深入理解golang：sync.map

疑惑开篇#
有了map为什么还要搞个sync.map 呢？它们之间有什么区别？
答：重要的一点是，map并发不是安全的。

在Go 1.6之前， 内置的map类型是部分goroutine安全的，并发的读没有问题，并发的写可能有问题。自go 1.6之后， 并发地读写map会报错，这在一些知名的开源库中都存在这个问题，所以go 1.9之前的解决方案是额外绑定一个锁，封装成一个新的struct或者单独使用锁都可以。

go version go1.13.9 windows/amd64

测试一波#
写一个简单的并发写map的测试代码看看：
testcurmap.go

Copy
package main

import (
“fmt”
“time”
)

func main() {
m := map[string]int{“age”: 10}

go func() {
	i := 0
	for i < 1000 {
		m["age"] = 10
		i++
	}
}()

go func() { //19 行
	i := 0
	for i < 1000 {
		m["age"] = 11 //22 行
		i++
	}
}()

time.Sleep(time.Second * 3)
fmt.Println(m)

}
多运行几次：go run testcurmap.go
会报错，错误的扼要信息如下：

fatal error: concurrent map writes

goroutine 7 [running]:
runtime.throw(0x4d49a3, 0x15)
/go/src/runtime/panic.go:774 +0x79 fp=0xc000041f30 sp=0xc000041f00 pc=0x42cf19
runtime.mapassign_faststr(0x4b4360, 0xc000066330, 0x4d168a, 0x3, 0x0)
/go/src/runtime/map_faststr.go:211 +0x41e fp=0xc000041f98 sp=0xc000041f30 pc=0x410f8e
main.main.func2(0xc000066330)
/mygo/src/study/go-practice2/map/curmap/testcurmap.go:22 +0x5c fp=0xc000041fd8 sp=0xc000041f98 pc=0x49ac9c
runtime.goexit()
/go/src/runtime/asm_amd64.s:1357 +0x1 fp=0xc000041fe0 sp=0xc000041fd8 pc=0x455391
created by main.main
/mygo/src/study/go-practice2/map/curmap/testcurmap.go:19 +0xb0

exit status 2

看报错信息是src/runtime/map_faststr.go:211 这个函数runtime.mapassign_faststr，它在runtime/map_faststr.go 中，简要代码如下：

Copy
func mapassign_faststr(t *maptype, h *hmap, s string) unsafe.Pointer {
… …

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

... ...

}
hashWriting = 4 // a goroutine is writing to the map goroutine写的一个标识，
这里h.flags与自己进行与运算，判断是否有其他goroutine在操作这个map，不是0说明有其他goroutine操作map，所以报错。

那咋防止map并发呢，一般有几种方式：

map+Mutex：
给map加一把大锁
map+RWMutex
给map加一个读写锁，给锁细分。适合读多写少场景
修改一下程序#
加一把读写锁防止并发，修改程序 testcurmap2.go：

Copy
package main

import (
“fmt”
“sync”
“time”
)

func main() {
m := map[string]int{“age”: 10}

var s sync.RWMutex
go func() {
	i := 0
	for i < 1000 {
		s.Lock()
		m["age"] = 10
		s.Unlock()
		i++
	}
}()

go func() {
	i := 0
	for i < 1000 {
		s.Lock()
		m["age"] = 11
		s.Unlock()
		i++
	}
}()

time.Sleep(time.Second * 3)
fmt.Println(m)

}
运行结果：
map[age:11]

没有报错了。

就到这里了吗？可以在思考思考，还有其他方法控制并发的方法没？有的，sync.map 登场

控制并的第三种方式：

sync.Map
官方实现的并发map。
原理是通过分离读写map和原子指令来实现读的近似无锁，并通过延迟更新的方式来保证读的无锁化。一般情况下可以替换上面2种锁。
sync.map#
先看一个简单的代码 testcurmap3.go

