思路
包括双向队列,让任务从一个方向出来,一个方向进入,相当于就是模仿消息队列,并且需要一个ReentrantLock来保证消息出去的是一个,存入消息的时候也只能有一个线程操作。而且需要两个条件变量控制消费者线程和生产者线程的消费和生产的数量以及阻塞时机。
class BlockedQueue<T>{
//阻塞队列
private Deque<T> queue=new ArrayDeque<>();
//锁
private ReentrantLock lock=new ReentrantLock();
//空队列条件
private Condition emptyCondition=lock.newCondition();
//满队列条件
private Condition fullCondition=lock.newCondition();
private int capcity;
public BlockedQueue(int capcity) {
this.capcity = capcity;
}
public T take(){
lock.lock();
try{
while(queue.isEmpty()){
try {
emptyCondition.await();
} catch (InterruptedException e) {
e.printStackTrace();
}
}
T t = queue.removeFirst();
//唤醒满队列
fullCondition.signal();
return t;
}finally {
lock.unlock();
}
}
public void put(T t){
lock.lock();
try{
while(queue.size()==capcity){
try {
fullCondition.await();
} catch (InterruptedException e) {
e.printStackTrace();
}
}
queue.addLast(t);
emptyCondition.signal();
}finally {
lock.unlock();
}
}
public int getCapcity(){
lock.lock();
try{
return this.capcity;
}finally {
lock.unlock();
}
}
}
设置等待时间,通过TimeUnit把秒转换成纳秒,最后awaitNacos返回的是还需要等待的时间,如果小于0那么就要结束等待。
public T poll(long timeout, TimeUnit unit){
lock.lock();
try{
long nacos = unit.toNanos(timeout);
while(queue.isEmpty()){
try {
//等待超时,后自动解锁。
if(nacos<=0){
return null;
}
nacos= emptyCondition.awaitNanos(nacos);
} catch (InterruptedException e) {
e.printStackTrace();
}
}
T t = queue.removeFirst();
//唤醒满队列
fullCondition.signal();
return t;
}finally {
lock.unlock();
}
}
包括核心线程数,任务阻塞队列,任务超时时间等。
class ThreadPool{
private BlockedQueue<Runnable> queue;
private HashSet<Worker> workers=new HashSet<>();
private int coreSize;
private long timeout;
private TimeUnit timeUnit;
/**
* 初始化
* @param coreSize
* @param timeout
* @param timeUnit
* @param queueCapcity
*/
public ThreadPool(int coreSize, long timeout, TimeUnit timeUnit,int queueCapcity) {
this.coreSize = coreSize;
this.timeout = timeout;
this.timeUnit = timeUnit;
queue=new BlockedQueue<Runnable>(queueCapcity);
}
class Worker{
}
}
public void execute(Runnable task){
synchronized (this){
//保证只有一个线程能够被创建
if(workers.size()<coreSize){
Worker worker = new Worker(task);
workers.add(worker);
}else{
//如果没有线程处理那么就放入到阻塞队列
queue.put(task);
}
}
}
class Worker extends Thread{
private Runnable task;
public Worker(Runnable task) {
this.task = task;
}
@Override
public void run() {
while(task!=null||(task=queue.take())!=null){
try {
task.run();
} catch (Exception e) {
e.printStackTrace();
}finally {
task=null;
}
}
//执行完任务之后立刻移除当前线程。
synchronized (workers){
workers.remove(this);
}
}
}
创建线程池之后,然后for循环让它执行任务,如果核心空闲的线程有那么就创建并且执行任务,如果没有多余的线程那么进入消息队列等待。线程空闲之后访问消息队列是不是还有任务,如果有那么就执行,没有就退出,并且退出线程池。
@Slf4j(topic = "c.test")
public class MyTestThreadPool {
public static void main(String[] args) {
ThreadPool threadPool = new ThreadPool(2, 1000, TimeUnit.MILLISECONDS, 10);
for(int i=0;i<5;i++){
int j=i;
threadPool.execute(()->{
log.debug("执行任务{}",j);
});
}
}
}
@Slf4j(topic = "c.pool")
class ThreadPool{
private BlockedQueue<Runnable> queue;
private HashSet<Worker> workers=new HashSet<>();
private int coreSize;
private long timeout;
private TimeUnit timeUnit;
public ThreadPool(int coreSize, long timeout, TimeUnit timeUnit,int queueCapcity) {
this.coreSize = coreSize;
this.timeout = timeout;
this.timeUnit = timeUnit;
queue=new BlockedQueue<Runnable>(queueCapcity);
