ConcurrentHashMap understand the implementation principle, it is recommended to first understand the next HashMap implementation principle.
HashMap source parsing (JDK1.8)
Why ConcurrentHashMap
HashMap thread-safe, while Hashtable is thread-safe, but it uses a synchronized method synchronized, insert, read data are used in synchronized, when inserted into the data can not be read (the equivalent of the entire Hashtable are locked, full table lock), in the case of multi-threaded, to be competing for the same lock, resulting in extremely inefficient. In order to improve the pain points JDK1.5 after the Hashtable, ConcurrentHashMap came into being.
Why ConcurrentHashMap highly efficient?
JDK1.5 Realization
ConcurrentHashMap using technology segment lock, the lock ConcurrentHashMap a section of the store, and then to each piece of data with a lock (segment), when a thread holds a lock (segment) where the piece of data access, the other segments the data can also be accessed by another thread, the default assignment 16 segment. 16 times faster than Hashtable default improve efficiency.
ConcurrentHashMap structure diagram is as follows (Fig users contribute, Kazakhstan):
Paste_Image.png
JDK1.8 Realization
ConcurrentHashMap canceled the segment segmented lock, while the use of CAS and synchronized to ensure concurrency safety. HashMap1.8 data structure with the same structure, array + list / red and black binary tree .
synchronized only lock the current head node linked list or binary tree of red and black, so long as the hash does not conflict, will not produce concurrent, but also enhance the efficiency of N times.
Structure of FIG JDK1.8 ConcurrentHashMap follows:
Paste_Image.png
TreeBin: 红黑二叉树节点
Node: 链表节点
ConcurrentHashMap 源码分析
ConcurrentHashMap 类结构参照HashMap，这里列出HashMap没有的几个属性。
/**
* Table initialization and resizing control. When negative, the
* table is being initialized or resized: -1 for initialization,
* else -(1 + the number of active resizing threads). Otherwise,
* when table is null, holds the initial table size to use upon
* creation, or 0 for default. After initialization, holds the
* next element count value upon which to resize the table.
hash表初始化或扩容时的一个控制位标识量。
负数代表正在进行初始化或扩容操作
-1代表正在初始化
-N 表示有N-1个线程正在进行扩容操作
正数或0代表hash表还没有被初始化，这个数值表示初始化或下一次进行扩容的大小
*/
private transient volatile int sizeCtl;
// 以下两个是用来控制扩容的时候 单线程进入的变量
/**
* The number of bits used for generation stamp in sizeCtl.
* Must be at least 6 for 32bit arrays.
*/
private static int RESIZE_STAMP_BITS = 16;
/**
* The bit shift for recording size stamp in sizeCtl.
*/
private static final int RESIZE_STAMP_SHIFT = 32 - RESIZE_STAMP_BITS;
/*
* Encodings for Node hash fields. See above for explanation.
*/
static final int MOVED = -1; // hash值是-1，表示这是一个forwardNode节点
static final int TREEBIN = -2; // hash值是-2 表示这时一个TreeBin节点
分析代码主要目的：分析是如果利用CAS和Synchronized进行高效的同步更新数据。
下面插入数据源码：
public V put(K key, V value) {
return putVal(key, value, false);
}
/** Implementation for put and putIfAbsent */
final V putVal(K key, V value, boolean onlyIfAbsent) {
//ConcurrentHashMap 不允许插入null键，HashMap允许插入一个null键
if (key == null || value == null) throw new NullPointerException();
//计算key的hash值
int hash = spread(key.hashCode());
int binCount = 0;
//for循环的作用：因为更新元素是使用CAS机制更新，需要不断的失败重试，直到成功为止。
for (Node<K,V>[] tab = table;;) {
// f：链表或红黑二叉树头结点，向链表中添加元素时，需要synchronized获取f的锁。
Node<K,V> f; int n, i, fh;
//判断Node[]数组是否初始化，没有则进行初始化操作
if (tab == null || (n = tab.length) == 0)
tab = initTable();
//通过hash定位Node[]数组的索引坐标，是否有Node节点，如果没有则使用CAS进行添加（链表的头结点），添加失败则进入下次循环。
else if ((f = tabAt(tab, i = (n - 1) & hash)) == null) {
if (casTabAt(tab, i, null,
new Node<K,V>(hash, key, value, null)))
break; // no lock when adding to empty bin
}
//检查到内部正在移动元素（Node[] 数组扩容）
else if ((fh = f.hash) == MOVED)
//帮助它扩容
tab = helpTransfer(tab, f);
else {
V oldVal = null;
//锁住链表或红黑二叉树的头结点
synchronized (f) {
//判断f是否是链表的头结点
if (tabAt(tab, i) == f) {
//如果fh>=0 是链表节点
if (fh >= 0) {
binCount = 1;
//遍历链表所有节点
for (Node<K,V> e = f;; ++binCount) {
K ek;
//如果节点存在，则更新value
if (e.hash == hash &&
((ek = e.key) == key ||
(ek != null && key.equals(ek)))) {
oldVal = e.val;
if (!onlyIfAbsent)
e.val = value;
break;
}
//不存在则在链表尾部添加新节点。
Node<K,V> pred = e;
if ((e = e.next) == null) {
pred.next = new Node<K,V>(hash, key,
value, null);
break;
}
}
}
//TreeBin是红黑二叉树节点
else if (f instanceof TreeBin) {
Node<K,V> p;
binCount = 2;
//添加树节点
if ((p = ((TreeBin<K,V>)f).putTreeVal(hash, key,
value)) != null) {
oldVal = p.val;
if (!onlyIfAbsent)
p.val = value;
}
}
}
}
if (binCount != 0) {
//如果链表长度已经达到临界值8 就需要把链表转换为树结构
if (binCount >= TREEIFY_THRESHOLD)
treeifyBin(tab, i);
if (oldVal != null)
return oldVal;
break;
}
}
}
//将当前ConcurrentHashMap的size数量+1
addCount(1L, binCount);
return null;
}
- 判断Node[]数组是否初始化，没有则进行初始化操作
- 通过hash定位Node[]数组的索引坐标，是否有Node节点，如果没有则使用CAS进行添加（链表的头结点），添加失败则进入下次循环。
- 检查到内部正在扩容，如果正在扩容，就帮助它一块扩容。
- 如果f!=null，则使用synchronized锁住f元素（链表/红黑二叉树的头元素）
4.1 如果是Node(链表结构)则执行链表的添加操作。
4.2 如果是TreeNode(树型结果)则执行树添加操作。 - 判断链表长度已经达到临界值8 就需要把链表转换为树结构。
总结：
JDK8中的实现也是锁分离的思想，它把锁分的比segment（JDK1.5）更细一些，只要hash不冲突，就不会出现并发获得锁的情况。它首先使用无锁操作CAS插入头结点，如果插入失败，说明已经有别的线程插入头结点了，再次循环进行操作。如果头结点已经存在，则通过synchronized获得头结点锁，进行后续的操作。性能比segment分段锁又再次提升。