码迷,mamicode.com
首页 > 其他好文 > 详细

golang map 内幕

时间:2020-07-13 13:33:53      阅读:54      评论:0      收藏:0      [点我收藏+]

标签:round   for loop   基本   执行流程   lex   ups   currently   cas   函数   

关键性数据结构

  • hmap: map 的 header结构
  • bmap: map 的 bucket结构
  • mapextra: map 的 拓展结构 不是每一个map都包含

golang map 是用 hash map实现的,首先,我们先看 hash map是怎么实现的;然后我们再看 golang map 是怎么基于 hash map 封装的 map 类型。

Bucket

// A bucket for a Go map.
type bmap struct {
	// tophash generally contains the top byte of the hash value
	// for each key in this bucket. If tophash[0] < minTopHash,
	// tophash[0] is a bucket evacuation state instead.
	tophash [bucketCnt]uint8
	// Followed by bucketCnt keys and then bucketCnt elems.
	// NOTE: packing all the keys together and then all the elems together makes the
	// code a bit more complicated than alternating key/elem/key/elem/... but it allows
	// us to eliminate padding which would be needed for, e.g., map[int64]int8.
	// Followed by an overflow pointer.
}

这里面的 bucketCnt 是一个常量:

const (
    // Maximum number of key/elem pairs a bucket can hold.
    bucketCntBits = 3
    bucketCnt     = 1 << bucketCntBits
    
    ... ...
)

这个topHash 就是指hash值的前8位, 每一个桶是大小为 8*8 大小。此外topHash 的元素还可能有以下状态:

const(
    emptyRest      = 0 // 此位置未被占用,且后面的位置也没被占用
	emptyOne       = 1 // 此位置未被占用,在delete后会设置此标志
	evacuatedX     = 2 // 此位置已经被占用,对应数据已经被迁移到新的buckets的first半区
	evacuatedY     = 3 // 此位置已经被占用,对应数据已经被迁移到新的buckets的second半区
	evacuatedEmpty = 4 // 此位置未被占用,但bucket已经被迁移
	minTopHash     = 5 // topHash的最小值,为了与前面4个值进行区分
)

Hmap

// A header for a Go map.
type hmap struct {
	// Note: the format of the hmap is also encoded in cmd/compile/internal/gc/reflect.go.
	// Make sure this stays in sync with the compiler‘s definition.
	count     int // # live cells == size of map.  Must be first (used by len() builtin)
	flags     uint8
	B         uint8  // log_2 of # of buckets (can hold up to loadFactor * 2^B items)
	noverflow uint16 // overflow buckets 的计数器;详细信息看 incrnoverflow 
	hash0     uint32 // hash seed

	buckets    unsafe.Pointer // array of 2^B Buckets. may be nil if count==0.
	oldbuckets unsafe.Pointer // previous bucket array of half the size, non-nil only when growing
	nevacuate  uintptr        // progress counter for evacuation (buckets less than this have been evacuated), 已迁移的bucket的计数器

	extra *mapextra // optional fields,记录next_overflow的地址,和已经分配的overflow
}

const (
    // flags
	iterator     = 1 // there may be an iterator using buckets,有一个iter在使用bucket
	oldIterator  = 2 // there may be an iterator using oldbuckets,有一个iter在使用old_bucket
	hashWriting  = 4 // a goroutine is writing to the map,有一个协程在写入
	sameSizeGrow = 8 // the current map growth is to a new map of the same size,新bucekt与old_bucket size相同.
)


// mapextra holds fields that are not present on all maps.
type mapextra struct {
	// If both key and elem do not contain pointers and are inline, then we mark bucket
	// type as containing no pointers. This avoids scanning such maps.
	// However, bmap.overflow is a pointer. In order to keep overflow buckets
	// alive, we store pointers to all overflow buckets in hmap.extra.overflow and hmap.extra.oldoverflow.
	// overflow and oldoverflow are only used if key and elem do not contain pointers.
	// overflow contains overflow buckets for hmap.buckets.
	// oldoverflow contains overflow buckets for hmap.oldbuckets.
	// The indirection allows to store a pointer to the slice in hiter.
	overflow    *[]*bmap
	oldoverflow *[]*bmap

	// nextOverflow holds a pointer to a free overflow bucket.
	nextOverflow *bmap
}

map解析

make map

在golang中可以通过 make(map[key]value, hint) 创建一个map实例,在runtime包中是通过如下函数实现的:

// makemap implements Go map creation for make(map[k]v, hint).
// If the compiler has determined that the map or the first bucket
// can be created on the stack, h and/or bucket may be non-nil.
// If h != nil, the map can be created directly in h.
// If h.buckets != nil, bucket pointed to can be used as the first bucket.
func makemap(t *maptype, hint int, h *hmap) *hmap {
    mem, overflow := math.MulUintptr(uintptr(hint), t.bucket.size)
	if overflow || mem > maxAlloc {
		hint = 0
	}

	// 1. 初始化hmap
	if h == nil {
		h = new(hmap)
	}
	h.hash0 = fastrand() // 取一个随机值作为hash seed

	// Find the size parameter B which will hold the requested # of elements.
	// For hint < 0 overLoadFactor returns false since hint < bucketCnt.
	B := uint8(0)
	for overLoadFactor(hint, B) {
		B++
	}
	h.B = B
    
    // allocate initial hash table
	// if B == 0, the buckets field is allocated lazily later (in mapassign)
	// If hint is large zeroing this memory could take a while.
	// 分配空间
	if h.B != 0 {
		var nextOverflow *bmap
		h.buckets, nextOverflow = makeBucketArray(t, h.B, nil)
		if nextOverflow != nil {
			h.extra = new(mapextra)
			h.extra.nextOverflow = nextOverflow // 保存overflow区域的首地址
		}
	}
	
