net/http.Header.Set
     0.06s 0.74% 16.60%      1.53s 18.96% net/textproto.MIMEHeader.Set
     0.09s 1.12% 17.72%      1.22s 15.12% runtime.newobject
     0.05s 0.62% 18.34%      1.09s 13.51% fmt.Sprint
     0.20s 2.48% 20.82%         1s 12.39% runtime.mapassign1
         0     0% 20.82%      0.81s 10.04% runtime.mcall
     0.01s 0.12% 20.94%      0.79s 9.79% runtime.schedule
     0.05s 0.62% 21.56%      0.76s 9.42% regexp.(*Regexp).MatchString
     0.09s 1.12% 22.68%      0.71s 8.80% regexp.(*Regexp).doExecute
     0.01s 0.12% 22.80%      0.71s 8.80% runtime.concatstring5
     0.20s 2.48% 25.28%      0.70s 8.67% runtime.concatstrings
         0     0% 25.28%      0.69s 8.55% runtime.gosweepone
     0.05s 0.62% 25.90%      0.69s 8.55% runtime.mSpan_Sweep
         0     0% 25.90%      0.68s 8.43% runtime.bgsweep
     0.04s   0.5% 26.39%      0.68s 8.43% runtime.newarray
     0.01s 0.12% 26.52%      0.67s 8.30% runtime.goschedImpl
     0.01s 0.12% 26.64%      0.65s 8.05% runtime.gosched_m
         0     0% 26.64%      0.65s 8.05% runtime.gosweepone.func1
     0.01s 0.12% 26.77%      0.65s 8.05% runtime.sweepone
     0.28s 3.47% 30.24%      0.62s 7.68% runtime.makemap
     0.17s 2.11% 32.34%      0.59s 7.31% runtime.heapBitsSweepSpan
     0.02s 0.25% 32.59%      0.58s 7.19% fmt.(*pp).doPrint
     0.11s 1.36% 33.95%      0.56s 6.94% fmt.(*pp).printArg

handleHi耗时有一定下降。

七、Mem Profiling

在step3目录下执行bench，获取mem分配数据：

$ go test -v -run=^$ -bench=^BenchmarkHi$ -benchtime=2s -memprofile=prof.mem
PASS
BenchmarkHi-4 2000000 1657 ns/op 720 B/op 9 allocs/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step3 5.002s

使用pprof工具分析mem：

$ go tool pprof –alloc_space step3.test prof.mem
Entering interactive mode (type "help" for commands)
(pprof) top
2065.91MB of 2067.41MB total (99.93%)
Dropped 14 nodes (cum <= 10.34MB)
      flat flat%   sum%        cum   cum%
1076.35MB 52.06% 52.06% 1076.35MB 52.06% net/textproto.MIMEHeader.Set
535.54MB 25.90% 77.97% 2066.91MB   100% step3.BenchmarkHi
406.52MB 19.66% 97.63% 1531.37MB 74.07% step3.handleHi
   47.50MB 2.30% 99.93%    48.50MB 2.35% fmt.Sprint
         0     0% 99.93% 1076.35MB 52.06% net/http.Header.Set
         0     0% 99.93% 2066.91MB   100% runtime.goexit
         0     0% 99.93% 2066.91MB   100% testing.(*B).launch
         0     0% 99.93% 2066.91MB   100% testing.(*B).runN

(pprof) top -cum
2065.91MB of 2067.41MB total (99.93%)
Dropped 14 nodes (cum <= 10.34MB)
      flat flat%   sum%        cum   cum%
535.54MB 25.90% 25.90% 2066.91MB   100% step3.BenchmarkHi
         0     0% 25.90% 2066.91MB   100% runtime.goexit
         0     0% 25.90% 2066.91MB   100% testing.(*B).launch
         0     0% 25.90% 2066.91MB   100% testing.(*B).runN
406.52MB 19.66% 45.57% 1531.37MB 74.07% step3.handleHi
         0     0% 45.57% 1076.35MB 52.06% net/http.Header.Set
1076.35MB 52.06% 97.63% 1076.35MB 52.06% net/textproto.MIMEHeader.Set
   47.50MB 2.30% 99.93%    48.50MB 2.35% fmt.Sprint

(pprof) list handleHi
Total: 2.02GB
     ROUTINE =========step3.handleHi in step3/demo.go
406.52MB     1.50GB (flat, cum) 74.07% of Total
         .          .     17:        http.Error(w, "Optional color is invalid", http.StatusBadRequest)
         .          .     18:        return
         .          .     19:    }
         .          .     20:
         .          .     21:    visitNum := atomic.AddInt64(&visitors, 1)
         .     1.05GB     22:    w.Header().Set("Content-Type", "text/html; charset=utf-8")
         .          .     23:    w.Write([]byte("<h1 style='color: " + r.FormValue("color") +
406.52MB   455.02MB     24:        "'>Welcome!</h1>You are visitor number " + fmt.Sprint(visitNum) + "!"))
         .          .     25:}
         .          .     26:
         .          .     27:func main() {
         .          .     28:    log.Printf("Starting on port 8080")
         .          .     29:    http.HandleFunc("/hi", handleHi)
(pprof)

