通过学习sync.Map的实现,可以学习以下性能优化的方向和技巧
- 成本分摊: 将一个明确的内存拷贝成本 变成 后续调用可能触发的内存拷贝成本。见Range实现
- 针对读多写少的场景,优化读取性能(尽量无锁化)。
- 由于read中存储的快照,dirty中存储的是最新的数据。为了防止两者差距过大导致Load miss,引入监听misses变量触发两者数据的同步
使用场景
- when the entry for a given key is only ever written once but read many times, as in caches that only grow(读多写少的场景)
- when multiple goroutines read, write, and overwrite entries for disjoint sets of keys
核心构成
mu Mutex
在操作dirty
以及处理删除元素等场景下保护临界资源read atomic.Value
只读快照 满足大多数Load场景dirty map[any]*entry
存放最新数据misses int
记录Load等场景下 需要取锁的次数。该值到达一定限值,会触发diry
到read
的替换并清空dirty
go
// Map is like a Go map[interface{}]interface{} but is safe for concurrent use
// by multiple goroutines without additional locking or coordination.
// Loads, stores, and deletes run in amortized constant time.
//
// The Map type is specialized. Most code should use a plain Go map instead,
// with separate locking or coordination, for better type safety and to make it
// easier to maintain other invariants along with the map content.
//
// The Map type is optimized for two common use cases: (1) when the entry for a given
// key is only ever written once but read many times, as in caches that only grow,
// or (2) when multiple goroutines read, write, and overwrite entries for disjoint
// sets of keys. In these two cases, use of a Map may significantly reduce lock
// contention compared to a Go map paired with a separate Mutex or RWMutex.
//
// The zero Map is empty and ready for use. A Map must not be copied after first use.
//
// In the terminology of the Go memory model, Map arranges that a write operation
// "synchronizes before" any read operation that observes the effect of the write, where
// read and write operations are defined as follows.
// Load, LoadAndDelete, LoadOrStore are read operations;
// Delete, LoadAndDelete, and Store are write operations;
// and LoadOrStore is a write operation when it returns loaded set to false.
type Map struct {
mu Mutex
// read contains the portion of the map's contents that are safe for
// concurrent access (with or without mu held).
//
// The read field itself is always safe to load, but must only be stored with
// mu held.
//
// Entries stored in read may be updated concurrently without mu, but updating
// a previously-expunged entry requires that the entry be copied to the dirty
// map and unexpunged with mu held.
read atomic.Value // readOnly
// dirty contains the portion of the map's contents that require mu to be
// held. To ensure that the dirty map can be promoted to the read map quickly,
// it also includes all of the non-expunged entries in the read map.
//
// Expunged entries are not stored in the dirty map. An expunged entry in the
// clean map must be unexpunged and added to the dirty map before a new value
// can be stored to it.
//
// If the dirty map is nil, the next write to the map will initialize it by
// making a shallow copy of the clean map, omitting stale entries.
dirty map[any]*entry
// misses counts the number of loads since the read map was last updated that
// needed to lock mu to determine whether the key was present.
//
// Once enough misses have occurred to cover the cost of copying the dirty
// map, the dirty map will be promoted to the read map (in the unamended
// state) and the next store to the map will make a new dirty copy.
misses int
}
核心方法
Store
-
如果在
read
存在该key,且该key没有被标记为已经从dirty map中删除
,则可以直接原子设置新的value
-
如果上一步没有成功,则需要加锁开始判断了
-
如果
read
中存在该key, 且这个key标记为已经从dirty中删除
,那么需要先在dirty
上添加该元素 , 原子设置新的value
( 为什么获取锁需要再次判断?因为可能其他协程先获取了锁,状态发生了变化 ) -
如果在
dirty
中找到则原子设置新的value
-
都没找到的话,那就往dirty map中添加一个新的元素了。由于
dirty
是懒加载的,所以可能不存在。初始化dirty map的流程如下:-
从read中"浅拷贝" 一份数据,并且剔除掉标记为删除的元素(
entry.p=nil --> entry.p=
expunged
) -
原子的设置
read.amended=true
,表明该map存在dirty
-
-
go
// Store sets the value for a key.
func (m *Map) Store(key, value any) {
read, _ := m.read.Load().(readOnly)
if e, ok := read.m[key]; ok && e.tryStore(&value) {
return
}
m.mu.Lock()
read, _ = m.read.Load().(readOnly)
if e, ok := read.m[key]; ok {
if e.unexpungeLocked() {
// The entry was previously expunged, which implies that there is a
// non-nil dirty map and this entry is not in it.
m.dirty[key] = e
}
e.storeLocked(&value)
} else if e, ok := m.dirty[key]; ok {
e.storeLocked(&value)
} else {
// 如果 read.amended=false 表明目前还没有建立dirty map
// 需要从read中"浅拷贝" 一份数据,并且剔除掉标记为删除的元素
if !read.amended {
// We're adding the first new key to the dirty map.