}
public void execute(Runnable task){
synchronized (this){
//保证只有一个线程能够被创建
if(workers.size()<coreSize){
Worker worker = new Worker(task);
log.debug("线程和任务创建{},{}",worker,task);
workers.add(worker);
worker.start();
}else{
log.debug("加入到任务队列{}",task);
//如果没有线程处理那么就放入到阻塞队列
queue.put(task);
}
}
}
// @Slf4j(topic = "worker")
class Worker extends Thread{
private Runnable task;
public Worker(Runnable task) {
this.task = task;
}
@Override
public void run() {
while(task!=null||(task=queue.poll(timeout,timeUnit))!=null){
try {
log.debug("任务被执行{}",task);
task.run();
} catch (Exception e) {
e.printStackTrace();
}finally {
task=null;
}
}
//执行完任务之后立刻移除当前线程。
synchronized (workers){
log.debug("线程完成任务退出池{}",this);
workers.remove(this);
}
}
}
}
class BlockedQueue<T>{
//阻塞队列
private Deque<T> queue=new ArrayDeque<>();
//锁
private ReentrantLock lock=new ReentrantLock();
//空队列条件
private Condition emptyCondition=lock.newCondition();
//满队列条件
private Condition fullCondition=lock.newCondition();
private int capcity;
public BlockedQueue(int capcity) {
this.capcity = capcity;
}
public T poll(long timeout, TimeUnit unit){
lock.lock();
try{
long nacos = unit.toNanos(timeout);
while(queue.isEmpty()){
try {
//等待超时,后自动解锁。
if(nacos<=0){
return null;
}
nacos= emptyCondition.awaitNanos(nacos);
} catch (InterruptedException e) {
e.printStackTrace();
}
}
T t = queue.removeFirst();
//唤醒满队列
fullCondition.signal();
return t;
}finally {
lock.unlock();
}
}
public T take(){
lock.lock();
try{
while(queue.isEmpty()){
try {
emptyCondition.await();
} catch (InterruptedException e) {
e.printStackTrace();
}
}
T t = queue.removeFirst();
//唤醒满队列
fullCondition.signal();
return t;
}finally {
lock.unlock();
}
}
public void put(T t){
lock.lock();
try{
while(queue.size()==capcity){
try {
fullCondition.await();
} catch (InterruptedException e) {
e.printStackTrace();
}
}
queue.addLast(t);
emptyCondition.signal();
}finally {
lock.unlock();
}
}
public int getCapcity(){
lock.lock();
try{
return this.capcity;
}finally {
lock.unlock();
}
}
}
解决方案
可以通过对等待加入任务队列的条件变量限制超时时间,如果一定时间没有空位,那么主线程主动放弃把新任务加入到任务队列的操作
public boolean offer(T t,long timeout,TimeUnit unit){
lock.lock();
try{
long nacos=unit.toNanos(timeout);
while(queue.size()==capcity){
try {
log.debug("任务队列满了...");
if(nacos<=0){
return false;
}
nacos = fullCondition.awaitNanos(nacos);
} catch (InterruptedException e) {
e.printStackTrace();
}
}
//任务加入到队列
log.debug("加入到任务队列{}",t);
queue.addLast(t);
emptyCondition.signal();
return true;
}finally {
lock.unlock();
}
}
阻塞添加的处理办法有很多
这里使用到了策略模式,实际上就是一个接口,然后如何实现并不知道,可以通过自己的实现来完成不同的拒绝策略。
本质就是通过接口,和lambda函数来完成拒绝策略的多样性编写。并且在需要执行拒绝策略的类中加入成员变量,通过局部变量来传输执行。实际上就是一个执行接口,只要出现该拒绝的时候就可以调用策略执行。
@FunctionalInterface
interface RejectPolicy<T>{
void reject(BlockedQueue<T> queue,T task);
}
public void execute(Runnable task){
synchronized (this){
//保证只有一个线程能够被创建
if(workers.size()<coreSize){
Worker worker = new Worker(task);
log.debug("线程和任务创建{},{}",worker,task);
workers.add(worker);
worker.start();
}else{
//如果没有线程处理那么就放入到阻塞队列
queue.tryput(rejectPolicy,task);
}
}
}
public void tryput(RejectPolicy<T> rejectPolicy, T task) {
//判断是否满,本队列在调用自己的加入方法,然后其他考自定义
lock.lock();
try{
if(queue.size()==capcity){
rejectPolicy.reject(this,task);
}else{
log.debug("加入到任务队列{}",task);
queue.addLast(task);
emptyCondition.signal();
