Poison

ThreadLocal

发表于

在Thread 类的源码中，可以看到以下声明：

1
2
3
4
5
6
7
8
public
class Thread implements Runnable {

    /* ThreadLocal values pertaining to this thread. This map is maintained
     * by the ThreadLocal class. */
    ThreadLocal.ThreadLocalMap threadLocals = null;

}

可以看出，Thread 的实例持有了对 ThreadLocal.ThreadLocalMap 实例 threadLocals 的引用。

ThreadLocal#get
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
/**
 * Returns the value in the current thread's copy of this
 * thread-local variable.  If the variable has no value for the
 * current thread, it is first initialized to the value returned
 * by an invocation of the {@link #initialValue} method.
 *
 * @return the current thread's value of this thread-local
 */
public T get() {
    Thread t = Thread.currentThread();
    ThreadLocalMap map = getMap(t);
    if (map != null) {
        ThreadLocalMap.Entry e = map.getEntry(this);
        if (e != null) {
            @SuppressWarnings("unchecked")
            T result = (T)e.value;
            return result;
        }
    }
    return setInitialValue();
}

/**
 * Get the map associated with a ThreadLocal. Overridden in
 * InheritableThreadLocal.
 *
 * @param  t the current thread
 * @return the map
 */
ThreadLocalMap getMap(Thread t) {
    return t.threadLocals;
}

在 ThreadLocal 类的 get() 方法中，首先获取到当前线程 t，然后使用 getMap(Thread t) 方法获取到当前线程 t 中持有的 ThreadLocal.ThreadLocalMap 实例 threadLocals，然后再尝试从该 threadLocals 中获取当前 ThreadLocal 实例对应的值，如果 threadLocals 为空或者未获取到 ThreadLocal 实例对应的值，则执行以下的初始化逻辑：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
/**
 * Variant of set() to establish initialValue. Used instead
 * of set() in case user has overridden the set() method.
 *
 * @return the initial value
 */
private T setInitialValue() {
    T value = initialValue();
    Thread t = Thread.currentThread();
    ThreadLocalMap map = getMap(t);
    if (map != null)
        map.set(this, value);
    else
        createMap(t, value);
    return value;
}

protected T initialValue() {
    return null;
}

/**
 * Create the map associated with a ThreadLocal. Overridden in
 * InheritableThreadLocal.
 *
 * @param t the current thread
 * @param firstValue value for the initial entry of the map
 */
void createMap(Thread t, T firstValue) {
    t.threadLocals = new ThreadLocalMap(this, firstValue);
}

其中 initialValue() 是留给用户覆写的方法。注意，ThreadLocalMap 中的 key 为 ThreadLocal 实例。

ThreadLocal#set
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
/**
 * Sets the current thread's copy of this thread-local variable
 * to the specified value.  Most subclasses will have no need to
 * override this method, relying solely on the {@link #initialValue}
 * method to set the values of thread-locals.
 *
 * @param value the value to be stored in the current thread's copy of
 *        this thread-local.
 */
public void set(T value) {
    Thread t = Thread.currentThread();
    ThreadLocalMap map = getMap(t);
    if (map != null)
        map.set(this, value);
    else
        createMap(t, value);
}

ThreadLocal 类的 set(T value) 方法实现与 get() 方法类似，此处不再赘述。

ThreadLocal#remove
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
/**
 * Removes the current thread's value for this thread-local
 * variable.  If this thread-local variable is subsequently
 * {@linkplain #get read} by the current thread, its value will be
 * reinitialized by invoking its {@link #initialValue} method,
 * unless its value is {@linkplain #set set} by the current thread
 * in the interim.  This may result in multiple invocations of the
 * {@code initialValue} method in the current thread.
 *
 * @since 1.5
 */
 public void remove() {
     ThreadLocalMap m = getMap(Thread.currentThread());
     if (m != null)
         m.remove(this);
 }

ThreadLocal 类的 remove() 方法实现也很简单，核心就是获取出当前线程持有的 threadLocals 实例，然后在该实例上操作即可。通过以上三个操作 ThreadLocal 实例的方法，我们知道，其操作都依赖当前线程持有的 ThreadLocal.ThreadLocalMap 实例 threadLocals，且在 threadLocals 中，key 为 ThreadLocal 实例，即对应源码中的 this，value 为用户使用 ThreadLocal 来实际存储的对象的引用。