// Make sure it is allocated and mark the read-only map as incomplete.
m.dirtyLocked()
m.read.Store(readOnly{m: read.m, amended: true})
}
m.dirty[key] = newEntry(value)
}
m.mu.Unlock()
}
// tryStore stores a value if the entry has not been expunged.
//
// If the entry is expunged, tryStore returns false and leaves the entry
// unchanged.
// 如果该entry标记为 已经从dirty map中删除
// 不能直接在read map中更新(因为会导致dirty map的状态落后于read map)
// 反之可以直接更新 因为entry是指针,修改同时也会作用到dirty map
func (e *entry) tryStore(i *any) bool {
for {
p := atomic.LoadPointer(&e.p)
if p == expunged {
return false
}
if atomic.CompareAndSwapPointer(&e.p, p, unsafe.Pointer(i)) {
return true
}
}
}
// expunged is an arbitrary pointer that marks entries which have been deleted
// from the dirty map.
var expunged = unsafe.Pointer(new(any))
func (m *Map) dirtyLocked() {
if m.dirty != nil {
return
}
read, _ := m.read.Load().(readOnly)
m.dirty = make(map[any]*entry, len(read.m))
for k, e := range read.m {
// 标记为删除的元素 不会放入dirty map 这时候才算真正删除
if !e.tryExpungeLocked() {
m.dirty[k] = e
}
}
}
// 尝试将 entry.p==nil 的元素 设置为expunged 标识
// 表明这个case会从dirty map中移除
func (e *entry) tryExpungeLocked() (isExpunged bool) {
p := atomic.LoadPointer(&e.p)
for p == nil {
if atomic.CompareAndSwapPointer(&e.p, nil, expunged) {
return true
}
p = atomic.LoadPointer(&e.p)
}
return p == expunged
}
// unexpungeLocked ensures that the entry is not marked as expunged.
//
// If the entry was previously expunged, it must be added to the dirty map
// before m.mu is unlocked.
// expunged 标记 --> 删除标记
func (e *entry) unexpungeLocked() (wasExpunged bool) {
return atomic.CompareAndSwapPointer(&e.p, expunged, nil)
}
Load
-
如果在
read
不存在该key,且不存在dirty
,那么可以直接返回了,数据就是不存在 -
如果存在
dirty
那就需要取锁继续判断了- 如果read中存在则直接返回对应数据
- 尝试从dirty中获取。并自增misses,到达限值则用
dirty
替换read
go
// Load returns the value stored in the map for a key, or nil if no
// value is present.
// The ok result indicates whether value was found in the map.
func (m *Map) Load(key any) (value any, ok bool) {
read, _ := m.read.Load().(readOnly)
e, ok := read.m[key]
if !ok && read.amended {
m.mu.Lock()
// Avoid reporting a spurious miss if m.dirty got promoted while we were
// blocked on m.mu. (If further loads of the same key will not miss, it's
// not worth copying the dirty map for this key.)
read, _ = m.read.Load().(readOnly)
e, ok = read.m[key]
if !ok && read.amended {
e, ok = m.dirty[key]
// Regardless of whether the entry was present, record a miss: this key
// will take the slow path until the dirty map is promoted to the read
// map.
m.missLocked()
}
m.mu.Unlock()
}
if !ok {
return nil, false
}
return e.load()
}
// 1. misses ++
// 2. miss次数达到限值 dirty替换read 同时清空dirty和计数misses
func (m *Map) missLocked() {
m.misses++
if m.misses < len(m.dirty) {
return
}
m.read.Store(readOnly{m: m.dirty})
m.dirty = nil
m.misses = 0
}
func (e *entry) load() (value any, ok bool) {
p := atomic.LoadPointer(&e.p)
if p == nil || p == expunged {
return nil , false
}
return *(*any)(p), true
}
// expunged is an arbitrary pointer that marks entries which have been deleted
// from the dirty map.
var expunged = unsafe.Pointer(new(any))
LoadAndDelete
-
如果在
read
中 可以直接删除(此处的删除并不会移除元素,只是原子的设置entry.p
为nil) -
如果
read
中没有,并且read
不是最新快照数据,尝试加锁,去dirty
中寻找 -
如果找到则
- dirty中删除该
entry
- 将
entry.p
设置为nil
- dirty中删除该
Q: 为什么在read中删除只是将 entry.p
设置nil,从dirty中删除需要同时删除dirty map中的key呢?