}
}finally {
lock.unlock();
}
}
各种策略
ThreadPool threadPool = new ThreadPool(2, 1000, TimeUnit.MILLISECONDS, 0
,(queue,task)->{
//死等
// queue.put(task);
//超时等待
// queue.offer(task,500,TimeUnit.MILLISECONDS);
//放弃执行
// log.debug("放弃任务{}",task);
//抛出异常
// throw new RuntimeException("任务执行失败");
//调用者自己执行,不能使用线程池了
task.run();
});
工作顺序首先是创建核心线程工作,核心线程满了那么就把任务放进去阻塞队列,阻塞队列也满了那么就放到救急线程处理,最后才是拒绝策略的使用。
核心线程和救急线程最大的不同就是救急线程是有限制时间的。
创建固定大小的线程池,这里的核心线程数和最大线程数一致,而且是LinkedBlockQueue。使用的线程工厂是Executors.defaultThreadFactory(),默认的。实际上就为线程创建好名字。而且创建的线程是非守护线程不会随着main线程结束而结束
static class DefaultThreadFactory implements ThreadFactory {
private static final AtomicInteger poolNumber = new AtomicInteger(1);
private final ThreadGroup group;
private final AtomicInteger threadNumber = new AtomicInteger(1);
private final String namePrefix;
DefaultThreadFactory() {
SecurityManager s = System.getSecurityManager();
group = (s != null) ? s.getThreadGroup() :
Thread.currentThread().getThreadGroup();
namePrefix = "pool-" +
poolNumber.getAndIncrement() +
"-thread-";
}
public Thread newThread(Runnable r) {
Thread t = new Thread(group, r,
namePrefix + threadNumber.getAndIncrement(),
0);
if (t.isDaemon())
t.setDaemon(false);
if (t.getPriority() != Thread.NORM_PRIORITY)
t.setPriority(Thread.NORM_PRIORITY);
return t;
}
}
public static ExecutorService newFixedThreadPool(int nThreads) {
return new ThreadPoolExecutor(nThreads, nThreads,
0L, TimeUnit.MILLISECONDS,
new LinkedBlockingQueue<Runnable>());
}
@Slf4j(topic = "c.test")
public class TestThreadPoolExecutors {
public static void main(String[] args) {
ExecutorService pool = Executors.newFixedThreadPool(2, new ThreadFactory() {
private AtomicInteger i=new AtomicInteger(1);
@Override
public Thread newThread(Runnable r) {
return new Thread(r,"线程"+i.getAndIncrement());
}
});
pool.execute(()->{
log.debug("1");
});
}
}
缓存线程池特点
public static ExecutorService newCachedThreadPool() {
return new ThreadPoolExecutor(0, Integer.MAX_VALUE,
60L, TimeUnit.SECONDS,
new SynchronousQueue<Runnable>());
}
@Slf4j(topic = "c.TestSynchronousQueue")
public class TestSynchronousQueue {
public static void main(String[] args) {
SynchronousQueue<Integer> integers = new SynchronousQueue<>();
new Thread(() -> {
try {
log.debug("putting {} ", 1);
integers.put(1);
log.debug("{} putted...", 1);
log.debug("putting...{} ", 2);
integers.put(2);
log.debug("{} putted...", 2);
} catch (InterruptedException e) {
e.printStackTrace();
}
},"t1").start();
sleep(1);
new Thread(() -> {
try {
log.debug("taking {}", 1);
integers.take();
} catch (InterruptedException e) {
e.printStackTrace();
}
},"t2").start();
sleep(1);
new Thread(() -> {
try {
log.debug("taking {}", 2);
integers.take();
} catch (InterruptedException e) {
e.printStackTrace();
}
},"t3").start();
}
}
特点
public static ExecutorService newSingleThreadExecutor() {
return new FinalizableDelegatedExecutorService
(new ThreadPoolExecutor(1, 1,
0L, TimeUnit.MILLISECONDS,
new LinkedBlockingQueue<Runnable>()));
}
有返回值的执行。通过Future进行接收类似,与通信的一个中间信箱。
@Slf4j(topic = "c.test")
public class TestSubmit {
public static void main(String[] args) throws ExecutionException, InterruptedException {
ExecutorService service = Executors.newFixedThreadPool(1);
Future<String> future = service.submit(new Callable<String>() {
@Override