Copy
package main

import (
“fmt”
“sync”
“time”
)

func main() {
smap := sync.Map{}

smap.Store("age", 10)

go func() {
	i := 0
	for i < 1000 {
		smap.Store("one", 10)
		i++
	}
}()

go func() {
	i := 0
	for i < 1000 {
		smap.Store("one", 11)
		i++
	}
}()

time.Sleep(time.Second * 2)
fmt.Println(smap.Load("one"))

}
运行输出：11 true
正常输出，没有报错。

sync.Map 的主要思想就是读写分离，空间换时间。

看看 sync.map 优点：

空间换时间：通过冗余的两个数据结构(read、dirty)，实现加锁对性能的影响。
使用只读数据(read)，避免读写冲突。
动态调整，miss次数多了之后，将dirty数据迁移到read中。
double-checking。
延迟删除。 删除一个键值只是打标记，只有在迁移dirty数据的时候才清理删除的数据。
优先从read读取、更新、删除，因为对read的读取不需要锁。
sync.Map 数据结构#
Map 数据结构#
在 src/sync/map.go 中

Copy
type Map struct {
// 当涉及到脏数据(dirty)操作时候，需要使用这个锁
mu Mutex

// read是一个只读数据结构，包含一个map结构，
// 读不需要加锁，只需要通过 atomic 加载最新的指正即可
read atomic.Value // readOnly

// dirty 包含部分map的键值对，如果操作需要mutex获取锁
// 最后dirty中的元素会被全部提升到read里的map去
dirty map[interface{}]*entry

// misses是一个计数器，用于记录read中没有的数据而在dirty中有的数据的数量。
// 也就是说如果read不包含这个数据，会从dirty中读取，并misses+1
// 当misses的数量等于dirty的长度，就会将dirty中的数据迁移到read中
misses int

}
read的数据结构 readOnly：#
Copy
// readOnly is an immutable struct stored atomically in the Map.read field.
type readOnly struct {
// m包含所有只读数据，不会进行任何的数据增加和删除操作
// 但是可以修改entry的指针因为这个不会导致map的元素移动
m map[interface{}]*entry

// 标志位，如果为true则表明当前read只读map的数据不完整，dirty map中包含部分数据
amended bool // true if the dirty map contains some key not in m.

}
只读map，对该map的访问不需要加锁，但是这个map也不会增加元素，元素会被先增加到dirty中，然后后续会迁移到read只读map中，通过原子操作所以不需要加锁操作。

entry#
readOnly.m和Map.dirty存储的值类型是*entry,它包含一个指针p, 指向用户存储的value值，结构如下：

Copy
type entry struct {
p unsafe.Pointer // *interface{}
}
p有三种值：

nil: entry已被删除了，并且m.dirty为nil
expunged: entry已被删除了，并且m.dirty不为nil，而且这个entry不存在于m.dirty中
其它： entry是一个正常的值
查找#
根据key来查找 value， 函数为 Load()，源码如下：

Copy
// src/sync/map.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 interface{}) (value interface{}, ok bool) {
// 首先从只读ready的map中查找，这时不需要加锁
read, _ := m.read.Load().(readOnly)
e, ok := read.m[key]

// 如果没有找到，并且read.amended为true，说明dirty中有新数据，从dirty中查找，开始加锁了
if !ok && read.amended {
    m.mu.Lock() // 加锁
    
   // 又在 readonly 中检查一遍，因为在加锁的时候 dirty 的数据可能已经迁移到了read中
    read, _ = m.read.Load().(readOnly)
    e, ok = read.m[key]
    
    // read 还没有找到，并且dirty中有数据
    if !ok && read.amended {
        e, ok = m.dirty[key] //从 dirty 中查找数据
        
        // 不管m.dirty中存不存在，都将misses + 1
        // missLocked() 中满足条件后就会把m.dirty中数据迁移到m.read中
        m.missLocked()
    }
    m.mu.Unlock()
}
if !ok {
    return nil, false
}
return e.load()

