ThreadLocal原理分析與代碼驗證
- 2019 年 11 月 26 日
- 筆記
ThreadLocal提供了線程安全的數據存儲和訪問方式,利用不帶key的get和set方法,居然能做到線程之間隔離,非常神奇。
比如
ThreadLocal<String> threadLocal = new ThreadLocal<>();
in thread 1
//in thread1 treadLocal.set("value1"); ..... //value的值是value1 String value = threadLocal.get();
in thread 2
//in thread2 treadLocal.set("value2"); ..... //value的值是value2 String value = threadLocal.get();
不論thread1和thread2是不是同時執行,都不會有線程安全問題,我們來測試一下。
線程安全測試
開10個線程,每個線程內都對同一個ThreadLocal對象set不同的值,會發現ThreadLocal在每個線程內部get出來的值,只會是自己線程內set進去的值,不會被別的線程影響。
static void testUsage() throws InterruptedException { Utils.println("-------------testUsage-------------------"); ThreadLocal<Long> threadLocal = new ThreadLocal<>(); AtomicBoolean threadSafe = new AtomicBoolean(true); int count = 10; CountDownLatch countDownLatch = new CountDownLatch(count); Random random = new Random(736832); for (int i = 0; i < count; i ++){ new Thread(() -> { try { //生成一個隨機數 Long value = System.nanoTime() + random.nextInt(); threadLocal.set(value); Thread.sleep(1000); Long value2 = threadLocal.get(); if (!value.equals(value2)) { //get和set的value不一致,說明被別的線程修改了,但這是不可能出現的 threadSafe.set(false); Utils.println("thread unsafe, this could not be happen!"); } } catch (InterruptedException e) { }finally { countDownLatch.countDown(); } }).start(); } countDownLatch.await(); Utils.println("all thread done, and threadSafe is " + threadSafe.get()); Utils.println("------------------------------------------"); }
輸出:
-------------testUsage------------------ all thread done, and threadSafe is true -----------------------------------------
原理淺析
翻開ThreadLocal的源碼,會發現ThreadLocal只是一個空殼子,它並不存儲具體的value,而是利用當前線程(Thread.currentThread())的threadLocalMap來存儲value,key就是這個threadLocal對象本身。
public void set(T value) { Thread t = Thread.currentThread(); ThreadLocalMap map = getMap(t); if (map != null) map.set(this, value); else createMap(t, value); } ThreadLocalMap getMap(Thread t) { return t.threadLocals; }
Thread的threadLocals字段是ThreadLocalMap類型(你可以簡單理解為一個key value的Map),key是ThreadLocal對象,value是我們在外層設置的值
- 當我們調用threadLocal.set(value)方法的時候,會找到當前線程的threadLocals這個map,然後以this作為key去set key value
- 當我們調用threadLocal.get()方法的時候,會找到當前線程的threadLocals這個map,然後以this作為key去get value
- 當我們調用threadLocal.remove()方法的時候,會找到當前線程的threadLocals這個map,然後以this作為key去remove
這就相當於:
Thread.currentThread().threadLocals.set(threadLocal1, "value1"); ..... //value的值是value1 String value = Thread.currentThread().threadLocals.get(threadLocal1);
因為每個Thread都是不同的對象,所以他們的threadLocals也是不同的map,threadLocal在不同的線程里工作時,實際上是從不同的map里get/set,這也就是線程安全的原因了,了解到這一點就差不多了。
再深入一些,ThreadLocalMap的結構
如果繼續翻ThreadLocalMap的源碼,會發現它有個字段table,是Entry類型的數組。
我們不妨寫段代碼,把ThreadLocalMap的結構輸出出來。
由於Thread.threadLocals和ThreadLocalMap類不是public的,我們只有通過反射來獲取它的值。反射的代碼如下(如果嫌長可以不看,直接看輸出):
static Object getThreadLocalMap(Thread thread) throws NoSuchFieldException, IllegalAccessException { //get thread.threadLocals Field threadLocals = Thread.class.getDeclaredField("threadLocals"); threadLocals.setAccessible(true); return threadLocals.get(thread); } static void printThreadLocalMap(Object threadLocalMap) throws NoSuchFieldException, IllegalAccessException { String threadName = Thread.currentThread().getName(); if(threadLocalMap == null){ Utils.println("threadMap is null, threadName:" + threadName); return; } Utils.println(threadName); //get threadLocalMap.table Field tableField = threadLocalMap.getClass().getDeclaredField("table"); tableField.setAccessible(true); Object[] table = (Object[])tableField.get(threadLocalMap); Utils.println("----threadLocals (ThreadLocalMap), table.length = " + table.length); for (int i = 0; i < table.length; i ++){ WeakReference<ThreadLocal<?>> entry = (WeakReference<ThreadLocal<?>>)table[i]; printEntry(entry, i); } } static void printEntry(WeakReference<ThreadLocal<?>> entry, int i) throws NoSuchFieldException, IllegalAccessException { if(entry == null){ Utils.println("--------table[" + i + "] -> null"); return; } ThreadLocal key = entry.get(); //get entry.value Field valueField = entry.getClass().getDeclaredField("value"); valueField.setAccessible(true); Object value = valueField.get(entry); Utils.println("--------table[" + i + "] -> entry key = " + key + ", value = " + value); }