可以看到handleHi22、23两行占用了较多内存。

八、第二次优化

第二次优化的方法：
1、删除w.Header().Set这行
2、用fmt.Fprintf替代w.Write

第二次优化的代码在step4目录中：

// go-debug-profile-optimization/step4/demo.go
… …
func handleHi(w http.ResponseWriter, r *http.Request) {
    if !rxOptionalID.MatchString(r.FormValue("color")) {
        http.Error(w, "Optional color is invalid", http.StatusBadRequest)
        return
    }

visitNum := atomic.AddInt64(&visitors, 1)
fmt.Fprintf(w, "<html><h1 stype='color: \"%s\"'>Welcome!</h1>You are visitor number %d!", r.FormValue("color"), visitNum)
}
… …

执行一遍pprof:

$ go test -v -run=^$ -bench=^BenchmarkHi$ -benchtime=2s -memprofile=prof.mem
PASS
BenchmarkHi-4 2000000 1428 ns/op 304 B/op 6 allocs/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step4 4.343s

$ go tool pprof –alloc_space step4.test prof.mem
Entering interactive mode (type "help" for commands)
(pprof) top
868.06MB of 868.56MB total (99.94%)
Dropped 5 nodes (cum <= 4.34MB)
      flat flat%   sum%        cum   cum%
559.54MB 64.42% 64.42%   868.06MB 99.94% step4.BenchmarkHi
219.52MB 25.27% 89.70%   219.52MB 25.27% bytes.makeSlice
      89MB 10.25% 99.94%   308.52MB 35.52% step4.handleHi
         0     0% 99.94%   219.52MB 25.27% bytes.(*Buffer).Write
         0     0% 99.94%   219.52MB 25.27% bytes.(*Buffer).grow
         0     0% 99.94%   219.52MB 25.27% fmt.Fprintf
         0     0% 99.94%   219.52MB 25.27% net/http/httptest.(*ResponseRecorder).Write
         0     0% 99.94%   868.06MB 99.94% runtime.goexit
         0     0% 99.94%   868.06MB 99.94% testing.(*B).launch
         0     0% 99.94%   868.06MB 99.94% testing.(*B).runN
(pprof) top –cum
868.06MB of 868.56MB total (99.94%)
Dropped 5 nodes (cum <= 4.34MB)
      flat flat%   sum%        cum   cum%
559.54MB 64.42% 64.42%   868.06MB 99.94% step4.BenchmarkHi
         0     0% 64.42%   868.06MB 99.94% runtime.goexit
         0     0% 64.42%   868.06MB 99.94% testing.(*B).launch
         0     0% 64.42%   868.06MB 99.94% testing.(*B).runN
      89MB 10.25% 74.67%   308.52MB 35.52% step4.handleHi
         0     0% 74.67%   219.52MB 25.27% bytes.(*Buffer).Write
         0     0% 74.67%   219.52MB 25.27% bytes.(*Buffer).grow
219.52MB 25.27% 99.94%   219.52MB 25.27% bytes.makeSlice
         0     0% 99.94%   219.52MB 25.27% fmt.Fprintf
         0     0% 99.94%   219.52MB 25.27% net/http/httptest.(*ResponseRecorder).Write
(pprof) list handleHi
Total: 868.56MB
ROUTINE ============ step4.handleHi in step4/demo.go
      89MB   308.52MB (flat, cum) 35.52% of Total
         .          .     17:        http.Error(w, "Optional color is invalid", http.StatusBadRequest)
         .          .     18:        return
         .          .     19:    }
         .          .     20:
         .          .     21:    visitNum := atomic.AddInt64(&visitors, 1)
      89MB   308.52MB     22:    fmt.Fprintf(w, "<html><h1 stype='color: \"%s\"'>Welcome!</h1>You are visitor number %d!", r.FormValue("color"), visitNum)
         .          .     23:}
         .          .     24:
         .          .     25:func main() {
         .          .     26:    log.Printf("Starting on port 8080")
         .          .     27:    http.HandleFunc("/hi", handleHi)
(pprof)

可以看出内存占用大幅减少。

九、Benchcmp

golang.org/x/tools中有一个工具：benchcmp，可以给出两次bench的结果对比。

github.com/golang/tools是golang.org/x/tools的一个镜像。安装benchcmp步骤：

1、go get -u github.com/golang/tools
2、mkdir -p $GOPATH/src/golang.org/x
3、mv $GOPATH/src/github.com/golang/tools $GOPATH/src/golang.org/x
4、go install golang.org/x/tools/cmd/benchcmp

我们分别在step2、step3和step4下执行如下命令：

$ go-debug-profile-optimization/step2$ go test -bench=. -memprofile=prof.mem | tee mem.2
PASS
BenchmarkHi-4 100000 14786 ns/op 4961 B/op 81 allocs/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step2 1.644s

go-debug-profile-optimization/step3$ go test -bench=. -memprofile=prof.mem | tee mem.3
PASS
BenchmarkHi-4 1000000 1662 ns/op 720 B/op 9 allocs/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step3 1.694s

go-debug-profile-optimization/step4$ go test -bench=. -memprofile=prof.mem | tee mem.4
PASS
BenchmarkHi-4 1000000 1428 ns/op 304 B/op 6 allocs/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step4 1.456s