}


// makeBucketArray initializes a backing array for map buckets.
// 1<<b is the minimum number of buckets to allocate.
// dirtyalloc should either be nil or a bucket array previously
// allocated by makeBucketArray with the same t and b parameters.
// If dirtyalloc is nil a new backing array will be alloced and
// otherwise dirtyalloc will be cleared and reused as backing array.
func makeBucketArray(t *maptype, b uint8, dirtyalloc unsafe.Pointer) (buckets unsafe.Pointer, nextOverflow *bmap) {
    // 1. 计算要使用的空间大小
	base := bucketShift(b)  // base = 2^(b‘), b‘ 32位系统时为b的低5位,最大值为31,64系统时为b的低6位,最大值为63
	nbuckets := base
	// For small b, overflow buckets are unlikely.
	// Avoid the overhead of the calculation.
	// 预分配内存
	if b >= 4 {
		// Add on the estimated number of overflow buckets
		// required to insert the median number of elements
		// used with this value of b.
		nbuckets += bucketShift(b - 4) // nbuckets = base + 2^((b-4)‘) 
		sz := t.bucket.size * nbuckets
		up := roundupsize(sz) // 内存页对齐,golang 中pagesize 为8kB
		if up != sz {
			nbuckets = up / t.bucket.size
		}
	}
    
    // 2. 空间重复利用
	if dirtyalloc == nil { 
	    // 没有脏空间,则新分配一个数组
		buckets = newarray(t.bucket, int(nbuckets))
	} else {
		// dirtyalloc was previously generated by
		// the above newarray(t.bucket, int(nbuckets))
		// but may not be empty.
		buckets = dirtyalloc
		size := t.bucket.size * nbuckets
		if t.bucket.ptrdata != 0 {
			memclrHasPointers(buckets, size)
		} else {
			memclrNoHeapPointers(buckets, size)
		}
	}
    
    // 3. 计算overflow 指针
	if base != nbuckets {
		// We preallocated some overflow buckets.
		// To keep the overhead of tracking these overflow buckets to a minimum,
		// we use the convention that if a preallocated overflow bucket‘s overflow
		// pointer is nil, then there are more available by bumping the pointer.
		// We need a safe non-nil pointer for the last overflow bucket; just use buckets.
		nextOverflow = (*bmap)(add(buckets, base*uintptr(t.bucketsize)))
		last := (*bmap)(add(buckets, (nbuckets-1)*uintptr(t.bucketsize)))
		last.setoverflow(t, (*bmap)(buckets)) // 将overflow最后一个bucket的末尾位置存储 buckets指针
	}
	return buckets, nextOverflow
}

// overLoadFactor reports whether count items placed in 1<<B buckets is over loadFactor.
// 检查当前数量是不是超过负载系数
func overLoadFactor(count int, B uint8) bool {
	return count > bucketCnt && uintptr(count) > loadFactorNum*(bucketShift(B)/loadFactorDen)
}

返回值类型是 hmap*, hmap.B 要符合以下规则

    loadFactorNum = 13
    loadFactorDen = 2
    count = hint
    uintptr(count) > loadFactorNum*(bucketShift(B)/loadFactorDen)

hmap.buckets 分配的内存: roudupsize(base=bucketsize*2^B + overflow=bucketsize*/2^B/4),bucketsize = bitmap_size + 8*(key_size) + 8*(elem_size) + ptr_size,在32和64位系统中ptr_size不同,所以bucket_size会也不同。
从上面的代码可以看出,在物理内存中是以一个数组来存储hmap table,内存分布大致如下:

# bucket
|bmap|key1~8|elem1~8|

#bucekts
|bucket1~N|overflow|

# overflow
|nextOverflow|...|last|

# last
|bmap|...|ptr_buckets|

bucket1~N 是base区域, overflow 是预留区域,降低内存重复分配的次数。

insert into map

// Like mapaccess, but allocates a slot for the key if it is not present in the map.
func mapassign(t *maptype, h *hmap, key unsafe.Pointer) unsafe.Pointer {
    ... ...
    // 1. 生成key的hash
    alg := t.key.alg
	hash := alg.hash(key, uintptr(h.hash0))
	
	if h.buckets == nil {
		h.buckets = newobject(t.bucket) // newarray(t.bucket, 1)
	}

again:
    // 2. 分桶
    // 取出hash的低 B 位,作为 bucket的编号
	bucket := hash & bucketMask(h.B)
	if h.growing() {
		growWork(t, h, bucket)
	}
	
	// 找到桶对应的数组首地址
	b := (*bmap)(unsafe.Pointer(uintptr(h.buckets) + bucket*uintptr(t.bucketsize)))
	// hash 的高 8 位, 如果小于5 则 +5 返回
	top := tophash(hash)

	var inserti *uint8 // bmap中写入 tophash 的地址
	var insertk unsafe.Pointer // 写入key的地址
	var elem unsafe.Pointer // 写入值的地址
bucketloop:
	for {
		for i := uintptr(0); i < bucketCnt; i++ { // 查找可使用的位置
			if b.tophash[i] != top {
				if isEmpty(b.tophash[i]) && inserti == nil {
					inserti = &b.tophash[i]
					insertk = add(unsafe.Pointer(b), dataOffset+i*uintptr(t.keysize))
					elem = add(unsafe.Pointer(b), dataOffset+bucketCnt*uintptr(t.keysize)+i*uintptr(t.elemsize))
				}
				if b.tophash[i] == emptyRest {  // 找到可使用的位置,跳出循环
					break bucketloop
				}
				continue
			}
			// 找到相同topHash,比较key
			k := add(unsafe.Pointer(b), dataOffset+i*uintptr(t.keysize))
			if t.indirectkey() {
				k = *((*unsafe.Pointer)(k))
			}
			if !alg.equal(key, k) { // key不相同继续查找
				continue
			}
			// already have a mapping for key. Update it.
			// key相同,找到elem位置,跳转到done
			if t.needkeyupdate() {
				typedmemmove(t.key, k, key) 
			}
			elem = add(unsafe.Pointer(b), dataOffset+bucketCnt*uintptr(t.keysize)+i*uintptr(t.elemsize))
			goto done
		}
		
		// 在N bucket 中没有找到,查找overflow
		ovf := b.overflow(t)
		if ovf == nil {
			break // overflow 为空跳出循环
		}
		b = ovf
	}

	// Did not find mapping for key. Allocate new cell & add entry.