測試代碼:
static void testStructure() throws InterruptedException { Utils.println("-------------testStructure----------------"); ThreadLocal<String> threadLocal1 = new ThreadLocal<>(); ThreadLocal<String> threadLocal2 = new ThreadLocal<>(); Thread thread1 = new Thread(() -> { threadLocal1.set("threadLocal1-value"); threadLocal2.set("threadLocal2-value"); try { Object threadLocalMap = getThreadLocalMap(Thread.currentThread()); printThreadLocalMap(threadLocalMap); } catch (NoSuchFieldException | IllegalAccessException e) { e.printStackTrace(); } }, "thread1"); thread1.start(); //wait thread1 done thread1.join(); Thread thread2 = new Thread(() -> { threadLocal1.set("threadLocal1-value"); try { Object threadLocalMap = getThreadLocalMap(Thread.currentThread()); printThreadLocalMap(threadLocalMap); } catch (NoSuchFieldException | IllegalAccessException e) { e.printStackTrace(); } }, "thread2"); thread2.start(); thread2.join(); Utils.println("------------------------------------------"); }
我們在創建了兩個ThreadLocal的對象threadLocal1和threadLocal2,在線程1里為這兩個對象設置值,在線程2里只為threadLocal1設置值。然後分別打印出這兩個線程的threadLocalMap。
輸出結果為:
-------------testStructure---------------- thread1 ----threadLocals (ThreadLocalMap), table.length = 16 --------table[0] -> null --------table[1] -> entry key = java.lang.ThreadLocal@33baa315, value = threadLocal2-value --------table[2] -> null --------table[3] -> null --------table[4] -> null --------table[5] -> null --------table[6] -> null --------table[7] -> null --------table[8] -> null --------table[9] -> null --------table[10] -> entry key = java.lang.ThreadLocal@4d42db5c, value = threadLocal1-value --------table[11] -> null --------table[12] -> null --------table[13] -> null --------table[14] -> null --------table[15] -> null thread2 ----threadLocals (ThreadLocalMap), table.length = 16 --------table[0] -> null --------table[1] -> null --------table[2] -> null --------table[3] -> null --------table[4] -> null --------table[5] -> null --------table[6] -> null --------table[7] -> null --------table[8] -> null --------table[9] -> null --------table[10] -> entry key = java.lang.ThreadLocal@4d42db5c, value = threadLocal1-value --------table[11] -> null --------table[12] -> null --------table[13] -> null --------table[14] -> null --------table[15] -> null ------------------------------------------
從結果上可以看出:
- 線程1和線程2的threadLocalMap對象的table字段,是個數組,長度都是16
- 由於線程1里給兩個threadLocal對象設置了值,所以線程1的ThreadLocalMap里有兩個entry,數組下標分別是1和10,其餘的是null(如果你自己寫代碼驗證,下標不一定是1和10,不需要糾結這個問題,只要前後對的上就行)
- 由於線程2里只給一個threadLocal對象設置了值,所以線程1的ThreadLocalMap里只有一個entry,數組下標是10,其餘的是null
- threadLocal1這個對象在兩個線程里都設置了值,所以當它作為key加入二者的threadLocalMap時,key是一樣的,都是java.lang.ThreadLocal@4d42db5c;下標也是一樣的,都是10。
為什麼是WeakReference
查看Entry的源碼,會發現Entry繼承自WeakReference:
static class Entry extends WeakReference<ThreadLocal<?>> { /** The value associated with this ThreadLocal. */ Object value; Entry(ThreadLocal<?> k, Object v) { super(k); value = v; } }
構造函數里把key傳給了super,也就是說,ThreadLocalMap中對key的引用,是WeakReference的。
Weak reference objects, which do not prevent their referents from being made finalizable, finalized, and then reclaimed. Weak references are most often used to implement canonicalizing mappings.