利用benchcmp工具对比结果（benchcmp old new）：

$ benchcmp step3/mem.3 step4/mem.4
benchmark old ns/op new ns/op delta
BenchmarkHi-4 1662 1428 -14.08%

benchmark old allocs new allocs delta
BenchmarkHi-4 9 6 -33.33%

benchmark old bytes new bytes delta
BenchmarkHi-4 720 304 -57.78%

$ benchcmp step2/mem.2 step4/mem.4
benchmark old ns/op new ns/op delta
BenchmarkHi-4 14786 1428 -90.34%

benchmark old allocs new allocs delta
BenchmarkHi-4 81 6 -92.59%

benchmark old bytes new bytes delta
BenchmarkHi-4 4961 304 -93.87%

可以看出优化后，内存分配大幅减少，gc的时间也随之减少。

十、内存来自哪

我们在BenchmarkHi中清理每次handleHi执行后的内存：

//step5/demo_test.go
… …
func BenchmarkHi(b *testing.B) {
b.ReportAllocs()

    req, err := http.ReadRequest(bufio.NewReader(strings.NewReader("GET / HTTP/1.0\r\n\r\n")))
    if err != nil {
        b.Fatal(err)
    }

    for i := 0; i < b.N; i++ {
        rw := httptest.NewRecorder()
        handleHi(rw, req)
        reset(rw)
    }
}

func reset(rw *httptest.ResponseRecorder) {
    m := rw.HeaderMap
    for k := range m {
        delete(m, k)
    }
    body := rw.Body
    body.Reset()
    *rw = httptest.ResponseRecorder{
        Body:      body,
        HeaderMap: m,
    }
}

… …
$ go test -v -run=^$ -bench=^BenchmarkHi$ -benchtime=2s -memprofile=prof.mem
PASS
BenchmarkHi-4 2000000 1518 ns/op 304 B/op 6 allocs/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step5 4.577s

$ go tool pprof –alloc_space step5.test prof.mem
Entering interactive mode (type "help" for commands)
(pprof) top –cum 10
290.52MB of 291.52MB total (99.66%)
Dropped 14 nodes (cum <= 1.46MB)
      flat flat%   sum%        cum   cum%
         0     0%     0%   291.02MB 99.83% runtime.goexit
179.01MB 61.41% 61.41%   290.52MB 99.66% step5.BenchmarkHi
         0     0% 61.41%   290.52MB 99.66% testing.(*B).launch
         0     0% 61.41%   290.52MB 99.66% testing.(*B).runN
   26.50MB 9.09% 70.50%   111.51MB 38.25% step5.handleHi
         0     0% 70.50%    85.01MB 29.16% bytes.(*Buffer).Write
         0     0% 70.50%    85.01MB 29.16% bytes.(*Buffer).grow
   85.01MB 29.16% 99.66%    85.01MB 29.16% bytes.makeSlice
         0     0% 99.66%    85.01MB 29.16% fmt.Fprintf
         0     0% 99.66%    85.01MB 29.16% net/http/httptest.(*ResponseRecorder).Write
(pprof) list handleHi
Total: 291.52MB
ROUTINE ======================== _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step5.handleHi in /home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step5/demo.go
   26.50MB   111.51MB (flat, cum) 38.25% of Total
         .          .     17:        http.Error(w, "Optional color is invalid", http.StatusBadRequest)
         .          .     18:        return
         .          .     19:    }
         .          .     20:
         .          .     21:    visitNum := atomic.AddInt64(&visitors, 1)
   26.50MB   111.51MB     22:    fmt.Fprintf(w, "<html><h1 stype='color: \"%s\"'>Welcome!</h1>You are visitor number %d!", r.FormValue("color"), visitNum)
         .          .     23:}
         .          .     24:
         .          .     25:func main() {
         .          .     26:    log.Printf("Starting on port 8080")
         .          .     27:    http.HandleFunc("/hi", handleHi)
(pprof)

内存从300MB降到111MB。内存来自哪？看到list handleHi，fmt.Fprintf分配了111.51MB。

我们来看这一行代码：
fmt.Fprintf(w, "<h1 style='color: %s'>Welcome!</h1>You are visitor number %d!",
r.FormValue("color"), num)

fmt.Fprintf的manual：

$ go doc fmt.Fprintf
func Fprintf(w io.Writer, format string, a …interface{}) (n int, err error)

Fprintf formats according to a format specifier and writes to w. It returns
the number of bytes written and any write error encountered.

这里回顾一下Go type在runtime中的内存占用：

A Go interface is 2 words of memory: (type, pointer).
A Go string is 2 words of memory: (base pointer, length)
A Go slice is 3 words of memory: (base pointer, length, capacity)

每次调用fmt.Fprintf，参数以value值形式传入函数时，程序就要为每个变参分配一个占用16bytes的empty interface，然后用传入的类型初始化该interface value。这就是这块累计分配内存较多的原因。

十一、消除所有内存分配

下面的优化代码可能在实际中并不需要，但一旦真的成为瓶颈，可以这么做：

//go-debug-profile-optimization/step6/demo.go
… …
var bufPool = sync.Pool{
    New: func() interface{} {
        return new(bytes.Buffer)
    },
}

    visitNum := atomic.AddInt64(&visitors, 1)
    buf := bufPool.Get().(*bytes.Buffer)
    defer bufPool.Put(buf)
    buf.Reset()
    buf.WriteString("<h1 style='color: ")
    buf.WriteString(r.FormValue("color"))
    buf.WriteString("'>Welcome!</h1>You are visitor number ")
    b := strconv.AppendInt(buf.Bytes(), int64(visitNum), 10)
    b = append(b, '!')
    w.Write(b)
}
… …