A: 因为read只是一份快照,从对应的map中删除没有意义,而dirty中需要保存最新的数据,以便于在合适的时机原子地替换read
。而原子的设置entry.p
为nil,是为了表明该key已经被 标记删除
Q: dirty的数据什么时候同步到read中?
A: 由于read中的数据可能是落后于dirty的。从read中读取无需加锁,而从dirty中读取需要加锁。所以为了优化性能,每次通过加锁进行相关操作的时候会执行s.miss++
,当m.misses >= len(m.dirty)
的时候,会原子地同步替换read
为dirty
从而降低取锁概率,并清空 dirty
(等待下次Store新元素的时候再次浅拷贝read, 这也说明该map的使用场景是读多写少,因为写多的场景会不断出现map的拷贝,性能可能不及分段锁map
)
go
// LoadAndDelete deletes the value for a key, returning the previous value if any.
// The loaded result reports whether the key was present.
func (m *Map) LoadAndDelete(key any) (value any, loaded bool) {
read, _ := m.read.Load().(readOnly)
e, ok := read.m[key]
if !ok && read.amended {
m.mu.Lock()
read, _ = m.read.Load().(readOnly)
e, ok = read.m[key]
if !ok && read.amended {
e, ok = m.dirty[key]
delete(m.dirty, key)
// Regardless of whether the entry was present, record a miss: this key
// will take the slow path until the dirty map is promoted to the read
// map.
m.missLocked()
}
m.mu.Unlock()
}
if ok {
return e.delete()
}
return nil, false
}
func (m *Map) missLocked() {
m.misses++
if m.misses < len(m.dirty) {
return
}
m.read.Store(readOnly{m: m.dirty})
m.dirty = nil
m.misses = 0
}
// 尝试原子地将 entry.p 设置nil的过程
func (e *entry) delete() (value any, ok bool) {
for {
p := atomic.LoadPointer(&e.p)
if p == nil || p == expunged {
return nil, false
}
if atomic.CompareAndSwapPointer(&e.p, p, nil) {
return *(*any)(p), true
}
}
}
Delete
调用 LoadAndDelete
scss
// Delete deletes the value for a key.
func (m *Map) Delete(key any) {
m.LoadAndDelete(key)
}
Range
-
如果不存在
dirty
则直接基于read快照做遍历 -
如果存在dirty说明read中数据不是最新的,考虑到加锁拷贝一份dirty耗时较长,便通过1. read替换为dirty 2. 清空dirty的方式 将拷贝的成本分摊到后面的调用中(Store方法可能会拷贝read->dirty)
go
// Range calls f sequentially for each key and value present in the map.
// If f returns false, range stops the iteration.
//
// Range does not necessarily correspond to any consistent snapshot of the Map's
// contents: no key will be visited more than once, but if the value for any key
// is stored or deleted concurrently (including by f), Range may reflect any
// mapping for that key from any point during the Range call. Range does not
// block other methods on the receiver; even f itself may call any method on m.
//
// Range may be O(N) with the number of elements in the map even if f returns
// false after a constant number of calls.
func (m *Map) Range(f func(key, value any) bool) {
// We need to be able to iterate over all of the keys that were already
// present at the start of the call to Range.
// If read.amended is false, then read.m satisfies that property without
// requiring us to hold m.mu for a long time.
read, _ := m.read.Load().(readOnly)
if read.amended {
// m.dirty contains keys not in read.m. Fortunately, Range is already O(N)
// (assuming the caller does not break out early), so a call to Range
// amortizes an entire copy of the map: we can promote the dirty copy
// immediately!
m.mu.Lock()
read, _ = m.read.Load().(readOnly)
if read.amended {
read = readOnly{m: m.dirty}
m.read.Store(read)
m.dirty = nil
m.misses = 0
}
m.mu.Unlock()
}
for k, e := range read.m {
v, ok := e.load()
if !ok {
continue
}
if !f(k, v) {
break
}
}
}