通俗點解釋:
當一個對象僅僅被weak reference(弱引用), 而沒有任何其他strong reference(強引用)的時候, 不論當前的內存空間是否足夠,當GC運行的時候, 這個對象就會被回收。
看不明白沒關係,還是寫代碼測試一下什麼是WeakReference吧…
static void testWeakReference(){ Object obj1 = new Object(); Object obj2 = new Object(); WeakReference<Object> obj1WeakRef = new WeakReference<>(obj1); WeakReference<Object> obj2WeakRf = new WeakReference<>(obj2); //obj32StrongRef是強引用 Object obj2StrongRef = obj2; Utils.println("before gc: obj1WeakRef = " + obj1WeakRef.get() + ", obj2WeakRef = " + obj2WeakRf.get() + ", obj2StrongRef = " + obj2StrongRef); //把obj1和obj2設為null obj1 = null; obj2 = null; //強制gc forceGC(); Utils.println("after gc: obj1WeakRef = " + obj1WeakRef.get() + ", obj2WeakRef = " + obj2WeakRf.get() + ", obj2StrongRef = " + obj2StrongRef); }
結果輸出:
before gc: obj1WeakRef = java.lang.Object@4554617c, obj2WeakRef = java.lang.Object@74a14482, obj2StrongRef = java.lang.Object@74a14482 after gc: obj1WeakRef = null, obj2WeakRef = java.lang.Object@74a14482, obj2StrongRef = java.lang.Object@74a14482
從結果上可以看出:
- 我們先new了兩個對象(為避免混淆,稱他們為Object1和Object2),分別用變量obj1和obj2指向它們,同時定義了一個obj2StrongRef,也指向Object2,最後把obj1和obj2均指向null
- 由於Object1沒有變量強引用它了,所以在gc後,Object1被回收了,obj1WeakRef.get()返回了null
- 由於Object2還有obj2StrongRef在引用它,所以gc後,Object2依然存在,沒有被回收。
那麼,ThreadLocalMap中對key的引用,為什麼是WeakReference的呢?
