Introduction
Go's built- map
in does not support concurrent write operations. The reason is that map
write operations are not concurrently safe. When you try to operate the same one with multiple Goroutines map
, an error will be reported: fatal error: concurrent map writes
.
Therefore, the official introduced additionally sync.Map
to meet the application in concurrent programming.
sync.Map
The realization principle can be summarized as:
- The read and write are separated by the two fields of read and dirty. The read data is stored in the read-only field read, and the latest written data is stored in the dirty field.
- When reading, it will query read first, and then query dirty if it does not exist, and only write dirty when writing.
- Reading read does not require locking, and reading or writing dirty requires locking
- In addition, there is a misses field to count the number of times the read has been penetrated (penetration refers to the need to read dirty), and if it exceeds a certain number of times, the dirty data will be synchronized to the read.
- For deleted data, the deletion is delayed directly by marking
data structure
Map
The data structure is as follows:
type Map struct {
// 加锁作用,保护 dirty 字段
mu Mutex
// 只读的数据,实际数据类型为 readOnly
read atomic.Value
// 最新写入的数据
dirty map[interface{}]*entry
// 计数器,每次需要读 dirty 则 +1
misses int
}
The data structure of readOnly is:
type readOnly struct {
// 内建 map
m map[interface{}]*entry
// 表示 dirty 里存在 read 里没有的 key,通过该字段决定是否加锁读 dirty
amended bool
}
entry
Data structures are used to store pointers to values:
type entry struct {
p unsafe.Pointer // 等同于 *interface{}
}
Property p has three states:
p == nil
: the key value has been deleted, andm.dirty == nil
p == expunged
: The key value has been deleted,m.dirty!=nil
butm.dirty
the key value does not exist (expunged is actually a null interface pointer)- Except for the above, the key-value pair exists, exists in
m.read.m
, and ifm.dirty!=nil
, also exists inm.dirty
Map
The following methods are commonly used:
Load
: Read the specified key and return the valueStore
: Store (add or change) key-valueDelete
: delete the specified key
Source code analysis
Load
func (m *Map) Load(key interface{}) (value interface{}, ok bool) {
// 首先尝试从 read 中读取 readOnly 对象
read, _ := m.read.Load().(readOnly)
e, ok := read.m[key]
// 如果不存在则尝试从 dirty 中获取
if !ok && read.amended {
m.mu.Lock()
// 由于上面 read 获取没有加锁,为了安全再检查一次
read, _ = m.read.Load().(readOnly)
e, ok = read.m[key]
// 确实不存在则从 dirty 获取
if !ok && read.amended {
e, ok = m.dirty[key]
// 调用 miss 的逻辑
m.missLocked()
}
m.mu.Unlock()
}
if !ok {
return nil, false
}
// 从 entry.p 读取值
return e.load()
}
func (m *Map) missLocked() {
m.misses++
if m.misses < len(m.dirty) {
return
}
// 当 miss 积累过多,会将 dirty 存入 read,然后 将 amended = false,且 m.dirty = nil
m.read.Store(readOnly{m: m.dirty})
m.dirty = nil
m.misses = 0
}
Store
func (m *Map) Store(key, value interface{}) {
read, _ := m.read.Load().(readOnly)
// 如果 read 里存在,则尝试存到 entry 里
if e, ok := read.m[key]; ok && e.tryStore(&value) {
return
}
// 如果上一步没执行成功,则要分情况处理
m.mu.Lock()
read, _ = m.read.Load().(readOnly)
// 和 Load 一样,重新从 read 获取一次
if e, ok := read.m[key]; ok {
// 情况 1:read 里存在
if e.unexpungeLocked() {
// 如果 p == expunged,则需要先将 entry 赋值给 dirty(因为 expunged 数据不会留在 dirty)
m.dirty[key] = e
}
// 用值更新 entry
e.storeLocked(&value)
} else if e, ok := m.dirty[key]; ok {
// 情况 2:read 里不存在,但 dirty 里存在,则用值更新 entry
e.storeLocked(&value)
} else {
// 情况 3:read 和 dirty 里都不存在
if !read.amended {
// 如果 amended == false,则调用 dirtyLocked 将 read 拷贝到 dirty(除了被标记删除的数据)
m.dirtyLocked()
// 然后将 amended 改为 true
m.read.Store(readOnly{m: read.m, amended: true})
}
// 将新的键值存入 dirty
m.dirty[key] = newEntry(value)
}
m.mu.Unlock()
}
func (e *entry) tryStore(i *interface{}) bool {
for {
p := atomic.LoadPointer(&e.p)
if p == expunged {
return false
}
if atomic.CompareAndSwapPointer(&e.p, p, unsafe.Pointer(i)) {
return true
}
}
}
func (e *entry) unexpungeLocked() (wasExpunged bool) {
return atomic.CompareAndSwapPointer(&e.p, expunged, nil)
}
func (e *entry) storeLocked(i *interface{}) {
atomic.StorePointer(&e.p, unsafe.Pointer(i))
}
func (m *Map) dirtyLocked() {
if m.dirty != nil {
return
}
read, _ := m.read.Load().(readOnly)
m.dirty = make(map[interface{}]*entry, len(read.m))
for k, e := range read.m {
// 判断 entry 是否被删除,否则就存到 dirty 中
if !e.tryExpungeLocked() {
m.dirty[k] = e
}
}
}
func (e *entry) tryExpungeLocked() (isExpunged bool) {
p := atomic.LoadPointer(&e.p)
for p == nil {
// 如果有 p == nil(即键值对被 delete),则会在这个时机被置为 expunged
if atomic.CompareAndSwapPointer(&e.p, nil, expunged) {
return true
}
p = atomic.LoadPointer(&e.p)
}
return p == expunged
}
Delete
func (m *Map) Delete(key interface{}) {
m.LoadAndDelete(key)
}
// LoadAndDelete 作用等同于 Delete,并且会返回值与是否存在
func (m *Map) LoadAndDelete(key interface{}) (value interface{}, loaded bool) {
// 获取逻辑和 Load 类似,read 不存在则查询 dirty
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]
m.missLocked()
}
m.mu.Unlock()
}
// 查询到 entry 后执行删除
if ok {
// 将 entry.p 标记为 nil,数据并没有实际删除
// 真正删除数据并被被置为 expunged,是在 Store 的 tryExpungeLocked 中
return e.delete()
}
return nil, false
}
Summarize
It can be seen that through this design of read-write separation, the write security in concurrent situations is solved, and the read speed can be close to the built-in in most cases map
, which is very suitable for the situation of reading more and writing less.
sync.Map
There are some other methods:
Range
: Traverse all key-value pairs, the parameter is the callback functionLoadOrStore
: Read the data, if it does not exist, save it and read it again
It will not be explained in detail here, please refer to the source code .