Gosched()

runtime.Gosched()表示让CPU把时间片让给别人,下次某个时候继续恢复执行该goroutine

import ("fmt""runtime"
)func main() {go say("world")say("hello")
}func say(s string) {for i := 0; i < 5; i++ {runtime.Gosched()fmt.Println(s)}
}

输出：

hello
world
hello
world
hello
world
hello
world
hello

运行时希望多少个goroutine来同时地运行代码

查看

  println(runtime.Version())      // go1.4.1println(runtime.NumGoroutine()) // 2println(runtime.NumCPU())       // 4println(runtime.GOMAXPROCS(-1)) // 1

func init() {numcpu := runtime.NumCPU()runtime.GOMAXPROCS(numcpu) // 尝试使用所有可用的CPU
}

gc

disable gc

defer debug.SetGCPercent(debug.SetGCPercent(-1))

运行gc

runtime.GC()

查看gc信息

GODEBUG=gctrace=1 ./test_server

将gc信息保存到文件：

GODEBUG=gctrace=1 go run main.go 2> gctrace.log

可视化信息 https://github.com/davecheney/gcvis

GODEBUG=gctrace=1 ./test_server 2>&1 | gcvis

方法调用栈

堆栈信息

  go func() {fmt.Println("i am a goroutine")time.Sleep(time.Second)}()time.Sleep(500 * time.Millisecond)buf := make([]byte, 1024)n := runtime.Stack(buf, false)fmt.Println(string(buf[:n]))fmt.Println("===================")n = runtime.Stack(buf, true)fmt.Println(string(buf[:n]))

第一个输出：

goroutine 1 [running]:
main.main()/项目路径/src/Test/Test.go:18 +0xa5

第二个输出：

goroutine 1 [running]:
main.main()/项目路径/src/Test/Test.go:23 +0x2c2goroutine 17 [sleep]:
time.Sleep(0x3b9aca00)/usr/local/go/src/runtime/time.go:59 +0xf9
main.main.func1()/项目路径/src/Test/Test.go:12 +0xd9
created by main.main/项目路径/src/Test/Test.go:13 +0x37

---

data := debug.Stack()

输出：

/项目路径/src/test/test.go:17 (0x400c49)main: data := debug.Stack()
/go安装路径/src/runtime/proc.go:111 (0x42846f)main: main_main()
/go安装路径/src/runtime/asm_amd64.s:1696 (0x454471)goexit: BYTE    $0x90   // NOP

runtime.Caller 的用法

函数的签名如下:

func runtime.Caller(skip int) (pc uintptr, file string, line int, ok bool)

runtime.Caller 返回当前 goroutine 的栈上的函数调用信息. 主要有当前的 pc 值和调用的文件和行号等信息. 若无法获得信息, 返回的 ok 值为 false.

其输入参数 skip 为要跳过的栈帧数, 若为 0 则表示 runtime.Caller 的调用者.

注意:由于历史原因, runtime.Caller 和 runtime.Callers 中的 skip 含义并不相同, 后面会讲到.

下面是一个简单的例子, 打印函数调用的栈帧信息:

package mainimport ("fmt""runtime"
)func main() {fun1()
}func fun1() {for skip := 0; ; skip++ {pc, file, line, ok := runtime.Caller(skip)if !ok {break}fmt.Printf("skip = %v, pc = %v, file = %v, line = %v\n", skip, pc, file, line)}
}

输出结果：

skip = 0, pc = 8274, file = /项目路径/src/test/main.go, line = 14
skip = 1, pc = 8219, file = /项目路径/src/test/main.go, line = 9
skip = 2, pc = 77123, file = /usr/local/go/src/runtime/proc.go, line = 63
skip = 3, pc = 227809, file = /usr/local/go/src/runtime/asm_amd64.s, line = 2232

其中 skip = 0 为当前文件的 main.main 函数, 以及对应的行号.

另外的 skip = 1 和 skip = 2 也分别对应2个函数调用. 通过查阅 runtime/proc.c 文件的代码, 我们可以知道对应的函数分别为 runtime.main 和 runtime.goexit.