}
从函数可以看出，如果查询的键值正好在m.read中，不需要加锁，直接返回结果，优化了性能。
即使不在read中，经过几次miss后， m.dirty中的数据也会迁移到m.read中，这时又可以从read中查找。
所以对于更新／增加较少，加载存在的key很多的case，性能基本和无锁的map类似。

missLockerd() 迁移数据：

Copy
// src/sync/map.go

func (m *Map) missLocked() {
m.misses++
if m.misses < len(m.dirty) {//misses次数小于 dirty的长度，就不迁移数据，直接返回
return
}
m.read.Store(readOnly{m: m.dirty}) //开始迁移数据
m.dirty = nil //迁移完dirty就赋值为nil
m.misses = 0 //迁移完 misses归0
}
新增和更新#
方法是 Store()， 更新或者新增一个 entry， 源码如下：

Copy
// src/sync/map.go

// Store sets the value for a key.
func (m *Map) Store(key, value interface{}) {
// 直接在read中查找值，找到了，就尝试 tryStore() 更新值
read, _ := m.read.Load().(readOnly)
if e, ok := read.m[key]; ok && e.tryStore(&value) {
return
}

// m.read 中不存在
m.mu.Lock()
read, _ = m.read.Load().(readOnly)
if e, ok := read.m[key]; ok {
    if e.unexpungeLocked() { // 未被标记成删除，前面讲到entry数据结构时，里面的p值有3种。1.nil 2.expunged，这个值含义有点复杂，可以看看前面entry数据结构 3.正常值
        
        m.dirty[key] = e // 加入到dirty里
    }
    e.storeLocked(&value) // 更新值
} else if e, ok := m.dirty[key]; ok { // 存在于 dirty 中，直接更新
    e.storeLocked(&value)
} else { // 新的值
    if !read.amended { // m.dirty 中没有新数据，增加到 m.dirty 中
        // 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() // 从 m.read中复制未删除的数据
        m.read.Store(readOnly{m: read.m, amended: true}) 
    }
    m.dirty[key] = newEntry(value) //将这个entry加入到m.dirty中
}
m.mu.Unlock()

}
操作都是先从m.read开始，不满足条件再加锁，然后操作m.dirty。

删除#
根据key删除一个值：

Copy
// src/sync/map.go

// Delete deletes the value for a key.
func (m *Map) Delete(key interface{}) {
// 从 m.read 中开始查找
read, _ := m.read.Load().(readOnly)
e, ok := read.m[key]

if !ok && read.amended { // m.read中没有找到，并且可能存在于m.dirty中，加锁查找
    m.mu.Lock() // 加锁
    read, _ = m.read.Load().(readOnly) // 再在m.read中查找一次
    e, ok = read.m[key]
    if !ok && read.amended { //m.read中又没找到，amended标志位true，说明在m.dirty中
        delete(m.dirty, key) // 删除
    }
    m.mu.Unlock()
}
if ok { // 在 m.ready 中就直接删除
    e.delete()
}

}
还有更好的方法没？java里面有一个分段锁，保证在操作不同 map 段的时候， 可以并发执行， 操作同段 map 的时候，进行锁的竞争和等待。从而达到线程安全， 且效率大于 synchronized。而不是直接加一把大锁，锁住整个map。

那go里面有木有？有人已经想到了

concurrent-map#
项目地址：concurrent-map

中文wiki：地址

正如 这里 和 这里 所描述的, Go语言原生的map类型并不支持并发读写。concurrent-map提供了一种高性能的解决方案:通过对内部map进行分片，降低锁粒度，从而达到最少的锁等待时间(锁冲突)。