	// If we hit the max load factor or we have too many overflow buckets,
	// and we‘re not already in the middle of growing, start growing.
	// 是否触发内存增长
	if !h.growing() && (overLoadFactor(h.count+1, h.B) || tooManyOverflowBuckets(h.noverflow, h.B)) {
		hashGrow(t, h)
		goto again // Growing the table invalidates everything, so try again
	}

	if inserti == nil { // 没有找到可以插入的位置,新分配一个overflow
		// all current buckets are full, allocate a new one.
		newb := h.newoverflow(t, b)
		inserti = &newb.tophash[0]
		insertk = add(unsafe.Pointer(newb), dataOffset)
		elem = add(insertk, bucketCnt*uintptr(t.keysize))
	}

	// store new key/elem at insert position
	if t.indirectkey() {
		kmem := newobject(t.key)
		*(*unsafe.Pointer)(insertk) = kmem
		insertk = kmem
	}
	if t.indirectelem() {
		vmem := newobject(t.elem)
		*(*unsafe.Pointer)(elem) = vmem
	}
	typedmemmove(t.key, insertk, key)
	*inserti = top
	h.count++

done:
	if h.flags&hashWriting == 0 {
		throw("concurrent map writes")
	}
	h.flags &^= hashWriting
	if t.indirectelem() {
		elem = *((*unsafe.Pointer)(elem))
	}
	return elem	
	
}

从上面代码解析我们能清楚一个写入的过程:

  1. 对key做hash,取hash的低B位,确定bucket的编号 N;
  2. 遍历 N bucket中每一个 位置 ,找到没有写入的 位置,写入topHash即hash的高8位;
  3. 如果 N bucket 中有相同的 topHash,则需要去出对应的key做比较,如果相同则修改elem,如果不同则继续向后遍历,寻找空闲的 位置 写入,以此来解决冲突的问题。

那如果出现N bucket 满了怎么办?虽然这种概率很低但是也难免会遇到,毕竟一个bmap中只能装的下8个key,这里就要用到我们刚才说的预分配内存 overflow,分配overflow的逻辑如下:

func (h *hmap) newoverflow(t *maptype, b *bmap) *bmap {
	var ovf *bmap
	if h.extra != nil && h.extra.nextOverflow != nil { // 存在overflow区域
		// We have preallocated overflow buckets available.
		// See makeBucketArray for more details.
		ovf = h.extra.nextOverflow
		if ovf.overflow(t) == nil { // ovf不是最后一块,将nextOverflow 指针向后移动
			// We‘re not at the end of the preallocated overflow buckets. Bump the pointer.
			h.extra.nextOverflow = (*bmap)(add(unsafe.Pointer(ovf), uintptr(t.bucketsize)))
		} else { // 最后一块 overflow,将nextOverflow 置空
			// This is the last preallocated overflow bucket.
			// Reset the overflow pointer on this bucket,
			// which was set to a non-nil sentinel value.
			ovf.setoverflow(t, nil) // 抹平末尾指针
			h.extra.nextOverflow = nil
		}
	} else {// overflow 用光,新建一块
		ovf = (*bmap)(newobject(t.bucket)) 
	}
	h.incrnoverflow() // 增加overflow 使用计数
	if t.bucket.ptrdata == 0 {
		h.createOverflow()
		*h.extra.overflow = append(*h.extra.overflow, ovf)
	}
	b.setoverflow(t, ovf) // 将overflow的地址加到末尾
	return ovf
}

通过上面的函数我们可以明白,当N bucket被用光后如何扩充,即从预留区域查找一块未使用的区域,将该区域的指针放在 N bucket 的末尾,作为 N bucket的扩充区域:

# N bucket 扩充
|bmap|key1~8|elem1~8|ptr_extra_bucket|......|extra_bucket|
                            ^---------------^

然后我们就可以愉快的将新的key-value放到extra_bucket 中,我们在结合上面一节的内容可以更加清晰的明白hashmap table在内存里的构造,即连续数组+跳转指针,这也是为什么访问能如此快速的原因,基本上都是在连续内存上指针位移操作。

看到这里我们对map的实现应该有一个粗浅的认识,但是这只是一部分。我们可以注意到hmap中还有一个oldbuckets,还有其他的成员变量的含义没有被解开。

在上面的段落中我们给出了bucket溢出的解决方案,但如果bucket溢出过多怎么办(即单桶数据过多)??我们可以假设一个极端情况 一个map中除了 bucket1,剩余的都是它的extra_bucket,当我在 bucket1中取查询的时候要遍历很长,最差的情况要遍历整个map。怎样去解决这个问题?

  • 对key进行排序,使key变成有序的。可以使用插入排序,实现简单但是内存位移的操作较多。使用二分查找时间复杂度可以优化到log(n)。当然也可以使用其它排序算法,来减少内存位移发生的几率,但是因为底层存储是使用的数组,内存位移难以避免。
  • 重建,通过重新划分桶,来解决单桶数据过多问题。

上面代码中不难展示出在golang中使用的是第二种方案,那么什么情况下会触发重建?如何重建?

map Grow

tooManyOverflowBucketsoverLoadFactor 这两个函数会去判断是否需要执行 Grow 流程:

func mapassign(t *maptype, h *hmap, key unsafe.Pointer) unsafe.Pointer {
    ... ...
    // Did not find mapping for key. Allocate new cell & add entry.