$ go test -v -run=^$ -bench=^BenchmarkHi$ -benchtime=2s -memprofile=prof.mem
PASS
BenchmarkHi-4 5000000 780 ns/op 192 B/op 3 allocs/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step6 4.709s

go tool pprof –alloc_space step6.test prof.mem
Entering interactive mode (type "help" for commands)
(pprof) top –cum 10
1.07GB of 1.07GB total ( 100%)
Dropped 5 nodes (cum <= 0.01GB)
      flat flat%   sum%        cum   cum%
    1.07GB   100%   100%     1.07GB   100% step6.BenchmarkHi
         0     0%   100%     1.07GB   100% runtime.goexit
         0     0%   100%     1.07GB   100% testing.(*B).launch
         0     0%   100%     1.07GB   100% testing.(*B).runN

$ go test -bench=. -memprofile=prof.mem | tee mem.6
PASS
BenchmarkHi-4 2000000 790 ns/op 192 B/op 3 allocs/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step6 2.401s

$ benchcmp step5/mem.5 step6/mem.6
benchmark old ns/op new ns/op delta
BenchmarkHi-4 1513 790 -47.79%

benchmark old allocs new allocs delta
BenchmarkHi-4 6 3 -50.00%

benchmark old bytes new bytes delta
BenchmarkHi-4 304 192 -36.84%

可以看到handleHi已经不在top列表中了。benchcmp结果也显示内存分配又有大幅下降！

十二、竞争(Contention)优化

为handleHi编写一个Parallel benchmark test:

//go-debug-profile-optimization/step7/demo_test.go
… …
func BenchmarkHiParallel(b *testing.B) {
    r, err := http.ReadRequest(bufio.NewReader(strings.NewReader("GET / HTTP/1.0\r\n\r\n")))
    if err != nil {
        b.Fatal(err)
    }

    b.RunParallel(func(pb *testing.PB) {
        rw := httptest.NewRecorder()
        for pb.Next() {
            handleHi(rw, r)
            reset(rw)
        }
    })
}
… …

执行测试，并分析结果:

$ go test -bench=Parallel -blockprofile=prof.block
PASS
BenchmarkHiParallel-4 5000000 305 ns/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step7 1.947s

$ go tool pprof step7.test prof.block
Entering interactive mode (type "help" for commands)
(pprof) top –cum 10
3.68s of 3.72s total (98.82%)
Dropped 29 nodes (cum <= 0.02s)
Showing top 10 nodes out of 20 (cum >= 1.84s)
      flat flat%   sum%        cum   cum%
         0     0%     0%      3.72s   100% runtime.goexit
     1.84s 49.46% 49.46%      1.84s 49.46% runtime.chanrecv1
         0     0% 49.46%      1.84s 49.46% main.main
         0     0% 49.46%      1.84s 49.46% runtime.main
         0     0% 49.46%      1.84s 49.46% testing.(*M).Run
         0     0% 49.46%      1.84s 49.43% testing.(*B).run
         0     0% 49.46%      1.84s 49.43% testing.RunBenchmarks
         0     0% 49.46%      1.84s 49.36% step7.BenchmarkHiParallel
     1.84s 49.36% 98.82%      1.84s 49.36% sync.(*WaitGroup).Wait
         0     0% 98.82%      1.84s 49.36% testing.(*B).RunParallel
(pprof) list BenchmarkHiParallel
Total: 3.72s
ROUTINE ====== step7.BenchmarkHiParallel in step7/demo_test.go
         0      1.84s (flat, cum) 49.36% of Total
         .          .    113:        rw := httptest.NewRecorder()
         .          .    114:        for pb.Next() {
         .          .    115:            handleHi(rw, r)
         .          .    116:            reset(rw)
         .          .    117:        }
         .      1.84s    118:    })
         .          .    119:}
ROUTINE ==== step7.BenchmarkHiParallel.func1 in step7/demo_test.go
         0    43.02ms (flat, cum) 1.16% of Total
         .          .    110:    }
         .          .    111:
         .          .    112:    b.RunParallel(func(pb *testing.PB) {
         .          .    113:        rw := httptest.NewRecorder()
         .          .    114:        for pb.Next() {
         .    43.02ms    115:            handleHi(rw, r)
         .          .    116:            reset(rw)
         .          .    117:        }
         .          .    118:    })
         .          .    119:}
(pprof) list handleHi
Total: 3.72s
ROUTINE =====step7.handleHi in step7/demo.go
         0    43.02ms (flat, cum) 1.16% of Total
         .          .     18:        return new(bytes.Buffer)
         .          .     19:    },
         .          .     20:}
         .          .     21:
         .          .     22:func handleHi(w http.ResponseWriter, r *http.Request) {
         .    43.01ms     23:    if !rxOptionalID.MatchString(r.FormValue("color")) {
         .          .     24:        http.Error(w, "Optional color is invalid", http.StatusBadRequest)
         .          .     25:        return
         .          .     26:    }
         .          .     27:
         .          .     28:    visitNum := atomic.AddInt64(&visitors, 1)
         .     2.50us     29:    buf := bufPool.Get().(*bytes.Buffer)
         .          .     30:    defer bufPool.Put(buf)
         .          .     31:    buf.Reset()
         .          .     32:    buf.WriteString("<h1 style='color: ")
         .          .     33:    buf.WriteString(r.FormValue("color"))
         .          .     34:    buf.WriteString("'>Welcome!</h1>You are visitor number ")
(pprof)

handleHi中MatchString这块是一个焦点，这里耗时较多。

优化方法（step8）：

//go-debug-profile-optimization/step8/demo.go
… …
var colorRxPool = sync.Pool{
New: func() interface{} { return regexp.MustCompile(`\w*$`) },
}

func handleHi(w http.ResponseWriter, r *http.Request) {
    if !colorRxPool.Get().(*regexp.Regexp).MatchString(r.FormValue("color")) {
        http.Error(w, "Optional color is invalid", http.StatusBadRequest)
        return
    }