在Go 1.9之前，go语言标准库中并没有实现并发map。在Go 1.9中，引入了sync.Map。新的sync.Map与此concurrent-map有几个关键区别。标准库中的sync.Map是专为append-only场景设计的。因此，如果您想将Map用于一个类似内存数据库，那么使用我们的版本可能会受益。你可以在golang repo上读到更多，这里 and 这里

译注: sync.Map在读多写少性能比较好，否则并发性能很差

有兴趣的可以自己研究下。

参考：#
Go 1.9 sync.Map揭秘
go sync.Map使用和介绍
golang sync.map源码
https://www.dwell.com/@T5g8NivaH2
https://www.dwell.com/@T5g8NivaH2/followers
https://www.dwell.com/@T5g8NivaH2/following
https://www.dwell.com/@42kZxCq00l
https://www.dwell.com/@42kZxCq00l/followers
https://www.dwell.com/@42kZxCq00l/following
https://www.dwell.com/@PdPnK4JsS
https://www.dwell.com/@PdPnK4JsS/followers
https://www.dwell.com/@PdPnK4JsS/following
https://www.dwell.com/@A98IAzZFK61
https://www.dwell.com/@A98IAzZFK61/followers
https://www.dwell.com/@A98IAzZFK61/following
https://www.dwell.com/@KOOlxW0I6X
https://www.dwell.com/@KOOlxW0I6X/followers
https://www.dwell.com/@KOOlxW0I6X/following
https://www.dwell.com/@LQ7v3iO5H7
https://www.dwell.com/@LQ7v3iO5H7/followers
https://www.dwell.com/@LQ7v3iO5H7/following
https://www.dwell.com/@l4rvlkJ4xdj
https://www.dwell.com/@l4rvlkJ4xdj/followers
https://www.dwell.com/@l4rvlkJ4xdj/following
https://www.dwell.com/@5aMtK1yKh
https://www.dwell.com/@5aMtK1yKh/followers
https://www.dwell.com/@5aMtK1yKh/following
https://www.dwell.com/@19613Z86A
https://www.dwell.com/@19613Z86A/followers
https://www.dwell.com/@19613Z86A/following
https://www.dwell.com/@0KuZCI9uTf
https://www.dwell.com/@0KuZCI9uTf/followers
https://www.dwell.com/@0KuZCI9uTf/following
https://www.dwell.com/@986q4pWDl
https://www.dwell.com/@986q4pWDl/followers
https://www.dwell.com/@986q4pWDl/following
https://www.dwell.com/@a64EhdqVq0
https://www.dwell.com/@a64EhdqVq0/followers
https://www.dwell.com/@a64EhdqVq0/following
https://www.dwell.com/@cR6ov3Nt28
https://www.dwell.com/@cR6ov3Nt28/followers
https://www.dwell.com/@cR6ov3Nt28/following
https://www.dwell.com/@09cDO66M9UH
https://www.dwell.com/@09cDO66M9UH/followers
https://www.dwell.com/@09cDO66M9UH/following
https://www.dwell.com/@1p2K8lMkoN
https://www.dwell.com/@1p2K8lMkoN/followers
https://www.dwell.com/@1p2K8lMkoN/following
https://www.dwell.com/@658wPF1aZ
https://www.dwell.com/@658wPF1aZ/followers
https://www.dwell.com/@658wPF1aZ/following
https://www.dwell.com/@aLsyDo5MY4
https://www.dwell.com/@aLsyDo5MY4/followers
https://www.dwell.com/@aLsyDo5MY4/following
https://www.dwell.com/@LkZ8DhKn7
https://www.dwell.com/@LkZ8DhKn7/followers
https://www.dwell.com/@LkZ8DhKn7/following
https://www.dwell.com/@ZdGelHDn9
https://www.dwell.com/@ZdGelHDn9/followers
https://www.dwell.com/@ZdGelHDn9/following
https://www.dwell.com/@Q4Fzexcd6rV
https://www.dwell.com/@Q4Fzexcd6rV/followers
https://www.dwell.com/@Q4Fzexcd6rV/following