public String call() throws Exception {
log.debug("好人");
return "ok";
}
});
log.debug("结果{}",future.get());
}
}
执行一系列的任务。返回一系列的结果
@Slf4j(topic = "c.test")
public class TestSubmit {
public static void main(String[] args) throws ExecutionException, InterruptedException {
ExecutorService service = Executors.newFixedThreadPool(1);
List<Future<Object>> futures = service.invokeAll(Arrays.asList(
() -> {
Sleeper.sleep(1);
log.debug("测试1");
return "1";
},
() -> {
Sleeper.sleep(0.5);
log.debug("测试2");
return "2";
},
() -> {
Sleeper.sleep(2);
log.debug("测试3");
return "3";
}
)
);
futures.forEach(f->{
try {
log.debug("结果{}",f.get());
} catch (InterruptedException e) {
e.printStackTrace();
} catch (ExecutionException e) {
e.printStackTrace();
}
});
}
}
最快执行完的任务返回结果并且截断其它任务的执行。
@Slf4j(topic = "c.test")
public class TestSubmit {
public static void main(String[] args) throws ExecutionException, InterruptedException {
ExecutorService service = Executors.newFixedThreadPool(1);
String s = service.invokeAny(Arrays.asList(
() -> {
Sleeper.sleep(1);
log.debug("{}", 1);
return "1";
},
() -> {
Sleeper.sleep(0.5);
log.debug("{}", 3);
return "1";
},
() -> {
Sleeper.sleep(2);
log.debug("{}", 2);
return "1";
}
));
log.debug("{}",s);
}
}
public void shutdown() {
final ReentrantLock mainLock = this.mainLock;
mainLock.lock();
try {
checkShutdownAccess();
//修改线程池状态
advanceRunState(SHUTDOWN);
//关闭所有空闲线程
interruptIdleWorkers();
onShutdown(); // hook for ScheduledThreadPoolExecutor
} finally {
mainLock.unlock();
}
tryTerminate();
}
public List<Runnable> shutdownNow() {
List<Runnable> tasks;
final ReentrantLock mainLock = this.mainLock;
mainLock.lock();
try {
checkShutdownAccess();
//修改状态为stop
advanceRunState(STOP);
//关闭所有线程
interruptWorkers();
tasks = drainQueue();
} finally {
mainLock.unlock();
}
tryTerminate();
return tasks;
}
shutdown之后主线程是不会等待线程池全部执行完的。但是可以通过调用方法awaitTerminated来要求主线程到底需要等待多久。但是shutdownNow需要主线程等待线程池中断之后的返回任务才能够继续执行。
@Slf4j(topic = "c.TestShutDown")
public class TestShutDown {
public static void main(String[] args) throws ExecutionException, InterruptedException {
ExecutorService pool = Executors.newFixedThreadPool(2);
Future<Integer> result1 = pool.submit(() -> {
log.debug("task 1 running...");
Thread.sleep(1000);
log.debug("task 1 finish...");
return 1;
});
Future<Integer> result2 = pool.submit(() -> {
log.debug("task 2 running...");
Thread.sleep(1000);
log.debug("task 2 finish...");
return 2;
});
Future<Integer> result3 = pool.submit(() -> {
log.debug("task 3 running...");
Thread.sleep(1000);
log.debug("task 3 finish...");
return 3;
});
log.debug("shutdown");
pool.shutdown();
pool.awaitTermination(3, TimeUnit.SECONDS);
log.debug("other");
// List runnables = pool.shutdownNow();
// log.debug("other.... {}" , runnables);
}
}
其实就是给线程进行分工。有限的线程处理多个异步任务,而且分工处理,增强效率。
出现的问题就是只有两个线程,而且同时点菜成功之后,需要等待其它线程做完菜才能结束运行。但是这里很明显已经没有线程去做菜了,导致了死锁的问题(不是synchronize的死锁),其实就是因为没有线程执行下一步导致大家都卡住在那里
ExecutorService pool = Executors.newFixedThreadPool(2);
pool.execute(()->{
log.debug("点菜");
Future<String> future = pool.submit(() -> {
log.debug("做菜");
return cooking();
});
try {
log.debug("上菜{}",future.get());
} catch (InterruptedException e) {
e.printStackTrace();
} catch (ExecutionException e) {
e.printStackTrace();
}
});
解决方案
可以把线程池的功能区分,一个线程做服务员,一个线程做厨师,那么就不会导致两个都去点菜导致没人做菜的问题.