测试执行与分析：

$ go test -bench=Parallel -blockprofile=prof.block
PASS
BenchmarkHiParallel-4 100000 19190 ns/op
ok _/home1/tonybai/proj/opensource/github/experiments/go-debug-profile-optimization/step8 2.219s

$ go tool pprof step8.test prof.block
Entering interactive mode (type "help" for commands)
(pprof) top –cum 10
4.22s of 4.23s total (99.69%)
Dropped 28 nodes (cum <= 0.02s)
Showing top 10 nodes out of 12 (cum >= 2.11s)
      flat flat%   sum%        cum   cum%
         0     0%     0%      4.23s   100% runtime.goexit
     2.11s 49.90% 49.90%      2.11s 49.90% runtime.chanrecv1
         0     0% 49.90%      2.11s 49.89% main.main
         0     0% 49.90%      2.11s 49.89% runtime.main
         0     0% 49.90%      2.11s 49.89% testing.(*M).Run
         0     0% 49.90%      2.11s 49.86% testing.(*B).run
         0     0% 49.90%      2.11s 49.86% testing.RunBenchmarks
         0     0% 49.90%      2.11s 49.79% step8.BenchmarkHiParallel
     2.11s 49.79% 99.69%      2.11s 49.79% sync.(*WaitGroup).Wait
         0     0% 99.69%      2.11s 49.79% testing.(*B).RunParallel
(pprof) list BenchmarkHiParallel
Total: 4.23s
ROUTINE ======step8.BenchmarkHiParallel in step8/demo_test.go
         0      2.11s (flat, cum) 49.79% of Total
         .          .    113:        rw := httptest.NewRecorder()
         .          .    114:        for pb.Next() {
         .          .    115:            handleHi(rw, r)
         .          .    116:            reset(rw)
         .          .    117:        }
         .      2.11s    118:    })
         .          .    119:}
ROUTINE ======step8.BenchmarkHiParallel.func1 in step8/demo_test.go
         0    11.68ms (flat, cum) 0.28% of Total
         .          .    110:    }
         .          .    111:
         .          .    112:    b.RunParallel(func(pb *testing.PB) {
         .          .    113:        rw := httptest.NewRecorder()
         .          .    114:        for pb.Next() {
         .    11.68ms    115:            handleHi(rw, r)
         .          .    116:            reset(rw)
         .          .    117:        }
         .          .    118:    })
         .          .    119:}
(pprof) list handleHi
Total: 4.23s
ROUTINE ======step8.handleHi in step8/demo.go
         0    11.68ms (flat, cum) 0.28% of Total
         .          .     21:var colorRxPool = sync.Pool{
         .          .     22:    New: func() interface{} { return regexp.MustCompile(`\w*$`) },
         .          .     23:}
         .          .     24:
         .          .     25:func handleHi(w http.ResponseWriter, r *http.Request) {
         .     5.66ms     26:    if !colorRxPool.Get().(*regexp.Regexp).MatchString(r.FormValue("color")) {
         .          .     27:        http.Error(w, "Optional color is invalid", http.StatusBadRequest)
         .          .     28:        return
         .          .     29:    }
         .          .     30:
         .          .     31:    visitNum := atomic.AddInt64(&visitors, 1)
         .     6.02ms     32:    buf := bufPool.Get().(*bytes.Buffer)
         .          .     33:    defer bufPool.Put(buf)
         .          .     34:    buf.Reset()
         .          .     35:    buf.WriteString("<h1 style='color: ")
         .          .     36:    buf.WriteString(r.FormValue("color"))
         .          .     37:    buf.WriteString("'>Welcome!</h1>You are visitor number ")
(pprof)

优化后，MatchString从43ms降到5.66ms。

Go 1.5中值得关注的几个变化

七月 10, 2015
4 条评论

在GopherCon2015开幕之际，Google Go Team终于放出了Go 1.5Beta1版本的安装包。在go 1.5Beta1的发布说明中，Go Team也诚恳地承认Go 1.5将打破之前6个月一个版本的发布周期，这是因为Go 1.5变动太大，需要更多时间来准备这次发布（fix bug, Write doc）。关于Go 1.5的变化，之前Go Team staff在各种golang技术会议的slide 中暴露不少，包括：

- 编译器和运行时由C改为Go（及少量汇编语言）重写，实现了Go的self Bootstrap(自举）
- Garbage Collector优化，大幅降低GC延迟（Stop The World），实现Gc在单独的goroutine中与其他user goroutine并行运行。
- 标准库变更以及一些go tools的引入。

每项变动都会让gopher激动不已。但之前也只是激动，这次beta1出来后，我们可以实际体会一下这些变动带来的“快感”了。Go 1.5beta1的发布文档目前还不全，有些地方还有“待补充”字样，可能与最终go 1.5发布时的版本有一定差异，不过大体内容应该是固定不变的了。这篇文章就想和大家一起浅显地体验一下go 1.5都给gophers们带来了哪些变化吧。

一、语言

【map literal】

go 1.5依旧兼容Go 1 language specification，但修正了之前的一个“小疏忽”。

Go 1.4及之前版本中，我们只能这么来写代码：

//testmapliteral.go
package main

import (
"fmt"
)

type Point struct {
x int
y int
}

func main() {
    var sl = []Point{{3, 4}, {5, 6}}
    var m = map[Point]string{
        Point{3,4}:"foo1",
        Point{5,6}:"foo2",
    }
    fmt.Println(sl)
    fmt.Println(m)
}