	// If we hit the max load factor or we have too many overflow buckets,
	// and we‘re not already in the middle of growing, start growing.
	if !h.growing() && (overLoadFactor(h.count+1, h.B) || tooManyOverflowBuckets(h.noverflow, h.B)) {
		hashGrow(t, h)
		goto again // Growing the table invalidates everything, so try again
	}
	... ...
}

const(
    loadFactorNum = 13
    loadFactorDen = 2
)

// overLoadFactor reports whether count items placed in 1<<B buckets is over loadFactor.
func overLoadFactor(count int, B uint8) bool {
	return count > bucketCnt && uintptr(count) > loadFactorNum*(bucketShift(B)/loadFactorDen)
}

// tooManyOverflowBuckets reports whether noverflow buckets is too many for a map with 1<<B buckets.
// Note that most of these overflow buckets must be in sparse use;
// if use was dense, then we‘d have already triggered regular map growth.
func tooManyOverflowBuckets(noverflow uint16, B uint8) bool {
	// If the threshold is too low, we do extraneous work.
	// If the threshold is too high, maps that grow and shrink can hold on to lots of unused memory.
	// "too many" means (approximately) as many overflow buckets as regular buckets.
	// See incrnoverflow for more details.
	if B > 15 {
		B = 15
	}
	// The compiler doesn‘t see here that B < 16; mask B to generate shorter shift code.
	return noverflow >= uint16(1)<<(B&15)
}

首先是overLoadFactor,当 key_count + 1 > 8 and key_count+1 > 13*2^(B-1) 时会触发 Grow 流程。其次是 tooManyOverflowBuckets,当noverflow >= 2^(B&15),也就是说overflow bucket的数量大于2^(B&15)就会触发Grow,这里分两种情况:

  1. B < 16,此时当overflow bucketbase bucket 相等的时候就会触发增长流程。
  2. B >= 16 时,此时当overflow bucket数量 > 2^15时就会触发增长流程。

这里需要注意的是,noverflow 并不是一个准确的计数,当数量过大的时候它只能显示一个近似的数量:

// incrnoverflow increments h.noverflow.
// noverflow counts the number of overflow buckets.
// This is used to trigger same-size map growth.
// See also tooManyOverflowBuckets.
// To keep hmap small, noverflow is a uint16.
// When there are few buckets, noverflow is an exact count.
// When there are many buckets, noverflow is an approximate count.
func (h *hmap) incrnoverflow() {
	// We trigger same-size map growth if there are
	// as many overflow buckets as buckets.
	// We need to be able to count to 1<<h.B.
	if h.B < 16 {
		h.noverflow++
		return
	}
	// Increment with probability 1/(1<<(h.B-15)).
	// When we reach 1<<15 - 1, we will have approximately
	// as many overflow buckets as buckets.
	mask := uint32(1)<<(h.B-15) - 1
	// Example: if h.B == 18, then mask == 7,
	// and fastrand & 7 == 0 with probability 1/8.
	if fastrand()&mask == 0 {
		h.noverflow++
	}
}

可以很明显的看出,当h.B>=16时候并不是每次都会累加,此时base bucket的数量至少为2^16,可以存储2^19=524288条数据,外加预分配的overflow 。桶的数量越多,分布的越离散,出现overflow的概率更低,即使出现overflow单桶过长的概率也会降低。

func hashGrow(t *maptype, h *hmap) {
	// If we‘ve hit the load factor, get bigger.
	// Otherwise, there are too many overflow buckets,
	// so keep the same number of buckets and "grow" laterally.
	bigger := uint8(1)
	// 判断是否超出loadFactor
	if !overLoadFactor(h.count+1, h.B) { // 没有超过则维持以前的size
		bigger = 0
		h.flags |= sameSizeGrow
	}
	oldbuckets := h.buckets
	newbuckets, nextOverflow := makeBucketArray(t, h.B+bigger, nil)
    
    // 更新flags
	flags := h.flags &^ (iterator | oldIterator)
	if h.flags&iterator != 0 {
		flags |= oldIterator
	}
	// commit the grow (atomic wrt gc)
	h.B += bigger
	h.flags = flags
	// 替换buckets
	h.oldbuckets = oldbuckets
	h.buckets = newbuckets
	h.nevacuate = 0
	h.noverflow = 0

	if h.extra != nil && h.extra.overflow != nil {
		// Promote current overflow buckets to the old generation.
		if h.extra.oldoverflow != nil {
			throw("oldoverflow is not nil")
		}
		h.extra.oldoverflow = h.extra.overflow
		h.extra.overflow = nil
	}
	if nextOverflow != nil {
		if h.extra == nil {
			h.extra = new(mapextra)
		}
		h.extra.nextOverflow = nextOverflow
	}

	// the actual copying of the hash table data is done incrementally
	// by growWork() and evacuate().
}

根据上面的hashGrow 函数,我们可以看出map内存增长的规则,在overLoadFactor的情况下,h.B = h.B + 1,即base bucket数量翻倍, 否则维持原size不变,创建一个新的 buckets数组。

func mapassign(t *maptype, h *hmap, key unsafe.Pointer) unsafe.Pointer{
    ... ...
    bucket := hash & bucketMask(h.B)
    if h.growing() {
		growWork(t, h, bucket)
	}
	... ...
}


func growWork(t *maptype, h *hmap, bucket uintptr) {
	// make sure we evacuate the oldbucket corresponding
	// to the bucket we‘re about to use
	
	evacuate(t, h, bucket&h.oldbucketmask())

	// evacuate one more oldbucket to make progress on growing
	if h.growing() {
		evacuate(t, h, h.nevacuate)
	}
}

// 迁移数据
func evacuate(t *maptype, h *hmap, oldbucket uintptr) {
	b := (*bmap)(add(h.oldbuckets, oldbucket*uintptr(t.bucketsize)))
	newbit := h.noldbuckets() //oldbuckets 的 size
	if !evacuated(b) {
		// TODO: reuse overflow buckets instead of using new ones, if there
		// is no iterator using the old buckets.  (If !oldIterator.)

		// xy contains the x and y (low and high) evacuation destinations.
		// 1. 确定两个潜在目的迁移地址,X/Y
		// X 地址为 new buckets中编号为 oldbucket 的bucket
		var xy [2]evacDst
		x := &xy[0]
		x.b = (*bmap)(add(h.buckets, oldbucket*uintptr(t.bucketsize)))
		x.k = add(unsafe.Pointer(x.b), dataOffset)
		x.e = add(x.k, bucketCnt*uintptr(t.keysize))
        
		if !h.sameSizeGrow() { // size 不变的情况下
			// Only calculate y pointers if we‘re growing bigger.
			// Otherwise GC can see bad pointers.
			// Y 地址为 new buckets中编号为 oldbucket + noldbuckets, 即second 半区
			y := &xy[1]
			y.b = (*bmap)(add(h.buckets, (oldbucket+newbit)*uintptr(t.bucketsize)))
			y.k = add(unsafe.Pointer(y.b), dataOffset)
			y.e = add(y.k, bucketCnt*uintptr(t.keysize))
		}
        