static final List<String> MENU = Arrays.asList("地三鲜", "宫保鸡丁", "辣子鸡丁", "烤鸡翅");
static Random RANDOM = new Random();
static String cooking() {
return MENU.get(RANDOM.nextInt(MENU.size()));
}
public static void main(String[] args) {
ExecutorService waiterPool = Executors.newFixedThreadPool(1);
ExecutorService cookPool = Executors.newFixedThreadPool(1);
waiterPool.execute(()->{
log.debug("点菜");
Future<String> future = cookPool.submit(() -> {
log.debug("做菜");
return cooking();
});
try {
log.debug("上菜{}",future.get());
} catch (InterruptedException e) {
e.printStackTrace();
} catch (ExecutionException e) {
e.printStackTrace();
}
});
waiterPool.execute(()->{
log.debug("点菜");
Future<String> future = cookPool.submit(() -> {
log.debug("做菜");
return cooking();
});
try {
log.debug("上菜{}",future.get());
} catch (InterruptedException e) {
e.printStackTrace();
} catch (ExecutionException e) {
e.printStackTrace();
}
});
}
也就是cpu计算时间占据大部分,那么就不需要那么多线程,只需要cpu数量+1的线程,多一个去处理页缺失问题也就是虚拟内存导致。cpu计算时间多,如果线程数还多的话问题就是上下文切换导致的时间浪费,性能消耗。所以只需要刚好这么多就可以了
cpu百分之50去计算,百分之50去等待,那么就需要多个线程去处理等待的问题,如果需要io等待,只需要分配一个线程去处理阻塞不占用cpu的计算时间,提高cpu的计算利用率。
问题就是只有一个线程在处理,限时任务如果出现异常或者是延迟就会影响后面的线程处理
@Slf4j(topic = "c.32")
public class MyTest32 {
public static void main(String[] args) {
Timer timer=new Timer();
TimerTask task1=new TimerTask() {
@Override
public void run() {
log.debug("执行1");
Sleeper.sleep(2);
}
};
TimerTask task2=new TimerTask() {
@Override
public void run() {
log.debug("执行2");
}
};
timer.schedule(task1,1000);
timer.schedule(task2,1000);
}
}
ScheduledThreadPool解决了Timer如果任务出现异常和延迟的问题。可以略过当前任务异常执行下一个任务,并且可以通过多个线程解决延迟问题。
@Slf4j(topic = "c.33")
public class MyTest33 {
public static void main(String[] args) {
ScheduledExecutorService scheduledThreadPool = Executors.newScheduledThreadPool(1);
scheduledThreadPool.schedule(()->{
log.debug("1");
int i=1/0;
Sleeper.sleep(1);
},1, TimeUnit.SECONDS);
scheduledThreadPool.schedule(()->{
log.debug("2");
},1,TimeUnit.SECONDS);
}
}
@Slf4j(topic = "c.33")
public class MyTest33 {
public static void main(String[] args) {
ScheduledExecutorService scheduledThreadPool = Executors.newScheduledThreadPool(1);
// scheduledThreadPool.scheduleAtFixedRate(()->{
// log.debug("运行");
// Sleeper.sleep(2);
// },1,1,TimeUnit.SECONDS);
scheduledThreadPool.scheduleWithFixedDelay(()->{
log.debug("运行");
Sleeper.sleep(2);
},1,1,TimeUnit.SECONDS);
}
}
可以直接通过future来返回异常,因为线程池处理的任务异常全部封装到了future
Future<String> future = scheduledThreadPool.submit(() -> {
log.debug("测试");
// int i = 1 / 0;
return "1";
});
log.debug("{}",future.get());
主要就是对间隔时间的一个计算,使用LocalDateTime进行处理。with设置时间,compareto来对比两个时间的大小,最后使用plus等方法来增加时间。Duration来计算时间差,然后就可以带入到上面的定时任务。
LocalDateTime now=LocalDateTime.now();
LocalDateTime time = now.withHour(18).withSecond(0).withMinute(0).withNano(0).with(DayOfWeek.THURSDAY);
log.debug("{}",now);
log.debug("{}",time);
if(now.compareTo(time)>0){
time=time.plusWeeks(1);
}
log.debug("{}",time);
long l = Duration.between(now, time).toMillis();
System.out.println(l);
每个线程的作用
分工处理,并发度更强。
最大和Executor创建的线程池的区别是execute是重写的,这里再救急线程用完之后不会立即调用拒绝策略,而是再次尝试把任务放入队,如果失败那么才调用拒绝策略
public void execute(Runnable command) {
if (this.executor != null) {
try {
//执行任务