可以看到，对于Point这个struct来说，在初始化一个slice时，slice value literal中无需显式的带上元素类型Point，即

var sl = []Point{{3, 4}, {5, 6}}

而不是

var sl = []Point{Point{3, 4}, Point{5, 6}}

但当Point作为map类型的key类型时，初始化map时则要显式带上元素类型Point。Go team承认这是当初的一个疏忽，在本次Go 1.5中将该问题fix掉了。也就是说，下面的代码在Go 1.5中可以顺利编译通过：

func main() {
    var sl = []Point{{3, 4}, {5, 6}}
    var m = map[Point]string{
        {3,4}:"foo1",
        {5,6}:"foo2",
    }
    fmt.Println(sl)
    fmt.Println(m)
}

【GOMAXPROCS】

就像这次GopherCon2015上现任Google Go project Tech Lead的Russ Cox的开幕Keynote中所说的那样：Go目标定位于高度并发的云环境。Go 1.5中将标识并发系统线程个数的GOMAXPROCS的初始值由1改为了运行环境的CPU核数。

// testgomaxprocs.go
package main

import (
"fmt"
"runtime"
)

func main() {
fmt.Println(runtime.GOMAXPROCS(-1))
fmt.Println(runtime.NumGoroutine())
}

这个代码在Go 1.4下（Mac OS X 4核）运行结果是：

$go run testgomaxprocs.go
1
4

而在go 1.5beta1下，结果为：

$go run testgomaxprocs.go
4
4

二、编译

【简化跨平台编译】

1.5之前的版本要想实现跨平台编译，需要到$GOROOT/src下重新执行一遍make.bash，执行前设置好目标环境的环境变量(GOOS和 GOARCH)，Go 1.5大大简化这个过程，使得跨平台编译几乎与普通编译一样简单。下面是一个简单的例子：

//testcrosscompile.go
package main

import (
"fmt"
"runtime"
)

func main() {
fmt.Println(runtime.GOOS)
}

在我的Mac上，本地编译执行：
$go build -o testcrosscompile_darwin testcrosscompile.go
$testcrosscompile_darwin
darwin

跨平台编译linux amd64上的目标程序：

$GOOS=linux GOARCH=amd64 go build -o testcrosscompile_linux testcrosscompile.go

上传testcrosscompile_linux到ubuntu 14.04上执行：
$testcrosscompile_linux
linux

虽然从用户角度跨平台编译命令很简单，但事实是go替你做了很多事情，我们可以通过build -x -v选项来输出编译的详细过程，你会发现go会先进入到$GOROOT/src重新编译runtime.a以及一些平台相关的包。编译输出的信息太多，这里就不贴出来了。但在1.5中这个过程非常快（10秒以内），与1.4之前版本的跨平台编译相比，完全不是一个级别，这也许就是编译器用Go重写完的好处之一吧。

除了直接使用go build，我们还可以使用go tool compile和go tool link来编译程序，实际上go build也是调用这两个工具完成编译过程的。

$go tool compile testcrosscompile.go
testcrosscompile.o
$go tool link testcrosscompile.o
a.out
$a.out
darwin

go 1.5移除了以前的6a,6l之类的编译连接工具，将这些工具整合到go tool中。并且go tool compile的输出默认改为.o文件，链接器输出默认改为了a.out。

【动态共享库】

个人不是很赞同Go语言增加对动态共享库的支持，.so和.dll这类十多年前的技术在如今内存、磁盘空间都“非常大”的前提下，似乎已经失去了以往的魅力。并且动态共享库所带来的弊端："DLL hell"会让程序后续的运维痛苦不已。Docker等轻量级容器的兴起，面向不变性的架构(immutable architecture)受到更多的关注。人们更多地会在container这一层进行操作，一个纯static link的应用在部署和维护方面将会有天然优势，.so只会增加复杂性。如果单纯从与c等其他语言互操作的角度，似乎用途也不会很广泛(但游戏或ui领域可能会用到)。不过go 1.5还是增加了对动态链接库的支持，不过从go tool compile和link的doc说明来看，目前似乎还处于实验阶段。

既然go 1.5已经支持了shared library，我们就来实验一下。我们先规划一下测试repository的目录结构：

$GOPATH
    /src
        /testsharedlib
            /shlib
                – lib.go
        /app
            /main.go

lib.go中的代码很简单：

//lib.go
package shlib

import "fmt"

// export Method1
func Method1() {
fmt.Println("shlib -Method1")
}

对于希望导出的方法，采用export标记。

我们来将这个lib.go编译成shared lib，注意目前似乎只有linux平台支持编译go shared library：

$ go build -buildmode=shared testsharedlib/shlib
# /tmp/go-build709704006/libtestsharedlib-shlib.so
warning: unable to find runtime/cgo.a

编译ok，那个warning是何含义不是很理解。

要想.so被其他go程序使用，需要将.so安装到相关目录下。我们install一下试试：

$ go install -buildmode=shared testsharedlib/shlib
multiple roots /home1/tonybai/test/go/go15/pkg/linux_amd64_dynlink & /home1/tonybai/.bin/go15beta1/go/pkg/linux_amd64_dynlink

go工具居然纠结了，不知道选择放在哪里，一个是$GOPATH/pkg/linux_amd64_dynlink，另外一个则是$GOROOT/pkg/linux_amd64_dynlink，我不清楚这是不是一个bug。