ThreadLocal.ThreadLocalMap

我们再看看 ThreadLocal 中用到的核心数据结构 ThreadLocalMap，该类作为 ThreadLocal 的静态内部类定义，如果没有看过 HashMap 实现原理的建议先看看 HashMap。回到 ThreadLocalMap，先看看 Entry 的定义：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
/**
 * The entries in this hash map extend WeakReference, using
 * its main ref field as the key (which is always a
 * ThreadLocal object).  Note that null keys (i.e. entry.get()
 * == null) mean that the key is no longer referenced, so the
 * entry can be expunged from table.  Such entries are referred to
 * as "stale entries" in the code that follows.
 */
static class Entry extends WeakReference<ThreadLocal<?>> {
    /** The value associated with this ThreadLocal. */
    Object value;

    Entry(ThreadLocal<?> k, Object v) {
        super(k);
        value = v;
    }
}

/**
 * The initial capacity -- MUST be a power of two.
 */
private static final int INITIAL_CAPACITY = 16;

/**
 * The table, resized as necessary.
 * table.length MUST always be a power of two.
 */
private Entry[] table;

可以看出，ThreadLocalMap 由 Entry 数组组成，初始容量为 16，且注释明确要求容量必须为 2 的 n 次幂，原因为当容量为 2 的 n 次幂时，定位元素所在数组下标的计算可以由 取余操作 优化为 与运算 实现，效率更高，关于 2 的 n 次幂带来的弊端可以参考 HashMap，此处不再赘述。仔细看 Entry 类的定义，其继承了 WeakReference 且将 Entry 的 key 作为弱引用，value 依然是强引用。再看看 ThreadLocalMap 的构造函数：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
/**
 * Construct a new map initially containing (firstKey, firstValue).
 * ThreadLocalMaps are constructed lazily, so we only create
 * one when we have at least one entry to put in it.
 */
ThreadLocalMap(ThreadLocal<?> firstKey, Object firstValue) {
    table = new Entry[INITIAL_CAPACITY];
    int i = firstKey.threadLocalHashCode & (INITIAL_CAPACITY - 1);
    table[i] = new Entry(firstKey, firstValue);
    size = 1;
    setThreshold(INITIAL_CAPACITY);
}

/**
 * The next size value at which to resize.
 */
private int threshold; // Default to 0

/**
 * Set the resize threshold to maintain at worst a 2/3 load factor.
 */
private void setThreshold(int len) {
    threshold = len * 2 / 3;
}

可以看出定位 key 所在数组下标采用了 与运算 实现，最后计算并设置了触发扩容的阈值，注意此处使用的负载因子为 2/3，原因可参考：Open Addressing。回到构造函数，其中不得不提的是 firstKey.threadLocalHashCode，我们看一下 ThreadLocal 中关于 threadLocalHashCode 的相关代码：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
/**
 * ThreadLocals rely on per-thread linear-probe hash maps attached
 * to each thread (Thread.threadLocals and
 * inheritableThreadLocals).  The ThreadLocal objects act as keys,
 * searched via threadLocalHashCode.  This is a custom hash code
 * (useful only within ThreadLocalMaps) that eliminates collisions
 * in the common case where consecutively constructed ThreadLocals
 * are used by the same threads, while remaining well-behaved in
 * less common cases.
 */
private final int threadLocalHashCode = nextHashCode();

/**
 * The next hash code to be given out. Updated atomically. Starts at
 * zero.
 */
private static AtomicInteger nextHashCode =
    new AtomicInteger();

/**
 * The difference between successively generated hash codes - turns
 * implicit sequential thread-local IDs into near-optimally spread
 * multiplicative hash values for power-of-two-sized tables.
 */
private static final int HASH_INCREMENT = 0x61c88647;

/**
 * Returns the next hash code.
 */
private static int nextHashCode() {
    return nextHashCode.getAndAdd(HASH_INCREMENT);
}

