Netty中FastThreadLocal源码分析
Netty中使用FastThreadLocal替代JDK中的ThreadLocal【JAVA】ThreadLocal源码分析,其用法和ThreadLocal 一样,只不过从名字FastThreadLocal来看,其处理效率要比JDK中的ThreadLocal要高
在类加载的时候,先初始化了一个静态成员:
1 private static final int variablesToRemoveIndex = InternalThreadLocalMap.nextVariableIndex();
实际上FastThreadLocal的操作都是通过对InternalThreadLocalMap的操作来实现的,
而InternalThreadLocalMap是UnpaddedInternalThreadLocalMap的子类,UnpaddedInternalThreadLocalMap的定义比较简单:
1 class UnpaddedInternalThreadLocalMap { 2 static final ThreadLocal<InternalThreadLocalMap> slowThreadLocalMap = new ThreadLocal(); 3 static final AtomicInteger nextIndex = new AtomicInteger(); 4 Object[] indexedVariables; 5 int futureListenerStackDepth; 6 int localChannelReaderStackDepth; 7 Map<Class<?>, Boolean> handlerSharableCache; 8 IntegerHolder counterHashCode; 9 ThreadLocalRandom random; 10 Map<Class<?>, TypeParameterMatcher> typeParameterMatcherGetCache; 11 Map<Class<?>, Map<String, TypeParameterMatcher>> typeParameterMatcherFindCache; 12 StringBuilder stringBuilder; 13 Map<Charset, CharsetEncoder> charsetEncoderCache; 14 Map<Charset, CharsetDecoder> charsetDecoderCache; 15 ArrayList<Object> arrayList; 16 17 UnpaddedInternalThreadLocalMap(Object[] indexedVariables) { 18 this.indexedVariables = indexedVariables; 19 } 20 }
可以看到在类加载时,会初始化一个泛型为InternalThreadLocalMap的JDK的ThreadLocal对象作为其静态成员slowThreadLocalMap ,还有一个原子化的Integer静态成员nextIndex
InternalThreadLocalMap的定义如下:
1 public final class InternalThreadLocalMap extends UnpaddedInternalThreadLocalMap { 2 private static final InternalLogger logger = InternalLoggerFactory.getInstance(InternalThreadLocalMap.class); 3 private static final int DEFAULT_ARRAY_LIST_INITIAL_CAPACITY = 8; 4 private static final int STRING_BUILDER_INITIAL_SIZE = SystemPropertyUtil.getInt("io.netty.threadLocalMap.stringBuilder.initialSize", 1024); 5 private static final int STRING_BUILDER_MAX_SIZE; 6 public static final Object UNSET = new Object(); 7 private BitSet cleanerFlags;
InternalThreadLocalMap的nextVariableIndex方法:
1 public static int nextVariableIndex() { 2 int index = nextIndex.getAndIncrement(); 3 if (index < 0) { 4 nextIndex.decrementAndGet(); 5 throw new IllegalStateException("too many thread-local indexed variables"); 6 } else { 7 return index; 8 } 9 }
这是一个CAS滞后自增操作,获取nextIndex自增前的值,那么variablesToRemoveIndex初始化时就是0,且恒为0,nextIndex此时变成了1
FastThreadLocal对象的初始化:
1 private final int index = InternalThreadLocalMap.nextVariableIndex(); 2 3 public FastThreadLocal() { 4 }
由上面可知,index成员恒等于nextVariableIndex的返回值,nextIndex 的CAS操作保障了每个FastThreadLocal对象的index是不同的
首先看到set方法:
1 public final void set(V value) { 2 if (value != InternalThreadLocalMap.UNSET) { 3 InternalThreadLocalMap threadLocalMap = InternalThreadLocalMap.get(); 4 if (this.setKnownNotUnset(threadLocalMap, value)) { 5 this.registerCleaner(threadLocalMap); 6 } 7 } else { 8 this.remove(); 9 } 10 11 }