在Google了之后，我尝试了网上的一个解决方法，先编译出runtime的动态共享库：

$go install -buildmode=shared runtime sync/atomic

编译安装后，你就会在$GOROOT/pkg下面看到多出来一个目录：linux_amd64_dynlink。这个目录下的结构如下：

$ ls -R
.:
libruntime,sync-atomic.so runtime runtime.a runtime.shlibname sync

./runtime:
cgo.a cgo.shlibname

./sync:
atomic.a atomic.shlibname

这里看到了之前warning提到的runtime/cgo.a，我们再来重新执行一下build，看看能不能消除warning:

$ go build -buildmode=shared testsharedlib/shlib
# /tmp/go-build086398801/libtestsharedlib-shlib.so
/home1/tonybai/.bin/go15beta1/go/pkg/tool/linux_amd64/link: cannot implicitly include runtime/cgo in a shared library

这回连warnning都没有了，直接是一个error。这里提示：无法在一个共享库中隐式包含runtime/cgo。也就是说我们在构建 testshared/shlib这个动态共享库时，还需要显式的link到runtime/cgo，这里就需要另外一个命令行标志：- linkshared。我们再来试试：

$ go build -linkshared -buildmode=shared testsharedlib/shlib

这回build成功！我们再来试试install：

$ go install -linkshared -buildmode=shared testsharedlib/shlib

同样成功了。并且我们在$GOPATH/pkg/linux_amd64_dynlink下发现了共享库：

$ ls -R
.:
libtestsharedlib-shlib.so testsharedlib

./testsharedlib:
shlib.a shlib.shlibname

$ ldd libtestsharedlib-shlib.so
    linux-vdso.so.1 => (0x00007fff93983000)
    libruntime,sync-atomic.so => /home1/tonybai/.bin/go15beta1/go/pkg/linux_amd64_dynlink/libruntime,sync-atomic.so (0x00007fa150f1b000)
    libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007fa150b3f000)
    libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007fa150921000)
    /lib64/ld-linux-x86-64.so.2 (0x00007fa1517a7000)

好了，既然共享库编译出来了。我们就来用一下这个共享库。

//app/main.go

package main

import (
"testsharedlib/shlib"
)

func main() {
shlib.Method1()
}

$ go build -linkshared main.go
$ ldd main
    linux-vdso.so.1 => (0x00007fff579f7000)
    libruntime,sync-atomic.so => /home1/tonybai/.bin/go15beta1/go/pkg/linux_amd64_dynlink/libruntime,sync-atomic.so (0x00007fa8d6df2000)
    libtestsharedlib-shlib.so => /home1/tonybai/test/go/go15/pkg/linux_amd64_dynlink/libtestsharedlib-shlib.so (0x00007fa8d6962000)
    libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007fa8d6586000)
    libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007fa8d6369000)
    /lib64/ld-linux-x86-64.so.2 (0x00007fa8d71ef000)

$ main
shlib -Method1

编译执行ok。从输出结果来看，我们可以清晰看到main依赖的.so以及so的路径。我们再来试试，如果将testsharedlib源码目录移除后，是否还能编译ok：

$ go build -linkshared main.go
main.go:4:2: cannot find package "testsharedlib/shlib" in any of:
/home1/tonybai/.bin/go15beta1/go/src/testsharedlib/shlib (from $GOROOT)
/home1/tonybai/test/go/go15/src/testsharedlib/shlib (from $GOPATH)

go编译器无法找到shlib，也就说即便是动态链接，我们也要有动态共享库的源码，应用才能编译通过。

【internal package】

internal包不是go 1.5的原创，在go 1.4中就已经提出对internal package的支持了。但go 1.4发布时，internal package只能用于GOROOT下的go core核心包，用户层面GOPATH不支持internal package。按原计划，go 1.5中会将internal包机制工作范围全面扩大到所有repository的。我原以为1.5beta1以及将internal package机制生效了，但实际结果呢，我们来看看示例代码：

测试目录结构如下：

testinternal/src
    mypkg/
        /internal
            /foo
                foo.go
        /pkg1
            main.go

otherpkg/
main.go

按照internal包的原理，预期mypkg/pkg1下的代码是可以import "mypkg/internal/foo"的，otherpkg/下的代码是不能import "mypkg/internal/foo"的。

//foo.go
package foo

import "fmt"

func Foo() {
fmt.Println("mypkg/internal/foo")
}

//main.go
package main

import "mypkg/internal/foo"

func main() {
foo.Foo()
}

在pkg1和otherpkg下分别run main.go：

mypkg/pkg1$ go run main.go
mypkg/internal/foo

otherpkg$ go run main.go
mypkg/internal/foo

可以看到在otherpkg下执行时，并没有任何build error出现。看来internal机制并未生效。

我们再来试试import $GOROOT下某些internal包，看看是否可以成功：

package main

import (
"fmt"
"image/internal/imageutil"
)

func main() {
fmt.Println(imageutil.DrawYCbCr)
}

我们run这个代码：

$go run main.go
0x6b7f0

同样没有出现任何error。

不是很清楚为何在1.5beta1中internal依旧无效。难道非要等最终1.5 release版么？

【Vendor】
Vendor机制是go team为了解决go第三方包依赖和管理而引入的实验性技术。你执行以下go env：

$go env
GOARCH="amd64"
GOBIN="/Users/tony/.bin/go15beta1/go/bin"
GOEXE=""
GOHOSTARCH="amd64"
GOHOSTOS="darwin"
GOOS="darwin"
GOPATH="/Users/tony/Test/GoToolsProjects"
GORACE=""
GOROOT="/Users/tony/.bin/go15beta1/go"
GOTOOLDIR="/Users/tony/.bin/go15beta1/go/pkg/tool/darwin_amd64"
GO15VENDOREXPERIMENT=""
CC="clang"
GOGCCFLAGS="-fPIC -m64 -pthread -fno-caret-diagnostics -Qunused-arguments -fmessage-length=0 -fno-common"
CXX="clang++"
CGO_ENABLED="1"