从注释中我们知道 ThreadLocalMap 采用线性探测处理 hash 冲突，每一个 ThreadLocal 实例的 threadLocalHashCode 都会使用线程安全的静态变量 nextHashCode 加上 0x61c88647 得到，那么这个累加的值是如何得到的呢？首先根据源码可以看出 threadLocalHashCode 字段定义的类型为 int，在 Java 中，基本类型是有符号的，那么可以知道 threadLocalHashCode 可以使用的范围为 [Integer.MIN_VALUE, Integer.MAX_VALUE]，这一段范围值的长度为 232 = 4294967296。我们将这段范围进行黄金分割，分割后较长的段长度为：(232) / ((1 + sqrt(5)) / 2) = 2654435769，较短的段长度为 4294967296 - 2654435769 = 1640531527。而 Java 中 Integer.MAX_VALUE 的十进制表示为 2147483647，较长段长度 2654435769 因为大于了 Integer.MAX_VALUE 无法用 int 直接表示，较短的段长度为 1640531527，可以用 int 表示，我们看看较长段长度 2654435769 的二进制表示：

1
2
2654435769:
10011110 00110111 01111001 10111001

可以看出如果把 2654435769 用 int 表示，则转换为十进制为 -1640531527，你会发现，数值部分竟然与较短段长度一致，这也是黄金分割点的神奇之处。将数值部分 1640531527 转换为十六进制表示则为 0x61c88647，与源码中的值一致，猜测这就是作者选择 0x61c88647 作为 HASH_INCREMENT 值的原因。即每个 ThreadLocal 实例的 threadLocalHashCode 相差的值为整个 int 范围段进行黄金分割后较短部分的长度。那么为何要对整数域进行黄金分割并选用上述计算的值作为 HASH_INCREMENT 呢？实际上这是一种特殊的乘法散列，只不过常数使用的 HASH_INCREMENT，也被称作斐波拉契哈希，其能保证在整个整数域上均匀分布，可参考：Fibonacci Hashing。当然，也有另一种说法是说采用的是黄金分割点的另一种变体，参见：What is the meaning of 0x61C88647 constant in ThreadLocal.java - Stack Overflow。

在 Kryo 的 ObjectIntMap 类中也可以看到类似的实现，只不过是根据 Long 进行的黄金分割：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
/** Returns an index >= 0 and <= {@link #mask} for the specified {@code item}.
* <p>
* The default implementation uses Fibonacci hashing on the item's {@link Object#hashCode()}: the hashcode is multiplied by a
* long constant (2 to the 64th, divided by the golden ratio) then the uppermost bits are shifted into the lowest positions to
* obtain an index in the desired range. Multiplication by a long may be slower than int (eg on GWT) but greatly improves
* rehashing, allowing even very poor hashcodes, such as those that only differ in their upper bits, to be used without high
* collision rates. Fibonacci hashing has increased collision rates when all or most hashcodes are multiples of larger
* Fibonacci numbers (see <a href=
* "https://probablydance.com/2018/06/16/fibonacci-hashing-the-optimization-that-the-world-forgot-or-a-better-alternative-to-integer-modulo/">Malte
* Skarupke's blog post</a>).
* <p>
* This method can be overriden to customizing hashing. This may be useful eg in the unlikely event that most hashcodes are
* Fibonacci numbers, if keys provide poor or incorrect hashcodes, or to simplify hashing if keys provide high quality
* hashcodes and don't need Fibonacci hashing: {@code return item.hashCode() & mask;} */
protected int place (K item) {
    return (int)(item.hashCode() * 0x9E3779B97F4A7C15L >>> shift);
}

我们再看看 ThreadLocalMap 的 set 方法实现，源码位于 ThreadLocal.java at jdk8-b120:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
/**
 * Set the value associated with key.
 *
 * @param key the thread local object
 * @param value the value to be set
 */
private void set(ThreadLocal<?> key, Object value) {

    // We don't use a fast path as with get() because it is at
    // least as common to use set() to create new entries as
    // it is to replace existing ones, in which case, a fast
    // path would fail more often than not.