https://www.dwell.com/@7oMed47Pb
https://www.dwell.com/@7oMed47Pb/followers
https://www.dwell.com/@7oMed47Pb/following
https://www.dwell.com/@cU2yGoY0UyW
https://www.dwell.com/@cU2yGoY0UyW/followers
https://www.dwell.com/@cU2yGoY0UyW/following
https://www.dwell.com/@a6S0F3uI0X
https://www.dwell.com/@a6S0F3uI0X/followers
https://www.dwell.com/@a6S0F3uI0X/following
https://www.dwell.com/@aGDSsRJ1V
https://www.dwell.com/@aGDSsRJ1V/followers
https://www.dwell.com/@aGDSsRJ1V/following
https://www.dwell.com/@mf2N8f6ayU
https://www.dwell.com/@mf2N8f6ayU/followers
https://www.dwell.com/@mf2N8f6ayU/following
https://www.dwell.com/@fxwEaT9wc
https://www.dwell.com/@fxwEaT9wc/followers
https://www.dwell.com/@fxwEaT9wc/following
https://www.dwell.com/@ww0DcWC0yz
https://www.dwell.com/@ww0DcWC0yz/followers
https://www.dwell.com/@ww0DcWC0yz/following
https://www.dwell.com/@9FieGsC6Meb
https://www.dwell.com/@9FieGsC6Meb/followers
https://www.dwell.com/@9FieGsC6Meb/following
https://www.dwell.com/@7wP08277pH
https://www.dwell.com/@7wP08277pH/followers
https://www.dwell.com/@7wP08277pH/following
https://www.dwell.com/@AN96JRmgR
https://www.dwell.com/@AN96JRmgR/followers
https://www.dwell.com/@AN96JRmgR/following
https://www.dwell.com/@zklUAHiUbQ
https://www.dwell.com/@zklUAHiUbQ/followers
https://www.dwell.com/@zklUAHiUbQ/following
https://www.dwell.com/@v48Y2j9Ltgo
https://www.dwell.com/@v48Y2j9Ltgo/followers
https://www.dwell.com/@v48Y2j9Ltgo/following
https://www.dwell.com/@rG676NVTx
https://www.dwell.com/@rG676NVTx/followers
https://www.dwell.com/@rG676NVTx/following
https://www.dwell.com/@fMdZvO3rQQJ
https://www.dwell.com/@fMdZvO3rQQJ/followers
https://www.dwell.com/@fMdZvO3rQQJ/following
https://www.dwell.com/@0pwC06aVqB
https://www.dwell.com/@0pwC06aVqB/followers
https://www.dwell.com/@0pwC06aVqB/following
https://www.dwell.com/@z8C3RwHcf
https://www.dwell.com/@z8C3RwHcf/followers
https://www.dwell.com/@z8C3RwHcf/following
https://www.dwell.com/@82D2zC1Ai
https://www.dwell.com/@82D2zC1Ai/followers
https://www.dwell.com/@82D2zC1Ai/following
https://www.dwell.com/@Ek6xcXmK4
https://www.dwell.com/@Ek6xcXmK4/followers
https://www.dwell.com/@Ek6xcXmK4/following
https://www.dwell.com/@07eoVymQmWB
https://www.dwell.com/@07eoVymQmWB/followers
https://www.dwell.com/@07eoVymQmWB/following
https://www.dwell.com/@0cb3lC0uI4o
https://www.dwell.com/@0cb3lC0uI4o/followers
https://www.dwell.com/@0cb3lC0uI4o/following
https://www.dwell.com/@vr8n934091
https://www.dwell.com/@vr8n934091/followers
https://www.dwell.com/@vr8n934091/following
https://www.dwell.com/@O2id6kR0V6j
https://www.dwell.com/@O2id6kR0V6j/followers
https://www.dwell.com/@O2id6kR0V6j/following
https://www.dwell.com/@8jc5IM8n4h