从结果中你会看到新增一个GO15VENDOREXPERIMENT变量，这个就是用来控制vendor机制是否开启的环境变量，默认不开启。若要开启，可以在环境变量文件中设置或export GO15VENDOREXPERIMENT=1临时设置。

vendor机制是在go 1.5beta1发布前不长时间临时决定加入到go 1.5中的，Russ Cox在Keith Rarick之前的一个Proposal的基础上重新做了设计而成，大致机制内容：

If there is a source directory d/vendor, then,
when compiling a source file within the subtree rooted at d,
import "p" is interpreted as import "d/vendor/p" if that exists.

When there are multiple possible resolutions,
the most specific (longest) path wins.

The short form must always be used: no import path can
contain “/vendor/” explicitly.

Import comments are ignored in vendored packages.

下面我们来测试一下这个机制。首先我们临时开启vendor机制，export GO15VENDOREXPERIMENT=1，我们的测试目录规划如下：

testvendor
    vendor/
        tonybai.com/
            foolib/
                foo.go
    main/
        main.go

$GOPATH/src/tonybai.com/foolib/foo.go

//vendor/tonybai.com/foolib/foo.go
package foo

import "fmt"

func Hello() {
fmt.Println("foo in vendor")
}

//$GOPATH/src/tonybai.com/foolib/foo.go
package foo

import "fmt"

func Hello() {
fmt.Println("foo in gopath")
}

vendor和gopath下的foo.go稍有不同，主要在输出内容上，以方便后续区分。

现在我们编译执行main.go

//main/main.go
package main

import (
"tonybai.com/foolib"
)

func main() {
foo.Hello()
}

$go run main.go
foo in gopath

显然结果与预期不符，我们通过go list -json来看main.go的依赖包路径：

$go list -json
{
… …
    "Imports": [
        "tonybai.com/foolib"
    ],
    "Deps": [
        "errors",
        "fmt",
        "io",
        "math",
        "os",
        "reflect",
        "runtime",
        "strconv",
        "sync",
        "sync/atomic",
        "syscall",
        "time",
        "tonybai.com/foolib",
        "unicode/utf8",
        "unsafe"
    ]
}

可以看出并没有看到vendor路径，main.go import的是$GOPATH下的foo。难道是go 1.5beta1的Bug？于是翻看各种资料，最后在go 1.5beta1发布前最后提交的revison的commit log中得到了帮助：

cmd/go: disable vendoredImportPath for code outside $GOPATH
It was crashing.
This fixes the build for
GO15VENDOREXPERIMENT=1 go test -short runtime

Fixes #11416.

Change-Id: I74a9114cdd8ebafcc9d2a6f40bf500db19c6e825
Reviewed-on: https://go-review.googlesource.com/11964
Reviewed-by: Russ Cox <rsc@golang.org>

从commit log来看，大致意思是$GOPATH之外的代码的vendor机制被disable了（因为某个bug）。也就是说只有$GOPATH路径下的包在 import时才会考虑vendor路径，我们的代码的确没有在$GOPATH下，我们重新设置一下$GOPATH。

$export GOPATH=~/test/go/go15
[tony@TonydeMacBook-Air-2 ~/test/go/go15/src/testvendor/main]$go list -json
{

… …
    "Imports": [
        "testvendor/vendor/tonybai.com/foolib"
    ],
    "Deps": [
        "errors",
        "fmt",
        "io",
        "math",
        "os",
        "reflect",
        "runtime",
        "strconv",
        "sync",
        "sync/atomic",
        "syscall",
        "testvendor/vendor/tonybai.com/foolib",
        "time",
        "unicode/utf8",
        "unsafe"
    ]
}

这回可以看到vendor机制生效了。执行main.go:

$go run main.go
foo in vendor

这回与预期结果就相符了。

前面提到，关闭GOPATH外的vendor机制是因为一个bug，相信go 1.5正式版发布时，这块会被enable的。

三、小结

Go 1.5还增加了很多工具，如trace，但因文档不全，尚不知如何使用。

Go 1.5标准库也有很多小的变化，这个只有到使用时才能具体深入了解。

Go 1.5更多是Go语言骨子里的变化，也就是runtime和编译器重写。语法由于兼容Go 1，所以基本frozen，因此从外在看来，基本没啥变动了。

至于Go 1.5的性能，官方的说法是，有的程序用1.5编译后会变得慢点，有的会快些。官方bench的结果是总体比1.4快上一些。但Go 1.5在性能方面主要是为了减少gc延迟，后续版本才会在性能上做进一步优化，优化空间还较大的，这次runtime、编译器由c变go，很多地方的go 代码并非是最优的，多是自动翻译，相信经过Go team的优化后，更idiomatic的Go code会让Go整体性能更为优异。

标签 goroutine 下的文章

Go程序调试、分析与优化

Go 1.5中值得关注的几个变化

欢迎使用邮件订阅我的博客

文章

评论

分类

归档

链接

开源项目

翻译项目