    Entry[] tab = table;
    int len = tab.length;
    int i = key.threadLocalHashCode & (len-1);

    for (Entry e = tab[i];
         e != null;
         e = tab[i = nextIndex(i, len)]) {
        ThreadLocal<?> k = e.get();

        if (k == key) {
            e.value = value;
            return;
        }

        if (k == null) {
            replaceStaleEntry(key, value, i);
            return;
        }
    }

    tab[i] = new Entry(key, value);
    int sz = ++size;
    if (!cleanSomeSlots(i, sz) && sz >= threshold)
        rehash();
}

/**
 * Increment i modulo len.
 */
private static int nextIndex(int i, int len) {
    return ((i + 1 < len) ? i + 1 : 0);
}

可以看出，其在定位 tab 数组的索引时没有像 HashMap 一样引入防御式的哈希函数，因为当前实现的哈希值由以上我们分析的逻辑计算得出，不像 HashMap 受用户编写的 hashCode() 函数的质量影响。所以可以认为当前哈希值计算函数质量良好，此处直接使用了 key.threadLocalHashCode 进行槽位索引计算。对于冲突，采用的是步长为 1 的基于线性探测的解决方案，相比 HashMap 的单独链表法，具有更好的缓存性能。

注意其中的 k == null 判断，此时是在判断 Entry 实例的 key 即弱引用指向的对象是否已经为空，如果为空，说明已经被 GC，此时将调用 replaceStaleEntry 方法将 value 设置至索引 i 对应的槽位：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
/**
 * Replace a stale entry encountered during a set operation
 * with an entry for the specified key.  The value passed in
 * the value parameter is stored in the entry, whether or not
 * an entry already exists for the specified key.
 *
 * As a side effect, this method expunges all stale entries in the
 * "run" containing the stale entry.  (A run is a sequence of entries
 * between two null slots.)
 *
 * @param  key the key
 * @param  value the value to be associated with key
 * @param  staleSlot index of the first stale entry encountered while
 *         searching for key.
 */
private void replaceStaleEntry(ThreadLocal<?> key, Object value,
                               int staleSlot) {
    Entry[] tab = table;
    int len = tab.length;
    Entry e;

    // Back up to check for prior stale entry in current run.
    // We clean out whole runs at a time to avoid continual
    // incremental rehashing due to garbage collector freeing
    // up refs in bunches (i.e., whenever the collector runs).
    // 以上注释表明为了避免 GC 释放连续的引用导致持续增量的 rehash, 所以我们清理整个 run
    int slotToExpunge = staleSlot;
    for (int i = prevIndex(staleSlot, len); // 从 staleSlot 左侧的槽开始向左侧扫描
         (e = tab[i]) != null; // 当遇到空槽时停止扫描，这与方法注释上的 run 定义匹配，run 为左端空槽与右端空槽之前连续的 Entry 序列
         i = prevIndex(i, len))
        if (e.get() == null)
            slotToExpunge = i; // 当 Entry 实例的 key 指向的 ThreadLocal 实例已经被 GC 时，此时 e.get() 为空，将当前索引 i 赋值给 slotToExpunge

    // Find either the key or trailing null slot of run, whichever
    // occurs first
    // 以上注释表明遇到匹配的 key 或者 run 右侧的空槽时停止扫描
    for (int i = nextIndex(staleSlot, len); // 从 staleSlot 右侧的槽开始向右侧扫描
         (e = tab[i]) != null; // 当遇到空槽时停止扫描
         i = nextIndex(i, len)) {
        ThreadLocal<?> k = e.get();

        // If we find key, then we need to swap it
        // with the stale entry to maintain hash table order.
        // The newly stale slot, or any other stale slot
        // encountered above it, can then be sent to expungeStaleEntry
        // to remove or rehash all of the other entries in run.
        // 以上注释表明如果我们找到匹配的 key, 那么我们需要将当前 i 指向的 Entry 实例与 staleSlot 指向的实例进行交换以保持哈希表的顺序，整个 run 中废弃的 Entry 实例将被统一使用 expungeStaleEntry 方法移除
        if (k == key) { // 找到需要设置的 ThreadLocal 实例
            e.value = value; // 将值设置至当前 Entry 实例的 value 引用

            tab[i] = tab[staleSlot]; // 将 staleSlot 指向的废弃 Entry 赋值给当前的 tab[i]
            tab[staleSlot] = e; // 将 tab[i] 被赋值前的 Entry 实例赋值给 tab[staleSlot]

