浅谈Golang内存逃逸

2024-04-02 19:55

短信预约 -IT技能 免费直播动态提醒

1.什么是内存逃逸

在一段程序中，每一个函数都会有自己的内存区域分配自己的局部变量，返回值，这些内存会由编译器在栈中进行分配，每一个函数会分配一个栈帧，在函数运行结束后销毁，但是有些变量我们想在函数运行结束后仍然使用，就需要把这个变量分配在堆上，这种从“栈”上逃逸到“堆”上的现象叫做内存逃逸

2.什么是逃逸分析

虽然Go语言引入的Gc，GC机制会对堆上的对象进行管理，当某个对象不可达(没有其他对象引用他)，他将会被回收。虽然GC可以降低工作人员负担，但是GC也会给程序带来性能损耗，当堆内存上有大量的堆内存对象，就会给GC很大的压力，虽然Go语言使用的是标记清除算法，并且在此基础上使用了三色标记法和写屏障技术，但是我们在堆上分配大量内存，仍然会对GC造成很大压力，Go引入了逃逸分析，就是想减少堆内存的分配，可以在栈分配的内存尽量分配在栈上

3.小结

逃逸分析就是在程序编译阶段根据代码中的数据流，对代码中哪些变量需要在栈上分配，哪些需要在对象分配的静态分析方法，堆和栈相比，堆适合分配不可预知大小的内存，但是付出代价是分配速度慢，容易产生碎片，栈分配十分快，栈分配只需要两个指令“Push”和"Release"分配和释放，而且堆分配需要先找一块适合大小的内存块分配，需要垃圾回收释放，所以逃逸分析可以更好的做内存分配

Go语言的逃逸分析

class="lazy" data-src/cmd/compile/internal/gc/escape.go

pointers to stack objects cannot be stored in the heap: 指向栈对象的指针不能存储在堆中
pointers to a stack object cannot outlive that object:指向栈对象的指针不能超过该对象的存活期，指针不能在栈对象销毁之后依然存活（例子：声明的函数返回并销毁了对象的栈帧，或者它在循环迭代中被重复用于逻辑上不同的变量）

既然逃逸分析是在编译阶段进行的，那我们就可以通过go build -gcflga '-m -m l'查看逃逸分析结果

4.逃逸分析案例

1.函数返回局部指针变量

func Add(x,y int) *int {
 res := 0
 res = x + y
 return &res
}
func main()  {
 Add(1,2)
}

.\pointer.go:4:2: res escapes to heap:
.\pointer.go:4:2: flow: ~r2 = &res:
.\pointer.go:4:2: from &res (address-of) at .\pointer.go:6:9
.\pointer.go:4:2: from return &res (return) at .\pointer.go:6:2
.\pointer.go:4:2: moved to heap: res

函数返回局部变量是一个指针变量，函数Add执行结束，对应栈帧就会销毁，但是引用返回到函数外部，如果我们外部解析地址，就会导致程序访问非法内存，所以经过编辑器分析过后将其在堆上分配

2.interface类型逃逸

1.interface产生逃逸

func main()  {
   str := "荔枝"
   fmt.Println(str)
}

E:\GoStudy\class="lazy" data-src\HighBase\Escape>go build -gcflags="-m -m -l" ./pointer.go
# command-line-arguments
.\pointer.go:20:13: str escapes to heap:
.\pointer.go:20:13: flow: {storage for ... argument} = &{storage for str}:
.\pointer.go:20:13: from str (spill) at .\pointer.go:20:13
.\pointer.go:20:13: from ... argument (slice-literal-element) at .\pointer.go:20:13
.\pointer.go:20:13: flow: {heap} = {storage for ... argument}:
.\pointer.go:20:13: from ... argument (spill) at .\pointer.go:20:13
.\pointer.go:20:13: from fmt.Println(... argument...) (call parameter) at .\pointer.go:20:13
.\pointer.go:20:13: ... argument does not escape
.\pointer.go:20:13: str escapes to heap

str是main的一个局部变量，传给 fmt.Printl()之后逃逸，因为fmt.Println()的入参是interface{}类型,如果参数为interface{}，那么编译期间就很难确定参数类型