整理之后可以知道, Go的普通程序的启动顺序如下:

1. runtime.goexit 为真正的函数入口(并不是main.main)
2. 然后 runtime.goexit 调用 runtime.main 函数
3. 最终 runtime.main 调用用户编写的 main.main 函数

runtime.Callers 的用法

函数的签名如下:

func runtime.Callers(skip int, pc []uintptr) int

runtime.Callers 函数和 runtime.Caller 函数虽然名字相似(多一个后缀s), 但是函数的参数/返回值和参数的意义都有很大的差异.

runtime.Callers 把调用它的函数Go程栈上的程序计数器填入切片 pc 中. 参数 skip 为开始在 pc 中记录之前所要跳过的栈帧数, 若为0则表示 runtime.Callers 自身的栈帧, 若为1则表示调用者的栈帧. 该函数返回写入到 pc 切片中的项数(受切片的容量限制).

下面是 runtime.Callers 的例子, 用于输出每个栈帧的 pc 信息:

func main() {fun1()
}func fun1() {pc := make([]uintptr, 1024)for skip := 0; ; skip++ {n := runtime.Callers(skip, pc)if n <= 0 {break}fmt.Printf("skip = %v, pc = %v\n", skip, pc[:n])}
}

输出：

skip = 0, pc = [28854 8368 8219 77155 227841]
skip = 1, pc = [8368 8219 77155 227841]
skip = 2, pc = [8219 77155 227841]
skip = 3, pc = [77155 227841]
skip = 4, pc = [227841]

输出新的 pc 长度和 skip 大小有逆相关性. skip = 0 为 runtime.Callers 自身的信息.

这个例子比前一个例子多输出了一个栈帧, 就是因为多了一个runtime.Callers栈帧的信息(前一个例子是没有runtime.Caller信息的(注意:没有s后缀)).

那么 runtime.Callers 和 runtime.Caller 有哪些关联和差异?

runtime.Callers 和 runtime.Caller 的异同

因为前面2个例子为不同的程序, 输出的 pc 值并不具备参考性. 现在我们看看在同一个例子的输出结果如何:

package mainimport ("fmt""runtime"
)func main() {fun1()
}func fun1() {for skip := 0; ; skip++ {pc, file, line, ok := runtime.Caller(skip)if !ok {break}fmt.Printf("skip = %v, pc = %v, file = %v, line = %v\n", skip, pc, file, line)}pc := make([]uintptr, 1024)for skip := 0; ; skip++ {n := runtime.Callers(skip, pc)if n <= 0 {break}fmt.Printf("skip = %v, pc = %v\n", skip, pc[:n])}
}

输出：

skip = 0, pc = 8277, file = /项目路径/src/test/main.go, line = 14
skip = 1, pc = 8219, file = /项目路径/src/test/main.go, line = 9
skip = 2, pc = 78179, file = /usr/local/go/src/runtime/proc.go, line = 63
skip = 3, pc = 228865, file = /usr/local/go/src/runtime/asm_amd64.s, line = 2232
skip = 0, pc = [29878 8449 8219 78179 228865]
skip = 1, pc = [8449 8219 78179 228865]
skip = 2, pc = [8219 78179 228865]
skip = 3, pc = [78179 228865]
skip = 4, pc = [228865]

比如输出结果可以发现, 8219 78179 228865 这个 pc 值是相同的. 它们分别对应 main.main, runtime.main 和 runtime.goexit 函数.

runtime.Caller 输出的 8277 和 runtime.Callers 输出的 8449 并不相同. 这是因为, 这两个函数的调用位置并不相同, 因此导致了 pc 值也不完全相同.

最后就是 runtime.Callers 多输出一个 29878 值, 对应runtime.Callers内部的调用位置.