            // 至此 tab[staleSlot] 上的 Entry 为存在 value 值的 Entry, tab[i] 指向废弃的 Entry

            // Start expunge at preceding stale entry if it exists
            if (slotToExpunge == staleSlot) // 如果 slotToExpunge 与 staleSlot 相等则说明 staleSlot 左侧没有废弃的 Entry 实例，此时需要从 i 指向的 Entry 开始清理，因为我们从 staleSlot 向右扫描至 i 发现了需要设置的 Entry, 且该 Entry 已经被交换至了 staleSlot, 所以此时需要将 i 赋值给 slotToExpunge
                slotToExpunge = i;
            cleanSomeSlots(expungeStaleEntry(slotToExpunge), len);
            return; // 注意此处，如果我们找到匹配的 key, 执行完上述逻辑后方法从此处返回
        }

        // If we didn't find stale entry on backward scan, the
        // first stale entry seen while scanning for key is the
        // first still present in the run.
        // 如果在之前左侧扫描的过程中没有发现废弃的 Entry, 即 slotToExpunge 与 staleSlot 相等时，当 k 为空时，此时记录下首个废弃的 Entry, 即记录下右侧扫描过程中的首个废弃的 Entry
        if (k == null && slotToExpunge == staleSlot)
            slotToExpunge = i; // 注意此处 slotToExpunge 被赋值后就不会再满足 slotToExpunge == staleSlot 的条件了，即仅将首个废弃的 Entry 的索引赋值给 slotToExpunge
    }

    // If key not found, put new entry in stale slot
    // 执行至此处说明 staleSlot 右侧连续的 Entry 实例中的 key 均不是需要设置的 key, 此时在 staleSlot 索引上创建新的 Entry
    tab[staleSlot].value = null; // 解除强引用，帮助 GC
    tab[staleSlot] = new Entry(key, value); // 新建 Entry 实例并设置至 tab[staleSlot]

    // If there are any other stale entries in run, expunge them
    // 如果 slotToExpunge 不等于 staleSlot, 则说明 staleSlot 左侧存在废弃的 Entry, 此时从 slotToExpunge 开始清理
    if (slotToExpunge != staleSlot)
        cleanSomeSlots(expungeStaleEntry(slotToExpunge), len);
}

根据代码中的注释我们知道，该方法除了设置值至对应的 Slot 外，还会清除废弃的 Entry 实例。接着我们看看在上面方法中调用的 cleanSomeSlots(expungeStaleEntry(slotToExpunge), len); 中的两个方法，先看 expungeStaleEntry(int staleSlot) 方法：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
/**
 * Expunge a stale entry by rehashing any possibly colliding entries
 * lying between staleSlot and the next null slot.  This also expunges
 * any other stale entries encountered before the trailing null.  See
 * Knuth, Section 6.4
 *
 * @param staleSlot index of slot known to have null key
 * @return the index of the next null slot after staleSlot
 * (all between staleSlot and this slot will have been checked
 * for expunging).
 */
private int expungeStaleEntry(int staleSlot) {
    Entry[] tab = table;
    int len = tab.length;

    // expunge entry at staleSlot
    tab[staleSlot].value = null;
    tab[staleSlot] = null;
    size--;

    // Rehash until we encounter null
    Entry e;
    int i;
    for (i = nextIndex(staleSlot, len);
         (e = tab[i]) != null; // 遇到空槽时停下
         i = nextIndex(i, len)) {
        ThreadLocal<?> k = e.get();
        if (k == null) { // 如果 k 指向的 ThreadLocal 实例已经被 GC
            e.value = null; // 解除 value 的强引用
            tab[i] = null; // 当前槽设置为空
            size--; // size 减 1
        } else {
            int h = k.threadLocalHashCode & (len - 1); // 重新计算槽索引
            if (h != i) { // 如果重新计算的槽索引不为 i, 则需要将 tab[i] 存放至 tab[h], 且需要处理 tab[h] 冲突的情况
                tab[i] = null; // 因为准备将 tab[i] 存储至 tab[h], 且 tab[i] 已经赋值给 e, 所以此处将 tab[i] 设置为空