        // 2. 开始迁移数据,包括bucket overflow部分的数据
		for ; b != nil; b = b.overflow(t) {
			k := add(unsafe.Pointer(b), dataOffset)
			e := add(k, bucketCnt*uintptr(t.keysize))
			for i := 0; i < bucketCnt; i, k, e = i+1, add(k, uintptr(t.keysize)), add(e, uintptr(t.elemsize)) {
				top := b.tophash[i]
				if isEmpty(top) {
					b.tophash[i] = evacuatedEmpty
					continue
				}
				if top < minTopHash {
					throw("bad map state")
				}
				k2 := k
				if t.indirectkey() {
					k2 = *((*unsafe.Pointer)(k2))
				}
				var useY uint8
				 // 计算迁移到first 半区还是 second半区
				if !h.sameSizeGrow() {
					// Compute hash to make our evacuation decision (whether we need
					// to send this key/elem to bucket x or bucket y).
					hash := t.key.alg.hash(k2, uintptr(h.hash0))
					if h.flags&iterator != 0 && !t.reflexivekey() && !t.key.alg.equal(k2, k2) {
						// If key != key (NaNs), then the hash could be (and probably
						// will be) entirely different from the old hash. Moreover,
						// it isn‘t reproducible. Reproducibility is required in the
						// presence of iterators, as our evacuation decision must
						// match whatever decision the iterator made.
						// Fortunately, we have the freedom to send these keys either
						// way. Also, tophash is meaningless for these kinds of keys.
						// We let the low bit of tophash drive the evacuation decision.
						// We recompute a new random tophash for the next level so
						// these keys will get evenly distributed across all buckets
						// after multiple grows.
						useY = top & 1
						top = tophash(hash)
					} else {
						if hash&newbit != 0 {
							useY = 1
						}
					}
				}

				if evacuatedX+1 != evacuatedY || evacuatedX^1 != evacuatedY {
					throw("bad evacuatedN")
				}
                // 修改topHash 为 evacuatedX 或 evacuatedY, 表示已经被迁移
				b.tophash[i] = evacuatedX + useY // evacuatedX + 1 == evacuatedY
				dst := &xy[useY]                 // evacuation destination
                // 迁移数据
				if dst.i == bucketCnt {
					dst.b = h.newoverflow(t, dst.b)
					dst.i = 0
					dst.k = add(unsafe.Pointer(dst.b), dataOffset)
					dst.e = add(dst.k, bucketCnt*uintptr(t.keysize))
				}
				dst.b.tophash[dst.i&(bucketCnt-1)] = top // mask dst.i as an optimization, to avoid a bounds check
				if t.indirectkey() {
					*(*unsafe.Pointer)(dst.k) = k2 // copy pointer
				} else {
					typedmemmove(t.key, dst.k, k) // copy elem
				}
				if t.indirectelem() {
					*(*unsafe.Pointer)(dst.e) = *(*unsafe.Pointer)(e)
				} else {
					typedmemmove(t.elem, dst.e, e)
				}
				dst.i++
				// These updates might push these pointers past the end of the
				// key or elem arrays.  That‘s ok, as we have the overflow pointer
				// at the end of the bucket to protect against pointing past the
				// end of the bucket.
				dst.k = add(dst.k, uintptr(t.keysize))
				dst.e = add(dst.e, uintptr(t.elemsize))
			}
		}
		// Unlink the overflow buckets & clear key/elem to help GC.
		// 清理数据
		if h.flags&oldIterator == 0 && t.bucket.ptrdata != 0 {
			b := add(h.oldbuckets, oldbucket*uintptr(t.bucketsize))
			// Preserve b.tophash because the evacuation
			// state is maintained there.
			ptr := add(b, dataOffset)
			n := uintptr(t.bucketsize) - dataOffset
			memclrHasPointers(ptr, n)
		}
	}
    
	if oldbucket == h.nevacuate {
	    // 统计是否完全迁移,如果完全迁移后,oldbuckets 会被释放掉(设置为nil)
		advanceEvacuationMark(h, t, newbit)
	}
}

上面我们提到了在Grow的流程中,新申请的buckets 可能会大小不变即same_size,也可能会变成oldbuckets的两倍即double_size,当double_size的情况下,会划分为两个半区firstsecond

# oldbuckets
|bucket1~N|

# newbuckets
|bucket1~N|bucketN+1~2N|
  first        second

我们以dobule_size为例,当插入一个新的key触发Grow操作的时候,整体的执行流程如下:

  1. 取hash(key)的低 B 位作为bucket编号 N, bucket_i = bucket_N;判断当前是否处于增长状态,如果不处于Grow_State,执行下一步;否则跳转到第 6 步;
  2. 如果bucket_i中可以找到插入的位置,则插入结束流程;否则执行下一步;
  3. 查找bucket_i 是否分配了overflow bucket,如果有分配overflow bucket,则 bucketi = overflwo_bucket,跳转到第2步;否则执行下一步;
  4. 判断是否需要执行Grow流程,如果需要则执行下一步;否则执行第 7 步;
  5. 创建新的 buckets并扩充到dobule_size,设置为Grow_State,跳转到第 1 步;
  6. oldbucketoldbuckets迁移到newbucketsoldbucket = N^(2^B-1),odlbucket中的数据会分布到firstsecond两个半区中(下面会详细说明),跳转到第2步执行;
  7. bucket_i分配的oveflow_bucketr,插入key,结束流程;