由于Go语言(Go1.2)采用分段堆栈, 因此不同的 pc 之间的大小关系并不明显.

runtime.FuncForPC 的用途

函数的签名如下:

func runtime.FuncForPC(pc uintptr) *runtime.Func
func (f *runtime.Func) FileLine(pc uintptr) (file string, line int)
func (f *runtime.Func) Entry() uintptr
func (f *runtime.Func) Name() string

其中 runtime.FuncForPC 返回包含给定 pc 地址的函数, 如果是无效 pc 则返回 nil .

runtime.Func.FileLine 返回与 pc 对应的源码文件名和行号. 安装文档的说明, 如果pc不在函数帧范围内, 则结果是不确定的.

runtime.Func.Entry 对应函数的地址. runtime.Func.Name 返回该函数的名称.

下面是 runtime.FuncForPC 的例子:

package mainimport ("fmt""runtime"
)func main() {fun1()
}func fun1() {for skip := 0; ; skip++ {pc, _, _, ok := runtime.Caller(skip)if !ok {break}p := runtime.FuncForPC(pc)file, line := p.FileLine(0)fmt.Printf("skip = %v, pc = %v\n", skip, pc)fmt.Printf("  file = %v, line = %d\n", file, line)fmt.Printf("  entry = %v\n", p.Entry())fmt.Printf("  name = %v\n", p.Name())}fmt.Println("-------------------------")pc := make([]uintptr, 1024)for skip := 0; ; skip++ {n := runtime.Callers(skip, pc)if n <= 0 {break}fmt.Printf("skip = %v, pc = %v\n", skip, pc[:n])for j := 0; j < n; j++ {p := runtime.FuncForPC(pc[j])file, line := p.FileLine(0)fmt.Printf("  skip = %v, pc = %v\n", skip, pc[j])fmt.Printf("    file = %v, line = %d\n", file, line)fmt.Printf("    entry = %v\n", p.Entry())fmt.Printf("    name = %v\n", p.Name())}break}
}

输出：

skip = 0, pc = 8277file = /项目路径/src/test/main.go, line = 12entry = 8224name = main.fun1
skip = 1, pc = 8219file = /项目路径/src/test/main.go, line = 8entry = 8192name = main.main
skip = 2, pc = 80579file = /usr/local/go/src/runtime/proc.go, line = 16entry = 80336name = runtime.main
skip = 3, pc = 231265file = /usr/local/go/src/runtime/asm_amd64.s, line = 2232entry = 231264name = runtime.goexit
-------------------------
skip = 0, pc = [32278 8634 8219 80579 231265]skip = 0, pc = 32278file = /usr/local/go/src/runtime/extern.go, line = 134entry = 32192name = runtime.Callersskip = 0, pc = 8634file = /项目路径/src/test/main.go, line = 12entry = 8224name = main.fun1skip = 0, pc = 8219file = /项目路径/src/test/main.go, line = 8entry = 8192name = main.mainskip = 0, pc = 80579file = /usr/local/go/src/runtime/proc.go, line = 16entry = 80336name = runtime.mainskip = 0, pc = 231265file = /usr/local/go/src/runtime/asm_amd64.s, line = 2232entry = 231264name = runtime.goexit

根据测试, 如果是无效 pc (比如0), runtime.Func.FileLine 一般会输出当前函数的开始行号. 不过在实践中, 一般会用 runtime.Caller 获取文件名和行号信息, runtime.Func.FileLine 很少用到(如何独立获取pc参数?).

定制的 CallerName 函数

基于前面的几个函数, 我们可以方便的定制一个 CallerName 函数. 函数 CallerName 返回调用者的函数名/文件名/行号等用户友好的信息.

函数实现如下:

package mainimport ("fmt""runtime"
)func main() {for skip := 0; ; skip++ {name, file, line, ok := CallerName(skip)if !ok {break}fmt.Printf("skip = %v\n", skip)fmt.Printf("  file = %v, line = %d\n", file, line)fmt.Printf("  name = %v\n", name)}
}func CallerName(skip int) (name, file string, line int, ok bool) {var pc uintptrif pc, file, line, ok = runtime.Caller(skip + 1); !ok {return}name = runtime.FuncForPC(pc).Name()return
}

输出：

skip = 0file = /Users/zhangyuchen/go/pro/src/test/main.go, line = 10name = main.main
skip = 1file = /usr/local/go/src/runtime/proc.go, line = 63name = runtime.main
skip = 2file = /usr/local/go/src/runtime/asm_amd64.s, line = 2232name = runtime.goexit

其中在执行 runtime.Caller 调用时, 参数 skip + 1 用于抵消 CallerName 函数自身的调用.