                // Unlike Knuth 6.4 Algorithm R, we must scan until
                // null because multiple entries could have been stale.
                while (tab[h] != null) // 当 tab[h] 不为空时，说明冲突，继续向后探测直到 tab[h] 为空
                    h = nextIndex(h, len);
                tab[h] = e; // 将原 tab[i] 赋值给 tab[h]
            }
        }
    }
    return i; // 返回空槽的索引
}

可见 expungeStaleEntry(int staleSlot) 方法的实现还是比较简单的，再看看 cleanSomeSlots(int i, int n) 方法的实现：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
/**
 * Heuristically scan some cells looking for stale entries.
 * This is invoked when either a new element is added, or
 * another stale one has been expunged. It performs a
 * logarithmic number of scans, as a balance between no
 * scanning (fast but retains garbage) and a number of scans
 * proportional to number of elements, that would find all
 * garbage but would cause some insertions to take O(n) time.
 *
 * @param i a position known NOT to hold a stale entry. The
 * scan starts at the element after i.
 *
 * @param n scan control: {@code log2(n)} cells are scanned,
 * unless a stale entry is found, in which case
 * {@code log2(table.length)-1} additional cells are scanned.
 * When called from insertions, this parameter is the number
 * of elements, but when from replaceStaleEntry, it is the
 * table length. (Note: all this could be changed to be either
 * more or less aggressive by weighting n instead of just
 * using straight log n. But this version is simple, fast, and
 * seems to work well.)
 *
 * @return true if any stale entries have been removed.
 */
private boolean cleanSomeSlots(int i, int n) {
    boolean removed = false;
    Entry[] tab = table;
    int len = tab.length;
    do {
        i = nextIndex(i, len);
        Entry e = tab[i];
        if (e != null && e.get() == null) {
            n = len;
            removed = true;
            i = expungeStaleEntry(i);
        }
    } while ( (n >>>= 1) != 0);
    return removed;
}

该方法比较简单，且注释非常清楚，启发式扫描，在不扫描与元素数量成比例扫描之前取得平衡。此时我们回到 set 方法：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
/**
 * Set the value associated with key.
 *
 * @param key the thread local object
 * @param value the value to be set
 */
private void set(ThreadLocal<?> key, Object value) {

    // We don't use a fast path as with get() because it is at
    // least as common to use set() to create new entries as
    // it is to replace existing ones, in which case, a fast
    // path would fail more often than not.

    Entry[] tab = table;
    int len = tab.length;
    int i = key.threadLocalHashCode & (len-1);

    for (Entry e = tab[i];
         e != null;
         e = tab[i = nextIndex(i, len)]) {
        ThreadLocal<?> k = e.get();

        if (k == key) {
            e.value = value;
            return;
        }

        if (k == null) {
            replaceStaleEntry(key, value, i);
            return;
        }
    }

    tab[i] = new Entry(key, value);
    int sz = ++size;
    if (!cleanSomeSlots(i, sz) && sz >= threshold)
        rehash();
}

可知，如果哈希表中之前不存在相同的 key 且没有废弃的 Entry 实例可以用于存放当前需要设置的 key，那么此时需要创建新的 Entry 实例并设置至 tab[i]，设置并更新 size 之后将会调用 cleanSomeSlots 方法清理部分槽位，如果连一个槽位也没有得到清理，那么此时将比较 sz 是否大于等于 threshold，即如果当前 size 已经达到阈值，则调用 rehash() 尝试进行重新哈希：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
/**
 * Re-pack and/or re-size the table. First scan the entire
 * table removing stale entries. If this doesn't sufficiently
 * shrink the size of the table, double the table size.
 */
private void rehash() {
    expungeStaleEntries(); // 清除 tab 数组中所有废弃的 Entry 实例

    // Use lower threshold for doubling to avoid hysteresis
    if (size >= threshold - threshold / 4)
        // 注意此处 size 为清除所有废弃 Entry 实例的个数，此处与 threshold - threshold / 4 进行比较
        // 在构造函数中我们知道 threshold = 2/3 * capacity, 那么此时的比较可以简化为 size >= capacity/2
        resize();
}