此处对第6步进行一下详细描述,我们假设 N = 12, B = 3,触发增长后B = 4, X = N^(2^B - 1) = 4, Y = X + 2^(B-1) = 12,新的key会被插到second半区;如果N = 4则会被插入到first半区,但无论如何都是在oldbucket迁移过来的数据桶中,以此来保证hash的一致性,这就是重构hash map的过程。

same_size的情况下执行过程是一样的,因为bucket总数不变,所以oldbucket对应迁移到new_buckets中相同编号的bucket中即可。

同时有一个细节值得我们注意,在bmap中被设置为emptyone的表示是已经被删除的数据,在迁移的过程中跳过即可,这样迁移后的数据会变的更加紧凑。

# 迁移前
# oldbucket_4
# emptyreset = 0, emptyone = 1
|xx|yy|1|ww|zz|0|0|0|key1~8|elem1~8|

# 迁移后
# newbucket
# newbucket_4
|xx|ww|0|0|0|0|0|0|key_xx|key_ww|...|elem_xx|elme_ww|...|

# newbucket_12
|yy|zz|0|0|0|0|0|0|key_yy|key_zz|...|elem_yy|elme_zz|...|

# oldbucket_4
# evacuatedX = 2, evacuatedY = 3, evacuatedEmpty = 4
|2|3|1|2|3|4|4|4|key1~8|elem1~8|

map access

golang map中访问一个map中数据有三种方式,我们以map[int]int为例:

    sets := map[int]int{1:2,3:4,5:6}
    value := map[1] // 返回值
    value, isExist := map[i] // 返回值和是否存在
    for key, value := range sets{ // 遍历
    }

前两种访问方式相同,都是通过key来访问,只是返回的值有所不同而已。


func mapaccess2(t *maptype, h *hmap, key unsafe.Pointer) (unsafe.Pointer, bool) {
	if raceenabled && h != nil {
		callerpc := getcallerpc()
		pc := funcPC(mapaccess2)
		racereadpc(unsafe.Pointer(h), callerpc, pc)
		raceReadObjectPC(t.key, key, callerpc, pc)
	}
	if msanenabled && h != nil {
		msanread(key, t.key.size)
	}
	if h == nil || h.count == 0 {
		if t.hashMightPanic() {
			t.key.alg.hash(key, 0) // see issue 23734
		}
		return unsafe.Pointer(&zeroVal[0]), false
	}
	if h.flags&hashWriting != 0 {
		throw("concurrent map read and map write")
	}
	
	// 1. 计算对应的桶编号,获取桶地址
	alg := t.key.alg
	hash := alg.hash(key, uintptr(h.hash0))
	m := bucketMask(h.B)
	b := (*bmap)(unsafe.Pointer(uintptr(h.buckets) + (hash&m)*uintptr(t.bucketsize))) // 取hash值的低B位为桶编号
	if c := h.oldbuckets; c != nil { // 当存在旧桶的时候,数据可能尚未迁移
		if !h.sameSizeGrow() { // 当扩充桶为double_size时, 桶编号要取 hash 值的低 B-1 位
			// There used to be half as many buckets; mask down one more power of two.
			m >>= 1
		}
		
		oldb := (*bmap)(unsafe.Pointer(uintptr(c) + (hash&m)*uintptr(t.bucketsize)))
		if !evacuated(oldb) { // 如果没有迁移则去老的桶取
			b = oldb
		}
	}
	top := tophash(hash)
bucketloop:
    // 2. 遍历寻找对应的key
	for ; b != nil; b = b.overflow(t) {
		for i := uintptr(0); i < bucketCnt; i++ {
			if b.tophash[i] != top {
				if b.tophash[i] == emptyRest {
					break bucketloop
				}
				continue
			}
			k := add(unsafe.Pointer(b), dataOffset+i*uintptr(t.keysize))
			if t.indirectkey() {
				k = *((*unsafe.Pointer)(k))
			}
			if alg.equal(key, k) {
				e := add(unsafe.Pointer(b), dataOffset+bucketCnt*uintptr(t.keysize)+i*uintptr(t.elemsize))
				if t.indirectelem() {
					e = *((*unsafe.Pointer)(e))
				}
				return e, true
			}
		}
	}
	return unsafe.Pointer(&zeroVal[0]), false
}

整个查询过程比较简单,如果你是顺序阅读到这里的话应该很好理解:

  1. 取低key_hash的低B位计算桶的编号N,bucket_i = bucket_N,如果此时有旧桶存在,执行第2步,否则执行第 4 步;
  2. 寻找oldbucket,判断Grow 类型,如果是same_size则编号编号仍然为N,如果是dobule_size则取key_hash低B-1位作为旧桶编号;执行下一步;
  3. 判断odlbucket是否被迁移,如果没被迁移则 bucket_i = old_bucket;执行下一步;
  4. bucket_ioverflow_bucket中寻找与key_topHash相等的tophash,找到后取对应的keyequal 比对,如果相等则返回 对应的elem,否则继续遍历,到结束为止。

相比较于按照key访问,遍历访问要更复杂一些:


// mapiterinit initializes the hiter struct used for ranging over maps.
// The hiter struct pointed to by ‘it‘ is allocated on the stack
// by the compilers order pass or on the heap by reflect_mapiterinit.
// Both need to have zeroed hiter since the struct contains pointers.
func mapiterinit(t *maptype, h *hmap, it *hiter) {
	if raceenabled && h != nil {
		callerpc := getcallerpc()
		racereadpc(unsafe.Pointer(h), callerpc, funcPC(mapiterinit))
	}

	if h == nil || h.count == 0 {
		return
	}

	if unsafe.Sizeof(hiter{})/sys.PtrSize != 12 {
		throw("hash_iter size incorrect") // see cmd/compile/internal/gc/reflect.go
	}
	it.t = t
	it.h = h

	// grab snapshot of bucket state
	// 1. 创建当前map状态快照
	it.B = h.B
	it.buckets = h.buckets
	if t.bucket.ptrdata == 0 {
		// Allocate the current slice and remember pointers to both current and old.
		// This preserves all relevant overflow buckets alive even if
		// the table grows and/or overflow buckets are added to the table
		// while we are iterating.
		h.createOverflow()
		it.overflow = h.extra.overflow
		it.oldoverflow = h.extra.oldoverflow
	}

	// decide where to start
	// 2. 决定起始位置
	// 取一个随机值
	r := uintptr(fastrand())
	if h.B > 31-bucketCntBits {
		r += uintptr(fastrand()) << 31
	}
	// 选取一个随机的起始bucket
	it.startBucket = r & bucketMask(h.B)
	// 选取一个随机的偏移量
	it.offset = uint8(r >> h.B & (bucketCnt - 1))