Go语言中函数的类型

在Go语言中, 除了语言定义的普通函数调用外, 还有闭包函数/init函数/全局变量初始化等不同的函数调用类型.

为了便于测试不同类型的函数调用, 我们包装一个 PrintCallerName 函数. 该函数用于输出调用者的信息.

package mainimport ("fmt""runtime"
)var a = PrintCallerName(0, "main.a")
var b = PrintCallerName(0, "main.b")func init() {a = PrintCallerName(0, "main.init.a")
}func init() {b = PrintCallerName(0, "main.init.b")func() {b = PrintCallerName(0, "main.init.b[1]")}()
}func main() {a = PrintCallerName(0, "main.main.a")b = PrintCallerName(0, "main.main.b")func() {b = PrintCallerName(0, "main.main.b[1]")func() {b = PrintCallerName(0, "main.main.b[1][1]")}()b = PrintCallerName(0, "main.main.b[2]")}()
}func PrintCallerName(skip int, comment string) bool {name, file, line, ok := CallerName(skip + 1)if !ok {return false}fmt.Printf("skip = %v, comment = %s\n", skip, comment)fmt.Printf("  file = %v, line = %d\n", file, line)fmt.Printf("  name = %v\n", name)return true
}func CallerName(skip int) (name, file string, line int, ok bool) {var pc uintptrif pc, file, line, ok = runtime.Caller(skip + 1); !ok {return}name = runtime.FuncForPC(pc).Name()return
}

输出：

skip = 0, comment = main.afile = /Users/zhangyuchen/go/pro/src/test/main.go, line = 8name = main.init
skip = 0, comment = main.bfile = /Users/zhangyuchen/go/pro/src/test/main.go, line = 9name = main.init
skip = 0, comment = main.init.afile = /Users/zhangyuchen/go/pro/src/test/main.go, line = 12name = main.init·1
skip = 0, comment = main.init.bfile = /Users/zhangyuchen/go/pro/src/test/main.go, line = 16name = main.init·2
skip = 0, comment = main.init.b[1]file = /Users/zhangyuchen/go/pro/src/test/main.go, line = 18name = main.func·001
skip = 0, comment = main.main.afile = /Users/zhangyuchen/go/pro/src/test/main.go, line = 23name = main.main
skip = 0, comment = main.main.bfile = /Users/zhangyuchen/go/pro/src/test/main.go, line = 24name = main.main
skip = 0, comment = main.main.b[1]file = /Users/zhangyuchen/go/pro/src/test/main.go, line = 26name = main.func·003
skip = 0, comment = main.main.b[1][1]file = /Users/zhangyuchen/go/pro/src/test/main.go, line = 28name = main.func·002
skip = 0, comment = main.main.b[2]file = /Users/zhangyuchen/go/pro/src/test/main.go, line = 30name = main.func·003

观察输出结果, 可以发现以下几个规律:

1. 全局变量的初始化调用者为 main.init 函数
2. 自定义的 init 函数有一个数字后缀, 根据出现的顺序进编号. 比如 main.init·1 和 main.init·2 等.
3. 闭包函数采用 main.func·001 格式命名, 安装闭包定义结束的位置顺序进编号.

比如以下全局变量的初始化调用者为 main.init 函数:

var a = PrintCallerName(0, "main.a")
var b = PrintCallerName(0, "main.b")

以下两个 init 函数根据出现顺序分别对应 main.init·1 和 main.init·2 :

func init() { // main.init·1//
}
func init() { // main.init·2//
}

以下三个闭包根据定义结束顺序分别为 001 / 002 / 003 :

func init() {func(){//}() // main.func·001
}func main() {func() {func(){//}() // main.func·002}() // main.func·003
}

因为, 这些特殊函数调用方式的存在, 我们需要进一步完善 CallerName 函数.