https://www.dwell.com/@8jc5IM8n4h/followers
https://www.dwell.com/@8jc5IM8n4h/following
https://www.dwell.com/@a067IUFV2
https://www.dwell.com/@a067IUFV2/followers
https://www.dwell.com/@a067IUFV2/following
https://www.dwell.com/@oA6nY1kL4
https://www.dwell.com/@oA6nY1kL4/followers
https://www.dwell.com/@oA6nY1kL4/following
https://www.dwell.com/@6ciLyOo30
https://www.dwell.com/@6ciLyOo30/followers
https://www.dwell.com/@6ciLyOo30/following
https://www.dwell.com/@0AMDgU2mw
https://www.dwell.com/@0AMDgU2mw/followers
https://www.dwell.com/@0AMDgU2mw/following
https://www.dwell.com/@Pn4N1y5hp
https://www.dwell.com/@Pn4N1y5hp/followers
https://www.dwell.com/@Pn4N1y5hp/following
https://www.dwell.com/@fs99s450c1
https://www.dwell.com/@fs99s450c1/followers
https://www.dwell.com/@fs99s450c1/following
https://www.dwell.com/@Q6bNV9687r8
https://www.dwell.com/@Q6bNV9687r8/followers
https://www.dwell.com/@Q6bNV9687r8/following
https://www.dwell.com/@Fi3l7tWT20
https://www.dwell.com/@Fi3l7tWT20/followers
https://www.dwell.com/@Fi3l7tWT20/following
https://www.dwell.com/@TLt632mit
https://www.dwell.com/@TLt632mit/followers
https://www.dwell.com/@TLt632mit/following
https://www.dwell.com/@6dUiFlX2Wb3
https://www.dwell.com/@6dUiFlX2Wb3/followers
https://www.dwell.com/@6dUiFlX2Wb3/following
https://www.dwell.com/@567U60eras
https://www.dwell.com/@567U60eras/followers
https://www.dwell.com/@567U60eras/following
https://www.dwell.com/@H7N1l7Qlj
https://www.dwell.com/@H7N1l7Qlj/followers
https://www.dwell.com/@H7N1l7Qlj/following
https://www.dwell.com/@DXt4sw2p7g
https://www.dwell.com/@DXt4sw2p7g/followers
https://www.dwell.com/@DXt4sw2p7g/following
https://www.dwell.com/@uJO5efoRG
https://www.dwell.com/@uJO5efoRG/followers
https://www.dwell.com/@uJO5efoRG/following
https://www.dwell.com/@fTDF0c7NJ9Q
https://www.dwell.com/@fTDF0c7NJ9Q/followers
https://www.dwell.com/@fTDF0c7NJ9Q/following
https://www.dwell.com/@B7h268v991
https://www.dwell.com/@B7h268v991/followers
https://www.dwell.com/@B7h268v991/following
https://www.dwell.com/@7QDg8wDGMK9
https://www.dwell.com/@7QDg8wDGMK9/followers
https://www.dwell.com/@7QDg8wDGMK9/following
https://www.dwell.com/@225W3x243
https://www.dwell.com/@225W3x243/followers
https://www.dwell.com/@225W3x243/following
https://www.dwell.com/@rkHacO43l81
https://www.dwell.com/@rkHacO43l81/followers
https://www.dwell.com/@rkHacO43l81/following
https://www.dwell.com/@38b9HHx684
https://www.dwell.com/@38b9HHx684/followers
https://www.dwell.com/@38b9HHx684/following
https://www.dwell.com/@13PniF508
https://www.dwell.com/@13PniF508/followers
https://www.dwell.com/@13PniF508/following
https://www.dwell.com/@kvi3RX97BA8
https://www.dwell.com/@kvi3RX97BA8/followers