	// iterator state
	it.bucket = it.startBucket

	// Remember we have an iterator.
	// Can run concurrently with another mapiterinit().
	// 写入标志位
	if old := h.flags; old&(iterator|oldIterator) != iterator|oldIterator {
		atomic.Or8(&h.flags, iterator|oldIterator)
	}

	mapiternext(it)
}


func mapiternext(it *hiter) {
	h := it.h
	if raceenabled {
		callerpc := getcallerpc()
		racereadpc(unsafe.Pointer(h), callerpc, funcPC(mapiternext))
	}
	if h.flags&hashWriting != 0 {
		throw("concurrent map iteration and map write")
	}
	t := it.t
	bucket := it.bucket
	b := it.bptr
	i := it.i
	checkBucket := it.checkBucket
	alg := t.key.alg

next:
	if b == nil {
	    // 判断是否已经循环遍历所有的Bucket
		if bucket == it.startBucket && it.wrapped {
			// end of iteration
			it.key = nil
			it.elem = nil
			return
		}
		
		if h.growing() && it.B == h.B {
		    //map 在grow state 且 iterInit在grow state 或者是 `same_size`
			// Iterator was started in the middle of a grow, and the grow isn‘t done yet.
			// If the bucket we‘re looking at hasn‘t been filled in yet (i.e. the old
			// bucket hasn‘t been evacuated) then we need to iterate through the old
			// bucket and only return the ones that will be migrated to this bucket.
			oldbucket := bucket & it.h.oldbucketmask()
			b = (*bmap)(add(h.oldbuckets, oldbucket*uintptr(t.bucketsize)))
			if !evacuated(b) { // 判断是否迁移
				checkBucket = bucket
			} else {
				b = (*bmap)(add(it.buckets, bucket*uintptr(t.bucketsize)))
				checkBucket = noCheck
			}
		} else {
		    // map未处于`grow state`,或`grow state`为`doubel_size`.
			b = (*bmap)(add(it.buckets, bucket*uintptr(t.bucketsize)))
			checkBucket = noCheck
		}
		bucket++
		if bucket == bucketShift(it.B) {
			bucket = 0
			it.wrapped = true
		}
		i = 0
	}
	for ; i < bucketCnt; i++ {
		offi := (i + it.offset) & (bucketCnt - 1)
		// 跳过空闲位置
		if isEmpty(b.tophash[offi]) || b.tophash[offi] == evacuatedEmpty {
			// TODO: emptyRest is hard to use here, as we start iterating
			// in the middle of a bucket. It‘s feasible, just tricky.
			continue
		}
		k := add(unsafe.Pointer(b), dataOffset+uintptr(offi)*uintptr(t.keysize))
		if t.indirectkey() {
			k = *((*unsafe.Pointer)(k))
		}
		e := add(unsafe.Pointer(b), dataOffset+bucketCnt*uintptr(t.keysize)+uintptr(offi)*uintptr(t.elemsize))
		
		if checkBucket != noCheck && !h.sameSizeGrow() {// 过滤不会被迁移过来的数据
			// Special case: iterator was started during a grow to a larger size
			// and the grow is not done yet. We‘re working on a bucket whose
			// oldbucket has not been evacuated yet. Or at least, it wasn‘t
			// evacuated when we started the bucket. So we‘re iterating
			// through the oldbucket, skipping any keys that will go
			// to the other new bucket (each oldbucket expands to two
			// buckets during a grow).
			if t.reflexivekey() || alg.equal(k, k) {
				// If the item in the oldbucket is not destined for
				// the current new bucket in the iteration, skip it.
				hash := alg.hash(k, uintptr(h.hash0))
				if hash&bucketMask(it.B) != checkBucket {
					continue
				}
			} else {
				// Hash isn‘t repeatable if k != k (NaNs).  We need a
				// repeatable and randomish choice of which direction
				// to send NaNs during evacuation. We‘ll use the low
				// bit of tophash to decide which way NaNs go.
				// NOTE: this case is why we need two evacuate tophash
				// values, evacuatedX and evacuatedY, that differ in
				// their low bit.
				if checkBucket>>(it.B-1) != uintptr(b.tophash[offi]&1) {
					continue
				}
			}
		}
		if (b.tophash[offi] != evacuatedX && b.tophash[offi] != evacuatedY) ||
			!(t.reflexivekey() || alg.equal(k, k)) {// 数据没有迁移,直接访问
			// This is the golden data, we can return it.
			// OR
			// key!=key, so the entry can‘t be deleted or updated, so we can just return it.
			// That‘s lucky for us because when key!=key we can‘t look it up successfully.
			it.key = k
			if t.indirectelem() {
				e = *((*unsafe.Pointer)(e))
			}
			it.elem = e
		} else { // 数据已经迁移,通过key定位访问
			// The hash table has grown since the iterator was started.
			// The golden data for this key is now somewhere else.
			// Check the current hash table for the data.
			// This code handles the case where the key
			// has been deleted, updated, or deleted and reinserted.
			// NOTE: we need to regrab the key as it has potentially been
			// updated to an equal() but not identical key (e.g. +0.0 vs -0.0).
			rk, re := mapaccessK(t, h, k)
			if rk == nil {
				continue // key has been deleted
			}
			it.key = rk
			it.elem = re
		}
		it.bucket = bucket
		if it.bptr != b { // avoid unnecessary write barrier; see issue 14921
			it.bptr = b
		}
		it.i = i + 1
		it.checkBucket = checkBucket
		return
	}
	b = b.overflow(t) // 继续遍历overflow bucket
	i = 0
	goto next
}

当我们去遍历一个map的时候可能有三种情况:

  1. iterInititerNext期间未发生Grow: 只要顺序遍历it.buckets数据即可;
  2. iterInititerNext 都发生在Grow state: 上述代码可以清晰看出iterInit 时候是对 h.bucketsh.B 做了 snapshot,此时iterNext是以 newBucket作为基础去遍历的,那么一个bucket_N可能有两个状态:已经迁移和尚未迁移。已经迁移的直接遍历桶即可,未迁移的则需要去oldbucket中遍历,不过需要注意的一点是,要过滤掉那些不可能被迁移到bucket_N的数据(在double_size情况下分上下半区);
  3. iterInitGrow state 之前,iterNextGrow State:此种情况更加复杂一些,因为iterInit是在Grow之前,iterNext的时候it.buckets实际对应的是h.oldbucket,也就是说是基于oldbuckets去遍历,此时bucket_N也有两种情况:已经迁移和没有迁移,没有迁移的直接取数据返回,已经迁移的则直接通过key访问,因为此时可能在新bucekt中已经被更新或者删除了。

map delete

上面我们也提到过,map的移除是通过修改topHashemptyOne完成,删除逻辑要比插入逻辑简单很多:

func mapdelete(t *maptype, h *hmap, key unsafe.Pointer) {
	if raceenabled && h != nil {
		callerpc := getcallerpc()
		pc := funcPC(mapdelete)
		racewritepc(unsafe.Pointer(h), callerpc, pc)
		raceReadObjectPC(t.key, key, callerpc, pc)
	}
	if msanenabled && h != nil {
		msanread(key, t.key.size)
	}
	if h == nil || h.count == 0 {
		if t.hashMightPanic() {
			t.key.alg.hash(key, 0) // see issue 23734
		}
		return
	}
	if h.flags&hashWriting != 0 {
		throw("concurrent map writes")
	}

	alg := t.key.alg
	hash := alg.hash(key, uintptr(h.hash0))

	// Set hashWriting after calling alg.hash, since alg.hash may panic,
	// in which case we have not actually done a write (delete).
	h.flags ^= hashWriting

	bucket := hash & bucketMask(h.B)
	if h.growing() { // 判断是否在增长,如果在增长,则对对应的oldbucket 进行迁移
		growWork(t, h, bucket)
	}
	b := (*bmap)(add(h.buckets, bucket*uintptr(t.bucketsize)))
	bOrig := b // bucket在base区域的位置,但key实际所在位置可能是overflow
	top := tophash(hash)
search:
	for ; b != nil; b = b.overflow(t) {
		for i := uintptr(0); i < bucketCnt; i++ {
			if b.tophash[i] != top {
				if b.tophash[i] == emptyRest { // 搜索到emptyRest,停止搜索
					break search
				}
				continue
			}
			k := add(unsafe.Pointer(b), dataOffset+i*uintptr(t.keysize))
			k2 := k
			if t.indirectkey() {
				k2 = *((*unsafe.Pointer)(k2))
			}
			if !alg.equal(key, k2) {
				continue
			}
			// Only clear key if there are pointers in it.
			// 清理内存
			if t.indirectkey() {
				*(*unsafe.Pointer)(k) = nil
			} else if t.key.ptrdata != 0 {
				memclrHasPointers(k, t.key.size) 
			}
			e := add(unsafe.Pointer(b), dataOffset+bucketCnt*uintptr(t.keysize)+i*uintptr(t.elemsize))
			if t.indirectelem() {
				*(*unsafe.Pointer)(e) = nil
			} else if t.elem.ptrdata != 0 {
				memclrHasPointers(e, t.elem.size)
			} else {
				memclrNoHeapPointers(e, t.elem.size)
			}
			
			b.tophash[i] = emptyOne
			// If the bucket now ends in a bunch of emptyOne states,
			// change those to emptyRest states.
			// It would be nice to make this a separate function, but
			// for loops are not currently inlineable.
			// 如果emptyOne 后面紧跟的 emptyRest,则把emptyOne设置为emptyRest
			if i == bucketCnt-1 {
				if b.overflow(t) != nil && b.overflow(t).tophash[0] != emptyRest {
					goto notLast
				}
			} else {
				if b.tophash[i+1] != emptyRest {
					goto notLast
				}
			}
			for {
				b.tophash[i] = emptyRest
				if i == 0 {
					if b == bOrig {
						break // beginning of initial bucket, we‘re done.
					}
					// Find previous bucket, continue at its last entry.
					c := b
					for b = bOrig; b.overflow(t) != c; b = b.overflow(t) { //查找b前面的overflow
					}
					i = bucketCnt - 1
				} else {
					i--
				}
				if b.tophash[i] != emptyOne {
					break
				}
			}
		notLast:
			h.count--
			break search
		}
	}

	if h.flags&hashWriting == 0 {
		throw("concurrent map writes")
	}
	h.flags &^= hashWriting
}

执行流程如下:

  1. 查找key所在bucket,将对应位置的topHash设置为emptyOne,清除对应的keyelem数据;
  2. 如果下一个位置为emptyRest(包括后面紧跟的overflow_bueckt),则将emptyOne修改为emptyRest,向下执行;否则结束流程;
  3. 向前查找前一个位置,如果到了base_bucekt的起始位置,则结束流程;否则跳转到第2步。

总结

  1. golang map的底层实现是通过hash table实现的,每个bucket可以存贮8个key-elem,通过key_hash的低B(总共划分2^B个桶)bit划分桶,桶内通过topHash(key_hash前8bit)做区分;
  2. hash table在内存中使用连续数组+跳转指针存储,跳转指针指向overflow_bucket,根据key查找的时候都是在连续内存上操作,以此来保证O(1)时间复杂度;
  3. 桶的Grow可以剔除被删除数据占用的空间,使得数据更加紧凑,同时overflow_bucket的排序会发生改变,优先迁移的bucket对应的overflow_bucket地址靠前。有两种形式:same_sizeGrow前后桶数不变),数据分桶格局不变;double_size(扩张后桶数*2), 根据key_hash的B-1 bit决定是划分到first半区还是second半区,完成桶的重新划分。
  4. 删除数据的时候会见对应位置的topHash设置为emptyOne,如果一个bucekt(这里的bucekt是指逻辑上的桶,包括base_bucekt和overflow_bucket)中最后的位置为emptyOne则修改为emptyreset

golang map 内幕

标签:round   for loop   基本   执行流程   lex   ups   currently   cas   函数   

原文地址:https://www.cnblogs.com/cnblogs-wangzhipeng/p/13292524.html

(0)
(0)
   
举报
评论 一句话评论(0
登录后才能评论!
© 2014 mamicode.com 版权所有  联系我们:gaon5@hotmail.com
迷上了代码!