只要set的value不是InternalThreadLocalMap.UNSET,会先调用InternalThreadLocalMap的get方法:
1 public static InternalThreadLocalMap get() { 2 Thread thread = Thread.currentThread(); 3 return thread instanceof FastThreadLocalThread ? fastGet((FastThreadLocalThread)thread) : slowGet(); 4 }
判断当前线程是否是FastThreadLocalThread,是则调用fastGet,否则调用slowGet
FastThreadLocalThread是经过包装后的Thread:
1 public class FastThreadLocalThread extends Thread { 2 private final boolean cleanupFastThreadLocals; 3 private InternalThreadLocalMap threadLocalMap; 4 5 public FastThreadLocalThread() { 6 this.cleanupFastThreadLocals = false; 7 } 8 9 public FastThreadLocalThread(Runnable target) { 10 super(FastThreadLocalRunnable.wrap(target)); 11 this.cleanupFastThreadLocals = true; 12 } 13 14 public FastThreadLocalThread(ThreadGroup group, Runnable target) { 15 super(group, FastThreadLocalRunnable.wrap(target)); 16 this.cleanupFastThreadLocals = true; 17 } 18 19 public FastThreadLocalThread(String name) { 20 super(name); 21 this.cleanupFastThreadLocals = false; 22 } 23 24 public FastThreadLocalThread(ThreadGroup group, String name) { 25 super(group, name); 26 this.cleanupFastThreadLocals = false; 27 } 28 29 public FastThreadLocalThread(Runnable target, String name) { 30 super(FastThreadLocalRunnable.wrap(target), name); 31 this.cleanupFastThreadLocals = true; 32 } 33 34 public FastThreadLocalThread(ThreadGroup group, Runnable target, String name) { 35 super(group, FastThreadLocalRunnable.wrap(target), name); 36 this.cleanupFastThreadLocals = true; 37 } 38 39 public FastThreadLocalThread(ThreadGroup group, Runnable target, String name, long stackSize) { 40 super(group, FastThreadLocalRunnable.wrap(target), name, stackSize); 41 this.cleanupFastThreadLocals = true; 42 } 43 44 public final InternalThreadLocalMap threadLocalMap() { 45 return this.threadLocalMap; 46 } 47 48 public final void setThreadLocalMap(InternalThreadLocalMap threadLocalMap) { 49 this.threadLocalMap = threadLocalMap; 50 } 51 52 public boolean willCleanupFastThreadLocals() { 53 return this.cleanupFastThreadLocals; 54 } 55 56 public static boolean willCleanupFastThreadLocals(Thread thread) { 57 return thread instanceof FastThreadLocalThread && ((FastThreadLocalThread)thread).willCleanupFastThreadLocals(); 58 } 59 }
如果看过我之前写的ThreadLocal源码分析,看到这就明白,JDK的ThreadLocal中很重要的一点是在Thread类中有一个ThreadLocalMap类型的成员,每个线程都维护这一张ThreadLocalMap,通过ThreadLocalMap来和ThreadLocal对象产生映射关系;而这里和JDK同理绑定的就是InternalThreadLocalMap。
fastGet方法:
1 private static InternalThreadLocalMap fastGet(FastThreadLocalThread thread) { 2 InternalThreadLocalMap threadLocalMap = thread.threadLocalMap(); 3 if (threadLocalMap == null) { 4 thread.setThreadLocalMap(threadLocalMap = new InternalThreadLocalMap()); 5 } 6 7 return threadLocalMap; 8 }
这里也和JDK的ThreadLocal类似,判断FastThreadLocalThread 线程的threadLocalMap成员是否为null,若是null,则先创建一个InternalThreadLocalMap实例:
1 private InternalThreadLocalMap() { 2 super(newIndexedVariableTable()); 3 }
先调用newIndexedVariableTable方法:
1 private static Object[] newIndexedVariableTable() { 2 Object[] array = new Object[32]; 3 Arrays.fill(array, UNSET); 4 return array; 5 }
创建了一个大小为32的数组,并且用UNSET这个Object填充了整个数组,然后调用UnpaddedInternalThreadLocalMap的构造,令indexedVariables成员保存该数组
再来看slowGet方法:
1 private static InternalThreadLocalMap slowGet() { 2 ThreadLocal<InternalThreadLocalMap> slowThreadLocalMap = UnpaddedInternalThreadLocalMap.slowThreadLocalMap; 3 InternalThreadLocalMap ret = (InternalThreadLocalMap)slowThreadLocalMap.get(); 4 if (ret == null) { 5 ret = new InternalThreadLocalMap(); 6 slowThreadLocalMap.set(ret); 7 } 8 9 return ret; 10 }
可以看到,其实这里为了提高效率,并没有直接使用JDK的ThreadLocal,而是给当前非FastThreadLocalThread线程绑定了一个ThreadLocal<InternalThreadLocalMap>对象,避免直接使用JDK的ThreadLocal效率低。
回到FastThreadLocal的set方法,在取得到了当前线程的InternalThreadLocalMap成员后,调用setKnownNotUnset方法:
1 private boolean setKnownNotUnset(InternalThreadLocalMap threadLocalMap, V value) { 2 if (threadLocalMap.setIndexedVariable(this.index, value)) { 3 addToVariablesToRemove(threadLocalMap, this); 4 return true; 5 } else { 6 return false; 7 } 8 }
首先调用了InternalThreadLocalMap的setIndexedVariable方法:
1 public boolean setIndexedVariable(int index, Object value) { 2 Object[] lookup = this.indexedVariables; 3 if (index < lookup.length) { 4 Object oldValue = lookup[index]; 5 lookup[index] = value; 6 return oldValue == UNSET; 7 } else { 8 this.expandIndexedVariableTableAndSet(index, value); 9 return true; 10 } 11 }
因为index是不可更改的常量,所以这里有两种情况:
当indexedVariables这个Object数组的长度大于index时,直接将value放在indexedVariables数组下标为index的位置,返回oldValue是否等于UNSET,若是不等于UNSET,说明已经set过了,直进行替换,若是等于UNSET,还要进行后续的registerCleaner
当indexedVariables这个Object数组的长度小于等于index时,调用expandIndexedVariableTableAndSet方法扩容
expandIndexedVariableTableAndSet方法:
1 private void expandIndexedVariableTableAndSet(int index, Object value) { 2 Object[] oldArray = this.indexedVariables; 3 int oldCapacity = oldArray.length; 4 int newCapacity = index | index >>> 1; 5 newCapacity |= newCapacity >>> 2; 6 newCapacity |= newCapacity >>> 4; 7 newCapacity |= newCapacity >>> 8; 8 newCapacity |= newCapacity >>> 16; 9 ++newCapacity; 10 Object[] newArray = Arrays.copyOf(oldArray, newCapacity); 11 Arrays.fill(newArray, oldCapacity, newArray.length, UNSET); 12 newArray[index] = value; 13 this.indexedVariables = newArray; 14 }
如果读过HashMap源码的话对上述的位运算操作因该不陌生,这个位运算产生的newCapacity的值是大于oldCapacity的最小的二的整数幂(【Java】HashMap中的tableSizeFor方法)
然后申请一个newCapacity大小的数组,将原数组的内容拷贝到新数组,并且用UNSET填充剩余部分,还是将value放在下标为index的位置,用indexedVariables保存新数组。
setIndexedVariable成立后,setKnownNotUnset继续调用addToVariablesToRemove方法:
1 private static void addToVariablesToRemove(InternalThreadLocalMap threadLocalMap, FastThreadLocal<?> variable) { 2 Object v = threadLocalMap.indexedVariable(variablesToRemoveIndex); 3 Set variablesToRemove; 4 if (v != InternalThreadLocalMap.UNSET && v != null) { 5 variablesToRemove = (Set)v; 6 } else { 7 variablesToRemove = Collections.newSetFromMap(new IdentityHashMap()); 8 threadLocalMap.setIndexedVariable(variablesToRemoveIndex, variablesToRemove); 9 } 10 11 variablesToRemove.add(variable); 12 }
上面说过variablesToRemoveIndex恒为0,调用InternalThreadLocalMap的indexedVariable方法:
1 public Object indexedVariable(int index) { 2 Object[] lookup = this.indexedVariables; 3 return index < lookup.length ? lookup[index] : UNSET; 4 }
由于variablesToRemoveIndex恒等于0,所以这里判断indexedVariables这个Object数组是否为空,若是为空,则返回第0个元素,若不是则返回UNSET
在addToVariablesToRemove中,接着对indexedVariables的返回值进行了判断,
判断不是UNSET,并且不等于null,则说明是set过的,然后将刚才的返回值强转为Set类型
若上述条件不成立,创建一个IdentityHashMap,将其包装成Set赋值给variablesToRemove,然后调用InternalThreadLocalMap的setIndexedVariable方法,这里就和上面不一样了,上面是将value放在下标为index的位置,而这里是将Set放在下标为0的位置。
看到这,再结合上面来看,其实已经有一个大致的想法了,一开始在set时,是将value放在InternalThreadLocalMap的Object数组下标为index的位置,然后在这里获取下标为0的Set,说明value是暂时放在下标为index的位置,然后判断下标为0的位置有没有Set,若是有,取出这个Set ,将当前FastThreadLocal对象放入Set中,则说明这个Set中存放的是FastThreadLocal集合
那么就有如下关系:
回到FastThreadLocal的set方法,在setKnownNotUnset成立后,调用registerCleaner方法:
1 private void registerCleaner(InternalThreadLocalMap threadLocalMap) { 2 Thread current = Thread.currentThread(); 3 if (!FastThreadLocalThread.willCleanupFastThreadLocals(current) && !threadLocalMap.isCleanerFlagSet(this.index)) { 4 threadLocalMap.setCleanerFlag(this.index); 5 } 6 }
willCleanupFastThreadLocals的返回值在前面FastThreadLocalThread的初始化时就确定了,看到isCleanerFlagSet方法:
1 public boolean isCleanerFlagSet(int index) { 2 return this.cleanerFlags != null && this.cleanerFlags.get(index); 3 }
cleanerFlags 是一个BitSet对象,在InternalThreadLocalMap初始化时是null,
若不是第一次的set操作,则根据index,获取index在BitSet对应位的值
这里使用BitSet,使其持有的位和indexedVariables这个Object数组形成了一一对应关系,每一位都是0和1代表当前indexedVariables的对应下标位置的使用情况,0表示没有使用对应UNSET,1则代表有value
在上面条件成立的情况下,调用setCleanerFlag方法:
1 public void setCleanerFlag(int index) { 2 if (this.cleanerFlags == null) { 3 this.cleanerFlags = new BitSet(); 4 } 5 6 this.cleanerFlags.set(index); 7 }
逻辑比较简单,判断cleanerFlags是否初始化,若没有,则立即初始化,再将cleanerFlags中对应index位的值设为1;
这里通过registerCleaner直接标记了所有set了value的下标可,为以后的removeAll 清除提高效率。
下来看FastThreadLocal的get方法:
1 public final V get() { 2 InternalThreadLocalMap threadLocalMap = InternalThreadLocalMap.get(); 3 Object v = threadLocalMap.indexedVariable(this.index); 4 if (v != InternalThreadLocalMap.UNSET) { 5 return v; 6 } else { 7 V value = this.initialize(threadLocalMap); 8 this.registerCleaner(threadLocalMap); 9 return value; 10 } 11 }
和上面一样,先取得当前线程持有的InternalThreadLocalMap ,调用indexedVariable方法,根据当前FastThreadLocal的index定位,判断是否是UNSET(set过),若没有set过则和JDK一样调用initialize先set:
1 private V initialize(InternalThreadLocalMap threadLocalMap) { 2 Object v = null; 3 4 try { 5 v = this.initialValue(); 6 } catch (Exception var4) { 7 PlatformDependent.throwException(var4); 8 } 9 10 threadLocalMap.setIndexedVariable(this.index, v); 11 addToVariablesToRemove(threadLocalMap, this); 12 return v; 13 }