https://www.dwell.com/@kvi3RX97BA8/following
https://www.dwell.com/@aFJlvp0Zo
https://www.dwell.com/@aFJlvp0Zo/followers
https://www.dwell.com/@aFJlvp0Zo/following
https://www.dwell.com/@y6P2YzB7D
https://www.dwell.com/@y6P2YzB7D/followers
https://www.dwell.com/@y6P2YzB7D/following
https://www.dwell.com/@4DTtSi1v2
https://www.dwell.com/@4DTtSi1v2/followers
https://www.dwell.com/@4DTtSi1v2/following
https://www.dwell.com/@NWfWtJ8SxQ
https://www.dwell.com/@NWfWtJ8SxQ/followers
https://www.dwell.com/@NWfWtJ8SxQ/following
https://www.dwell.com/@XncA96kcvH9
https://www.dwell.com/@XncA96kcvH9/followers
https://www.dwell.com/@XncA96kcvH9/following
https://www.dwell.com/@3P0QQhXqOkL
https://www.dwell.com/@3P0QQhXqOkL/followers
https://www.dwell.com/@3P0QQhXqOkL/following
https://www.dwell.com/@4WrVAw9lke
https://www.dwell.com/@4WrVAw9lke/followers
https://www.dwell.com/@4WrVAw9lke/following
https://www.dwell.com/@QQ7Q51d6Ej9
https://www.dwell.com/@QQ7Q51d6Ej9/followers
https://www.dwell.com/@QQ7Q51d6Ej9/following
https://www.dwell.com/@iX7y60o74
https://www.dwell.com/@iX7y60o74/followers
https://www.dwell.com/@iX7y60o74/following
https://www.dwell.com/@Mu1iK8A06m
https://www.dwell.com/@Mu1iK8A06m/followers
https://www.dwell.com/@Mu1iK8A06m/following
https://www.dwell.com/@FeDuJ48w959
https://www.dwell.com/@FeDuJ48w959/followers
https://www.dwell.com/@FeDuJ48w959/following
https://www.dwell.com/@F353X5pZ5Xh
https://www.dwell.com/@F353X5pZ5Xh/followers
https://www.dwell.com/@F353X5pZ5Xh/following
https://www.dwell.com/@7cH8275Sl35
https://www.dwell.com/@7cH8275Sl35/followers
https://www.dwell.com/@7cH8275Sl35/following
https://www.dwell.com/@sNgcG9z3nzK
https://www.dwell.com/@sNgcG9z3nzK/followers
https://www.dwell.com/@sNgcG9z3nzK/following
https://www.dwell.com/@O5W095R2O
https://www.dwell.com/@O5W095R2O/followers
https://www.dwell.com/@O5W095R2O/following
https://www.dwell.com/@Bz0yWoqK6ef
https://www.dwell.com/@Bz0yWoqK6ef/followers
https://www.dwell.com/@Bz0yWoqK6ef/following
https://www.dwell.com/@Xfr3IgxCp2
https://www.dwell.com/@Xfr3IgxCp2/followers
https://www.dwell.com/@Xfr3IgxCp2/following
https://www.dwell.com/@si7YE32ZLq
https://www.dwell.com/@si7YE32ZLq/followers
https://www.dwell.com/@si7YE32ZLq/following
https://www.dwell.com/@d4zKvtOM4
https://www.dwell.com/@d4zKvtOM4/followers
https://www.dwell.com/@d4zKvtOM4/following
https://www.dwell.com/@48cmLhViH
https://www.dwell.com/@48cmLhViH/followers
https://www.dwell.com/@48cmLhViH/following
https://www.dwell.com/@4I0bmn5dn9
https://www.dwell.com/@4I0bmn5dn9/followers
https://www.dwell.com/@4I0bmn5dn9/following
https://www.dwell.com/@GmF0I6Rjo6
https://www.dwell.com/@GmF0I6Rjo6/followers
https://www.dwell.com/@GmF0I6Rjo6/following