因為大部分情況下,線程不死
大部分情況下,線程不會頻繁的創建和銷毀,一般都會用線程池。所以線程對象一般不會被清除,線程的threadLocalMap就一直存在。 如果key對ThreadLocal是強引用,那麼key永遠不會被回收,即使我們程序里再也不用它了。
但是key是弱引用的話,情況就會得到改善:只要沒有指向threadLocal的強引用了,這個ThreadLocal對象就會被清理。
我們還是寫代碼測試一下吧。
/** * 測試ThreadLocal對象什麼時候被回收 * @throws InterruptedException */ static void testGC() throws InterruptedException { Utils.println("-----------------testGC-------------------"); Thread thread1 = new Thread(() -> { ThreadLocal<String> threadLocal1 = new ThreadLocal<>(); ThreadLocal<String> threadLocal2 = new ThreadLocal<>(); threadLocal1.set("threadLocal1-value"); threadLocal2.set("threadLocal2-value"); try { Object threadLocalMap = getThreadLocalMap(Thread.currentThread()); Utils.println("print threadLocalMap before gc"); printThreadLocalMap(threadLocalMap); //set threadLocal1 unreachable threadLocal1 = null; forceGC(); Utils.println("print threadLocalMap after gc"); printThreadLocalMap(threadLocalMap); } catch (NoSuchFieldException | IllegalAccessException e) { e.printStackTrace(); } }, "thread1"); thread1.start(); thread1.join(); Utils.println("------------------------------------------"); }
我們在一個線程里為兩個ThreadLocal對象賦值,最後把其中一個對象的強引用移除,gc後打印當前線程的threadLocalMap。 輸出結果如下:
-----------------testGC------------------- print threadLocalMap before gc thread1 ----threadLocals (ThreadLocalMap), table.length = 16 --------table[0] -> null --------table[1] -> entry key = java.lang.ThreadLocal@7bf9cebf, value = threadLocal2-value --------table[2] -> null --------table[3] -> null --------table[4] -> null --------table[5] -> null --------table[6] -> null --------table[7] -> null --------table[8] -> null --------table[9] -> null --------table[10] -> entry key = java.lang.ThreadLocal@56342d38, value = threadLocal1-value --------table[11] -> null --------table[12] -> null --------table[13] -> null --------table[14] -> null --------table[15] -> null print threadLocalMap after gc thread1 ----threadLocals (ThreadLocalMap), table.length = 16 --------table[0] -> null --------table[1] -> entry key = java.lang.ThreadLocal@7bf9cebf, value = threadLocal2-value --------table[2] -> null --------table[3] -> null --------table[4] -> null --------table[5] -> null --------table[6] -> null --------table[7] -> null --------table[8] -> null --------table[9] -> null --------table[10] -> entry key = null, value = threadLocal1-value --------table[11] -> null --------table[12] -> null --------table[13] -> null --------table[14] -> null --------table[15] -> null ------------------------------------------
從輸出結果可以看到,當我們把threadLocal1的強引用移除並gc之後,table[10]的key變成了null,說明threadLocal1這個對象被回收了;threadLocal2的強引用還在,所以table[1]的key不是null,沒有被回收。
但是你發現沒有,table[10]的key雖然是null了,但value還活着! table[10]這個entry對象,也活着!
是的,因為只有key是WeakReference….
無用的entry什麼時候被回收?
通過查看ThreadLocal的源碼,發現在ThreadLocal對象的get/set/remove方法執行時,都有機會清除掉map中已經無用的entry。
最容易驗證清除無用entry的場景分別是:
- remove:這個不用說了,這哥們本來就是做這個的
- get:當一個新的threadLocal對象(沒有set過value)發生get調用時,也會作為新的entry加入map,在加入的過程中,有機會清除掉無用的entry,邏輯和下面的set相同。
- set: 當一個新的threadLocal對象(沒有set過value)發生set調用時,會在map中加入新的entry,此時有機會清除掉無用的entry,清除的邏輯是:
- 清除掉table數組中的那些無用entry中的一部分,記住是一部分,這個一部分可能全部,也可能是0,具體算法請看ThreadLocalMap.cleanSomeSlots,這裡不解釋了。
- 如果上一步的"一部分"是0(即清除了0個),並且map的size(是真實size,不是table.length)大於等於threshold(table.length的2/3),會執行一次rehash,在rehash的過程中,清理掉所有無用的entry,並減小size,清理後的size如果還大於等於threshold – threshold/4,則把table擴容為原來的兩倍大小。
還有其他場景,但不好驗證,這裡就不提了。
ThreadLocal源碼就不貼了,貼了也講不明白,相關邏輯在setInitialValue、cleanSomeSlots、expungeStaleEntries、rehash、resize等方法里。
在我們寫代碼驗證entry回收邏輯之前,還需要簡單的提一下ThreadLocalMap的hash算法。
entry數組的下標如何確定?