initialValue()方法就是对外提供的,需要手动覆盖:
1 protected V initialValue() throws Exception { 2 return null; 3 }
后面的操作就和set的逻辑一样。
remove方法:
1 public final void remove() { 2 this.remove(InternalThreadLocalMap.getIfSet()); 3 }
getIfSet方法:
1 public static InternalThreadLocalMap getIfSet() { 2 Thread thread = Thread.currentThread(); 3 return thread instanceof FastThreadLocalThread ? ((FastThreadLocalThread)thread).threadLocalMap() : (InternalThreadLocalMap)slowThreadLocalMap.get(); 4 }
和上面的get方法思路相似,只不过在这里如果获取不到不会创建
然后调用remove重载:
1 public final void remove(InternalThreadLocalMap threadLocalMap) { 2 if (threadLocalMap != null) { 3 Object v = threadLocalMap.removeIndexedVariable(this.index); 4 removeFromVariablesToRemove(threadLocalMap, this); 5 if (v != InternalThreadLocalMap.UNSET) { 6 try { 7 this.onRemoval(v); 8 } catch (Exception var4) { 9 PlatformDependent.throwException(var4); 10 } 11 } 12 13 } 14 }
先检查threadLocalMap是否存在,若存在才进行后续操作:
调用removeIndexedVariable方法:
1 public Object removeIndexedVariable(int index) { 2 Object[] lookup = this.indexedVariables; 3 if (index < lookup.length) { 4 Object v = lookup[index]; 5 lookup[index] = UNSET; 6 return v; 7 } else { 8 return UNSET; 9 } 10 }
和之前的setIndexedVariable逻辑相似,只不过现在是把index位置的元素设置为UNSET
接着调用removeFromVariablesToRemove方法:
1 private static void removeFromVariablesToRemove(InternalThreadLocalMap threadLocalMap, FastThreadLocal<?> variable) { 2 Object v = threadLocalMap.indexedVariable(variablesToRemoveIndex); 3 if (v != InternalThreadLocalMap.UNSET && v != null) { 4 Set<FastThreadLocal<?>> variablesToRemove = (Set)v; 5 variablesToRemove.remove(variable); 6 } 7 }
之前说过variablesToRemoveIndex恒为0,在Object数组中下标为0存储的Set<FastThreadLocal<?>>集合(不为UNSET情况下),从集合中,将当前FastThreadLocal移除掉
最后调用了onRemoval方法,该方法需要由用户去覆盖:
1 protected void onRemoval(V value) throws Exception { 2 }
removeAll方法,是一个静态方法:
1 public static void removeAll() { 2 InternalThreadLocalMap threadLocalMap = InternalThreadLocalMap.getIfSet(); 3 if (threadLocalMap != null) { 4 try { 5 Object v = threadLocalMap.indexedVariable(variablesToRemoveIndex); 6 if (v != null && v != InternalThreadLocalMap.UNSET) { 7 Set<FastThreadLocal<?>> variablesToRemove = (Set)v; 8 FastThreadLocal<?>[] variablesToRemoveArray = (FastThreadLocal[])variablesToRemove.toArray(new FastThreadLocal[0]); 9 FastThreadLocal[] var4 = variablesToRemoveArray; 10 int var5 = variablesToRemoveArray.length; 11 12 for(int var6 = 0; var6 < var5; ++var6) { 13 FastThreadLocal<?> tlv = var4[var6]; 14 tlv.remove(threadLocalMap); 15 } 16 } 17 } finally { 18 InternalThreadLocalMap.remove(); 19 } 20 21 } 22 }
首先获取当前线程的InternalThreadLocalMap,若是存在继续后续操作:
通过indexedVariable方法,取出Object数组中下标为0的Set集合(如果不是UNSET情况下),将其转换为FastThreadLocal数组,遍历这个数组调用上面的remove方法。
FastThreadLocal源码分析到此结束。