/**
 * Double the capacity of the table.
 */
private void resize() {
    Entry[] oldTab = table;
    int oldLen = oldTab.length;
    int newLen = oldLen * 2; // 双倍扩容，保证长度为 2 的 n 次幂
    Entry[] newTab = new Entry[newLen];
    int count = 0;

    for (int j = 0; j < oldLen; ++j) {
        Entry e = oldTab[j];
        if (e != null) {
            ThreadLocal<?> k = e.get();
            if (k == null) {
                // 当前的 Entry 实例 e 为废弃的实例，无需迁移至新数组，所以仅解除了强引用以帮助 GC
                e.value = null; // Help the GC
            } else {
                int h = k.threadLocalHashCode & (newLen - 1);
                while (newTab[h] != null) // 处理冲突
                    h = nextIndex(h, newLen);
                newTab[h] = e; // 找到空槽位，赋值
                count++; // 维护 count
            }
        }
    }

    setThreshold(newLen);
    size = count;
    table = newTab;
}

/**
 * Expunge all stale entries in the table.
 */
private void expungeStaleEntries() {
    Entry[] tab = table;
    int len = tab.length;
    for (int j = 0; j < len; j++) {
        Entry e = tab[j];
        if (e != null && e.get() == null)
            expungeStaleEntry(j);
    }
}

以上几个方法的实现都比较简单，注释直接写在源码中了，此处不再赘述。

关于该类的使用场景，可以看几个例子：

  • 在 DataSourceTransactionManager.java at v5.3.15 中，会将获取到的 JDBC 连接进行包装并绑定至当前线程，核心代码如下：

    1
    2
    3
    4
    // Bind the connection holder to the thread.
    if (txObject.isNewConnectionHolder()) {
    TransactionSynchronizationManager.bindResource(obtainDataSource(), txObject.getConnectionHolder());
    }

    TransactionSynchronizationManager 的部分源码位于 TransactionSynchronizationManager.java at v5.3.15:

    1
    2
    3
    4
    5
    6
    7
    8
    9
    10
    11
    12
    13
    14
    15
    16
    17
    18
    19
    20
    21
    22
    23
    24
    25
    26
    27
    28
    29
    private static final ThreadLocal<Map<Object, Object>> resources =
        new NamedThreadLocal<>("Transactional resources");

    /**
    * Bind the given resource for the given key to the current thread.
    * @param key the key to bind the value to (usually the resource factory)
    * @param value the value to bind (usually the active resource object)
    * @throws IllegalStateException if there is already a value bound to the thread
    * @see ResourceTransactionManager#getResourceFactory()
    */
    public static void bindResource(Object key, Object value) throws IllegalStateException {
    Object actualKey = TransactionSynchronizationUtils.unwrapResourceIfNecessary(key);
    Assert.notNull(value, "Value must not be null");
    Map<Object, Object> map = resources.get();
    // set ThreadLocal Map if none found
    if (map == null) {
        map = new HashMap<>();
        resources.set(map);
    }
    Object oldValue = map.put(actualKey, value);
    // Transparently suppress a ResourceHolder that was marked as void...
    if (oldValue instanceof ResourceHolder && ((ResourceHolder) oldValue).isVoid()) {
        oldValue = null;
    }
    if (oldValue != null) {
        throw new IllegalStateException(
            "Already value [" + oldValue + "] for key [" + actualKey + "] bound to thread");
    }
    }

  • 在 StringCoding 中，使用 ThreadLocal 保证了字符串编码器与解码器的线程安全。在 HikariCP 连接池的核心类 ConcurrentBag.java 中也有对 ThreadLocal 的使用。

  • 在 Dubbo 的 ThreadLocalKryoFactory 类中，使用 ThreadLocal 为每个线程创建 Kryo 实例以规避 Kryo 实例被多线程操作的线程安全问题，因为 Kryo 实例不是线程安全的，参见：Thread safety.

Reference

ThreadLocal (Java Platform SE 8 )
Golden ratio - Wikipedia
Why should Java ThreadLocal variables be static - Stack Overflow
Fibonacci Hashing: The Optimization that the World Forgot (or: a Better Alternative to Integer Modulo) | Probably Dance
Why does ThreadLocalMap.Entry extend WeakReference? - Stack Overflow