2.指向栈对象的指针不能在堆中

我们把代码改成这样

func main()  {
   str := "苏珊"
   fmt.Println(&str)
}

# command-line-arguments
.\pointer.go:19:2: str escapes to heap:
.\pointer.go:19:2: flow: {storage for ... argument} = &str:
.\pointer.go:19:2: from &str (address-of) at .\pointer.go:20:14
.\pointer.go:19:2: from &str (interface-converted) at .\pointer.go:20:14
.\pointer.go:19:2: from ... argument (slice-literal-element) at .\pointer.go:20:13
.\pointer.go:19:2: flow: {heap} = {storage for ... argument}:
.\pointer.go:19:2: from ... argument (spill) at .\pointer.go:20:13
.\pointer.go:19:2: from fmt.Println(... argument...) (call parameter) at .\pointer.go:20:13
.\pointer.go:19:2: moved to heap: str
.\pointer.go:20:13: ... argument does not escape

这次str也逃逸到堆上面了，在堆上面进行分配，因为入参是interface，变量str的地址被以实参的方式传入fmt.Println被装箱到一个interface｛｝

装箱的形参变量要在堆上分配，但是还需要存储一个栈上的地址,这和之前说的第一条不符，所以str也会分配到堆上

3.闭包产生逃逸

func Increase() func() int {
 n := 0
 return func() int {
  n++
  return n
 }
}

func main() {
 in := Increase()
 fmt.Println(in()) // 1
}

E:\GoStudy\class="lazy" data-src\HighBase\Escape>go build -gcflags "-m -m -l" ./pointer.go
# command-line-arguments
.\pointer.go:27:2: Increase capturing by ref: n (addr=false assign=true width=8)
.\pointer.go:28:9: func literal escapes to heap:
.\pointer.go:28:9: flow: ~r0 = &{storage for func literal}:
.\pointer.go:28:9: from func literal (spill) at .\pointer.go:28:9
.\pointer.go:28:9: from return func literal (return) at .\pointer.go:28:2
.\pointer.go:27:2: n escapes to heap:
.\pointer.go:27:2: flow: {storage for func literal} = &n:
.\pointer.go:27:2: from n (captured by a closure) at .\pointer.go:29:3
.\pointer.go:27:2: from n (reference) at .\pointer.go:29:3
.\pointer.go:27:2: moved to heap: n
.\pointer.go:28:9: func literal escapes to heap
.\pointer.go:36:16: in() escapes to heap:
.\pointer.go:36:16: flow: {storage for ... argument} = &{storage for in()}:
.\pointer.go:36:16: from in() (spill) at .\pointer.go:36:16
.\pointer.go:36:16: from ... argument (slice-literal-element) at .\pointer.go:36:13
.\pointer.go:36:16: flow: {heap} = {storage for ... argument}:
.\pointer.go:36:16: from ... argument (spill) at .\pointer.go:36:13
.\pointer.go:36:16: from fmt.Println(... argument...) (call parameter) at .\pointer.go:36:13
.\pointer.go:36:13: ... argument does not escape
.\pointer.go:36:16: in() escapes to heap

因为函数是指针类型，所以匿名函数当做返回值产生逃逸，匿名函数使用外部变量n,这个n会一直存在知道in被销毁

4. 变量大小不确定及栈空间不足引发逃逸

import (
    "math/rand"
)

func LessThan8192()  {
    nums := make([]int, 100) // = 64KB
    for i := 0; i < len(nums); i++ {
        nums[i] = rand.Int()
    }
}


func MoreThan8192(){
    nums := make([]int, 1000000) // = 64KB
    for i := 0; i < len(nums); i++ {
        nums[i] = rand.Int()
    }
}


func NonConstant() {
    number := 10
    s := make([]int, number)
    for i := 0; i < len(s); i++ {
        s[i] = i
    }
}

func main() {
    NonConstant()
    MoreThan8192()
    LessThan8192()
}

# command-line-arguments
.\pointer.go:43:14: make([]int, 100) does not escape
.\pointer.go:51:14: make([]int, 1000000) escapes to heap:
.\pointer.go:51:14: flow: {heap} = &{storage for make([]int, 1000000)}:
.\pointer.go:51:14: from make([]int, 1000000) (too large for stack) at .\pointer.go:51:14
.\pointer.go:51:14: make([]int, 1000000) escapes to heap
.\pointer.go:60:11: make([]int, number) escapes to heap:
.\pointer.go:60:11: flow: {heap} = &{storage for make([]int, number)}:
.\pointer.go:60:11: from make([]int, number) (non-constant size) at .\pointer.go:60:11
.\pointer.go:60:11: make([]int, number) escapes to heap

栈空间足够不会发生逃逸，但是变量过大，已经超过栈空间，会逃逸到堆上