每個ThreadLocal對象,都有一個threadLocalHashCode變量,在加入ThreadLocalMap的時候,根據這個threadLocalHashCode的值,對entry數組的長度取余(hash & (len – 1)),餘數作為下標。
那麼threadLocalHashCode是怎麼計算的呢?看源碼:
public class ThreadLocal<T>{ private final int threadLocalHashCode = nextHashCode(); private static AtomicInteger nextHashCode = new AtomicInteger(); private static final int HASH_INCREMENT = 0x61c88647; private static int nextHashCode() { return nextHashCode.getAndAdd(HASH_INCREMENT); } ... }
ThreadLocal類維護了一個全局靜態字段nextHashCode,每new一個ThreadLocal對象,nextHashCode都會遞增0x61c88647,作為下一個ThreadLocal對象的threadLocalHashCode。
這個0x61c88647,是個神奇的數字,只要以它為遞增值,那麼和2的N次方取余時,在有限的次數內不會發生重複。 比如和16取余,那麼在16次遞增內,不會發生重複。還是寫代碼驗證一下吧。
int hashCode = 0; int HASH_INCREMENT = 0x61c88647; int length = 16; for(int i = 0; i < length ; i ++){ int h = hashCode & (length - 1); hashCode += HASH_INCREMENT; System.out.println("h = " + h + ", i = " + i); }
輸出結果為:
h = 0, i = 0 h = 7, i = 1 h = 14, i = 2 h = 5, i = 3 h = 12, i = 4 h = 3, i = 5 h = 10, i = 6 h = 1, i = 7 h = 8, i = 8 h = 15, i = 9 h = 6, i = 10 h = 13, i = 11 h = 4, i = 12 h = 11, i = 13 h = 2, i = 14 h = 9, i = 15
你看,h的值在16次遞增內,沒有發生重複。 但是要記住,2的N次方作為長度才會有這個效果,這也解釋了為什麼ThreadLocalMap的entry數組初始長度是16,每次都是2倍的擴容。
驗證新threadLocal的get和set時回收部分無效的entry
為了驗證出結果,我們需要先給ThreadLocal的nextHashCode重置一個初始值,這樣在測試的時候,每個threadLocal的數組下標才會按照我們設計的思路走。
static void resetNextHashCode() throws NoSuchFieldException, IllegalAccessException { Field nextHashCodeField = ThreadLocal.class.getDeclaredField("nextHashCode"); nextHashCodeField.setAccessible(true); nextHashCodeField.set(null, new AtomicInteger(1253254570)); }
然後在測試代碼里,我們先調用resetNextHashCode方法,然後加兩個ThreadLocal對象並set值,gc前把強引用去除,gc後再new兩個新的theadLocal對象,分別調用他們的get和set方法。 在每個關鍵點打印出threadLocalMap做比較。
static void testExpungeSomeEntriesWhenGetOrSet() throws InterruptedException { Utils.println("----------testExpungeStaleEntries----------"); Thread thread1 = new Thread(() -> { try { resetNextHashCode(); //注意,這裡必須有兩個ThreadLocal,才能驗證出threadLocal1被清理 ThreadLocal<String> threadLocal1 = new ThreadLocal<>(); ThreadLocal<String> threadLocal2 = new ThreadLocal<>(); threadLocal1.set("threadLocal1-value"); threadLocal2.set("threadLocal2-value"); Object threadLocalMap = getThreadLocalMap(Thread.currentThread()); //set threadLocal1 unreachable threadLocal1 = null; threadLocal2 = null; forceGC(); Utils.println("print threadLocalMap after gc"); printThreadLocalMap(threadLocalMap); ThreadLocal<String> newThreadLocal1 = new ThreadLocal<>(); newThreadLocal1.get(); Utils.println("print threadLocalMap after call a new newThreadLocal1.get"); printThreadLocalMap(threadLocalMap); ThreadLocal<String> newThreadLocal2 = new ThreadLocal<>(); newThreadLocal2.set("newThreadLocal2-value"); Utils.println("print threadLocalMap after call a new newThreadLocal2.set"); printThreadLocalMap(threadLocalMap); } catch (NoSuchFieldException | IllegalAccessException e) { e.printStackTrace(); } }, "thread1"); thread1.start(); thread1.join(); Utils.println("------------------------------------------"); }
程序輸出結果為:
----------testExpungeStaleEntries---------- print threadLocalMap after gc thread1 ----threadLocals (ThreadLocalMap), table.length = 16 --------table[0] -> null --------table[1] -> entry key = null, value = threadLocal2-value --------table[2] -> null --------table[3] -> null --------table[4] -> null --------table[5] -> null --------table[6] -> null --------table[7] -> null --------table[8] -> null --------table[9] -> null --------table[10] -> entry key = null, value = threadLocal1-value --------table[11] -> null --------table[12] -> null --------table[13] -> null --------table[14] -> null --------table[15] -> null print threadLocalMap after call a new newThreadLocal1.get thread1 ----threadLocals (ThreadLocalMap), table.length = 16 --------table[0] -> null --------table[1] -> entry key = null, value = threadLocal2-value --------table[2] -> null --------table[3] -> null --------table[4] -> null --------table[5] -> null --------table[6] -> null --------table[7] -> null --------table[8] -> entry key = java.lang.ThreadLocal@2b63dc81, value = null --------table[9] -> null --------table[10] -> null --------table[11] -> null --------table[12] -> null --------table[13] -> null --------table[14] -> null --------table[15] -> null print threadLocalMap after call a new newThreadLocal2.set thread1 ----threadLocals (ThreadLocalMap), table.length = 16 --------table[0] -> null --------table[1] -> null --------table[2] -> null --------table[3] -> null --------table[4] -> null --------table[5] -> null --------table[6] -> null --------table[7] -> null --------table[8] -> entry key = java.lang.ThreadLocal@2b63dc81, value = null --------table[9] -> null --------table[10] -> null --------table[11] -> null --------table[12] -> null --------table[13] -> null --------table[14] -> null --------table[15] -> entry key = java.lang.ThreadLocal@2e93c547, value = newThreadLocal2-value ------------------------------------------
從結果上來看,
- gc後table[1]和table[10]的key變成了null
- new newThreadLocal1.get後,新增了table[8],table[10]被清理了,但table[1]還在(這就是cleanSomeSlots中some的意思)
- new newThreadLocal2.set後,新增了table[15],table[1]被清理了。
驗證map的size大於等於table.length的2/3時回收所有無效的entry
static void testExpungeAllEntries() throws InterruptedException { Utils.println("----------testExpungeStaleEntries----------"); Thread thread1 = new Thread(() -> { try { resetNextHashCode(); int threshold = 16 * 2 / 3; ThreadLocal[] threadLocals = new ThreadLocal[threshold - 1]; for(int i = 0; i < threshold - 1; i ++){ threadLocals[i] = new ThreadLocal<String>(); threadLocals[i].set("threadLocal" + i + "-value"); } Object threadLocalMap = getThreadLocalMap(Thread.currentThread()); threadLocals[1] = null; threadLocals[8] = null; //threadLocals[6] = null; //threadLocals[4] = null; //threadLocals[2] = null; forceGC(); Utils.println("print threadLocalMap after gc"); printThreadLocalMap(threadLocalMap); ThreadLocal<String> newThreadLocal1 = new ThreadLocal<>(); newThreadLocal1.set("newThreadLocal1-value"); Utils.println("print threadLocalMap after call a new newThreadLocal1.get"); printThreadLocalMap(threadLocalMap); } catch (NoSuchFieldException | IllegalAccessException e) { e.printStackTrace(); } }, "thread1"); thread1.start(); thread1.join(); Utils.println("------------------------------------------"); }
我們先創建了9個threadLocal對象並設置了值,然後去掉了其中2個的強引用(注意這2個可不是隨意挑選的)。 gc後再添加一個新的threadLocal,最後打印出最新的map。輸出為:
----------testExpungeStaleEntries---------- print threadLocalMap after gc thread1 ----threadLocals (ThreadLocalMap), table.length = 16 --------table[0] -> null --------table[1] -> entry key = null, value = threadLocal1-value --------table[2] -> entry key = null, value = threadLocal8-value --------table[3] -> null --------table[4] -> entry key = java.lang.ThreadLocal@60523912, value = threadLocal6-value --------table[5] -> null --------table[6] -> entry key = java.lang.ThreadLocal@48fccd7a, value = threadLocal4-value --------table[7] -> null --------table[8] -> entry key = java.lang.ThreadLocal@188bbe72, value = threadLocal2-value --------table[9] -> null --------table[10] -> entry key = java.lang.ThreadLocal@19e0ebe8, value = threadLocal0-value --------table[11] -> entry key = java.lang.ThreadLocal@688bcb6f, value = threadLocal7-value --------table[12] -> null --------table[13] -> entry key = java.lang.ThreadLocal@46324c19, value = threadLocal5-value --------table[14] -> null --------table[15] -> entry key = java.lang.ThreadLocal@38f1283, value = threadLocal3-value print threadLocalMap after call a new newThreadLocal1.get thread1 ----threadLocals (ThreadLocalMap), table.length = 32 --------table[0] -> null --------table[1] -> null --------table[2] -> null --------table[3] -> null --------table[4] -> null --------table[5] -> null --------table[6] -> entry key = java.lang.ThreadLocal@48fccd7a, value = threadLocal4-value --------table[7] -> null --------table[8] -> null --------table[9] -> entry key = java.lang.ThreadLocal@1dae16b1, value = newThreadLocal1-value --------table[10] -> entry key = java.lang.ThreadLocal@19e0ebe8, value = threadLocal0-value --------table[11] -> null --------table[12] -> null --------table[13] -> entry key = java.lang.ThreadLocal@46324c19, value = threadLocal5-value --------table[14] -> null --------table[15] -> null --------table[16] -> null --------table[17] -> null --------table[18] -> null --------table[19] -> null --------table[20] -> entry key = java.lang.ThreadLocal@60523912, value = threadLocal6-value --------table[21] -> null --------table[22] -> null --------table[23] -> null --------table[24] -> entry key = java.lang.ThreadLocal@188bbe72, value = threadLocal2-value --------table[25] -> null --------table[26] -> null --------table[27] -> entry key = java.lang.ThreadLocal@688bcb6f, value = threadLocal7-value --------table[28] -> null --------table[29] -> null --------table[30] -> null --------table[31] -> entry key = java.lang.ThreadLocal@38f1283, value = threadLocal3-value ------------------------------------------
從結果上看:
- gc後table[1]和table[2](即threadLocal1和threadLocal8)的key變成了null
- 加入新的threadLocal後,table的長度從16變成了32(因為此時的size是8,正好等於10 – 10/4,所以擴容),並且threadLocal1和threadLocal8這兩個entry不見了。
如果在gc前,我們把threadLocals[1、8、6、4、2]都去掉強引用,加入新threadLocal後會發現1、8、6、4、2被清除了,但沒有擴容,因為此時size是5,小於10-10/4。這個邏輯就不貼測試結果了,你可以取消注釋上面代碼中相關的邏輯試試。
大部分場景下,ThreadLocal對象的生命周期是和app一致的,弱引用形同虛設
回到現實中。
我們用ThreadLocal的目的,無非是在跨方法調用時更方便的線程安全地存儲和使用變量。這就意味着ThreadLocal的生命周期很長,甚至和app是一起存活的,強引用一直在。
既然強引用一直存在,那麼弱引用就形同虛設了。
所以在確定不再需要ThreadLocal中的值的情況下,還是老老實實的調用remove方法吧!
代碼地址
https://github.com/kongxiangxin/pine/tree/master/threadlocal