改进的 CallerName 函数

两类特殊的调用是 init 类函数调用 和 闭包函数调用.

改进后的 CallerName 函数对 init 类函数调用者统一处理为 init 函数. 将闭包函数调用这处理为调用者的函数名.

func CallerName(skip int) (name, file string, line int, ok bool) {var (reInit    = regexp.MustCompile(`init·\d+$`) // main.init·1reClosure = regexp.MustCompile(`func·\d+$`) // main.func·001)for {var pc uintptrif pc, file, line, ok = runtime.Caller(skip + 1); !ok {return}name = runtime.FuncForPC(pc).Name()if reInit.MatchString(name) {name = reInit.ReplaceAllString(name, "init")return}if reClosure.MatchString(name) {skip++continue}return}return
}

输出：

skip = 0, comment = main.afile = /Users/zhangyuchen/go/pro/src/test/main.go, line = 9name = main.init
skip = 0, comment = main.bfile = /Users/zhangyuchen/go/pro/src/test/main.go, line = 10name = main.init
skip = 0, comment = main.init.afile = /Users/zhangyuchen/go/pro/src/test/main.go, line = 13name = main.init
skip = 0, comment = main.init.bfile = /Users/zhangyuchen/go/pro/src/test/main.go, line = 17name = main.init
skip = 0, comment = main.init.b[1]file = /Users/zhangyuchen/go/pro/src/test/main.go, line = 20name = main.init
skip = 0, comment = main.main.afile = /Users/zhangyuchen/go/pro/src/test/main.go, line = 24name = main.main
skip = 0, comment = main.main.bfile = /Users/zhangyuchen/go/pro/src/test/main.go, line = 25name = main.main
skip = 0, comment = main.main.b[1]file = /Users/zhangyuchen/go/pro/src/test/main.go, line = 32name = main.main
skip = 0, comment = main.main.b[1][1]file = /Users/zhangyuchen/go/pro/src/test/main.go, line = 32name = main.main
skip = 0, comment = main.main.b[2]file = /Users/zhangyuchen/go/pro/src/test/main.go, line = 32name = main.main

CallerName 函数的不足之处

有以下的代码:

func init() {myInit("1")
}
func main() {myInit("2")
}var myInit = func(name string) {PrintCallerName(0, name+":main.myInit.b")
}

输出：

skip = 0, comment = 1:main.myInit.bfile = /Users/zhangyuchen/go/pro/src/test/main.go, line = 10name = main.init
skip = 0, comment = 2:main.myInit.bfile = /Users/zhangyuchen/go/pro/src/test/main.go, line = 13name = main.main

从直观上看, myInit闭包函数在执行时, 最好输出 main.myInit 函数名. 但是 main.myInit 只是一个绑定到闭包函数的变量, 而闭包的真正名字是 main.func·???(这里如果用改进之前的CallerName的话，输出是main.func·001). 在运行时是无法得到 main.myInit 这个名字的.

不同Go程序启动流程

基于函数调用者信息可以很容易的验证各种环境的程序启动流程.

test:

package mainimport ("fmt""testing"
)func TestPrintCallerName(t *testing.T) {for skip := 0; ; skip++ {name, file, line, ok := CallerName(skip)if !ok {break}fmt.Printf("skip = %v, name = %v, file = %v, line = %v\n", skip, name, file, line)}t.Fail()
}

example:

package mainimport (myMain ".""fmt"
)func Example() {for skip := 0; ; skip++ {name, file, line, ok := myMain.CallerName(skip)if !ok {break}fmt.Printf("skip = %v, name = %v, file = %v, line = %v\n", skip, name, file, line)}// Output: ?
}

运行 go test , 得到的输出：

=== RUN TestPrintCallerName
skip = 0, name = test.TestPrintCallerName, file = /Users/zhangyuchen/go/pro/src/test/main_test.go, line = 10
skip = 1, name = testing.tRunner, file = /usr/local/go/src/testing/testing.go, line = 447
skip = 2, name = runtime.goexit, file = /usr/local/go/src/runtime/asm_amd64.s, line = 2232
--- FAIL: TestPrintCallerName (0.00s)
=== RUN: Example
--- FAIL: Example (0.00s)
got:
skip = 0, name = test.Example, file = /Users/zhangyuchen/go/pro/src/test/example_test.go, line = 10
skip = 1, name = testing.runExample, file = /usr/local/go/src/testing/example.go, line = 98
skip = 2, name = testing.RunExamples, file = /usr/local/go/src/testing/example.go, line = 36
skip = 3, name = testing.(*M).Run, file = /usr/local/go/src/testing/testing.go, line = 486
skip = 4, name = main.main, file = test/_test/_testmain.go, line = 54
skip = 5, name = runtime.main, file = /usr/local/go/src/runtime/proc.go, line = 63
skip = 6, name = runtime.goexit, file = /usr/local/go/src/runtime/asm_amd64.s, line = 2232
want:
?
FAIL
exit status 1
FAIL    test    0.008s

分析输出数据我们可以发现, 测试代码和例子代码的启动流程和普通的程序流程都不太一样.

测试代码的启动流程:

1. runtime.goexit 还是入口
2. 但是 runtime.goexit 不在调用 runtime.main 函数, 而是调用 testing.tRunner 函数
3. testing.tRunner 函数由 go test 命令生成, 用于执行各个测试函数

例子代码的启动流程:

1. runtime.goexit 还是入口
2. 然后 runtime.goexit 调用 runtime.main 函数
3. 最终 runtime.main 调用go test 命令生成的 main.main 函数, 在 _test/_testmain.go 文件
4. 然后调用 testing.Main, 改函数执行各个例子函数

另外, 从这个例子我们可以发现, 我们自己写的 main.main 函数所在的 main 包也可以被其他包导入. 但是其他包导入之后的 main 包里的 main 函数就不再是main.main 函数了. 因此, 程序的入口也就不是自己写的 main.main 函数了.

内存使用情况

堆内存

    var m runtime.MemStatsruntime.ReadMemStats(&m)format := "%-40s : %d bytes\n"fmt.Printf(format, "bytes allocated and still in use", m.HeapAlloc)fmt.Printf(format, "bytes obtained from system", m.HeapSys)fmt.Printf(format, "bytes in idle spans", m.HeapIdle)fmt.Printf(format, "bytes in non-idle span", m.HeapInuse)fmt.Printf(format, "bytes released to the OS", m.HeapReleased)fmt.Printf(format, "total number of allocated objects", m.HeapObjects)

输出：

bytes allocated and still in use         : 38928 bytes
bytes obtained from system               : 851968 bytes
bytes in idle spans                      : 696320 bytes
bytes in non-idle span                   : 155648 bytes
bytes released to the OS                 : 0 bytes
total number of allocated objects        : 113 bytes

在web页面展示debug信息

  router := httprouter.New()router.HandlerFunc("GET", "/debug/pprof", pprof.Index)router.Handler("GET", "/debug/heap", pprof.Handler("heap"))router.Handler("GET", "/debug/goroutine", pprof.Handler("goroutine"))router.Handler("GET", "/debug/block", pprof.Handler("block"))router.Handler("GET", "/debug/threadcreate", pprof.Handler("threadcreate"))// 启动时的命令,比如 bin/debug -a=1router.HandlerFunc("GET", "/debug/pprof/cmdline", pprof.Cmdline)router.HandlerFunc("GET", "/debug/pprof/symbol", pprof.Symbol)router.HandlerFunc("GET", "/debug/pprof/profile", pprof.Profile)router.HandlerFunc("GET", "/debug/pprof/trace", pprof.Trace)http.ListenAndServe(":8080", router)