this.executor.execute(command);
} catch (RejectedExecutionException var3) {
//发现线程和队列都没有空闲,先尝试把任务再次加入到队列中
if (!((TaskQueue)this.executor.getQueue()).force(command)) {
throw new RejectedExecutionException(sm.getString("standardThreadExecutor.queueFull"));
}
}
} else {
throw new IllegalStateException(sm.getString("standardThreadExecutor.notStarted"));
}
}
调用超时拒绝策略。
public boolean force(Runnable o, long timeout, TimeUnit unit) throws InterruptedException {
if (this.parent != null && !this.parent.isShutdown()) {
//调用超时任务加入
return super.offer(o, timeout, unit);
} else {
throw new RejectedExecutionException(sm.getString("taskQueue.notRunning"));
}
}
其实就是分治,把大任务拆分成小任务交给线程执行。
思路
其实就是通过把1+2+3+4+5进行了任务拆分,变成5+(4),4+(3)。。。这样的任务,接着交给不同的线程来进行处理,而且每个线程处理的结果依赖于上一次的小任务处理的结果。
@Slf4j(topic = "c.TestForkJoin2")
public class TestForkJoin2 {
public static void main(String[] args) {
ForkJoinPool pool = new ForkJoinPool(4);
System.out.println(pool.invoke(new MyTask(5)));
// new MyTask(5) 5+ new MyTask(4) 4 + new MyTask(3) 3 + new MyTask(2) 2 + new MyTask(1)
}
}
// 1~n 之间整数的和
@Slf4j(topic = "c.MyTask")
class MyTask extends RecursiveTask<Integer> {
private int n;
public MyTask(int n) {
this.n = n;
}
@Override
public String toString() {
return "{" + n + '}';
}
@Override
protected Integer compute() {
// 如果 n 已经为 1,可以求得结果了
if (n == 1) {
log.debug("join() {}", n);
return n;
}
// 将任务进行拆分(fork)
AddTask1 t1 = new AddTask1(n - 1);
t1.fork();
log.debug("fork() {} + {}", n, t1);
// 合并(join)结果
int result = n + t1.join();
log.debug("join() {} + {} = {}", n, t1, result);
return result;
}
}
优化思路就是二分拆解,把任务拆成两个范围,相当于就是减小了树的深度,返回的速度也就更快了。
相当于这次是拆分成两个节点去处理,而不是一个一个处理,两个处理可以分配多个线程,那么就不需要等待那么多次前面线程处理完之后返回的结果
特点
自定义的锁的方法基本上都是通过AQS来进行实现的
最后就是实现锁的方法,基本上都是间接调用同步器的方法来执行
@Slf4j(topic = "c.test111")
public class MyLockTest {
public static void main(String[] args) {
MyLock lock = new MyLock();
new Thread(()->{
System.out.println("线程1执行");
lock.lock();
try{
log.debug("上锁1");
Sleeper.sleep(2);
}finally {
log.debug("解锁1");
lock.unlock();
}
},"t1").start();
new Thread(()->{
lock.lock();
try{
log.debug("上锁2");
Sleeper.sleep(1);
}finally {
log.debug("解锁2");
lock.unlock();
}
},"t2").start();
}
}
class MyLock implements Lock {
class MySync extends AbstractQueuedSynchronizer{
@Override
protected boolean tryAcquire(int arg) {
//尝试加锁
if(compareAndSetState(0,1)){
setExclusiveOwnerThread(Thread.currentThread());
return true;
}
return false;
}
@Override
protected boolean tryRelease(int arg) {
//尝试释放锁
setExclusiveOwnerThread(null);
setState(0);
return true;
}
@Override
protected boolean isHeldExclusively() {
//锁是不是被线程持有
return getState()==1;
}
public Condition newCondition(){
return new ConditionObject();
}
}
private MySync sync=new MySync();
@Override//加锁
public void lock() {
sync.acquire(1);
}
@Override//加锁,可以被中断
public void lockInterruptibly() throws InterruptedException {
sync.acquireInterruptibly(1);
}
@Override//尝试加锁,不成功就放弃
public boolean tryLock() {
return sync.tryAcquire(1);
}
//尝试加锁,超时就进入队列
@Override
public boolean tryLock(long time, TimeUnit unit) throws InterruptedException {
long waitTime = unit.toNanos(time);
return sync.tryAcquireNanos(1,waitTime);
}
@Override
public void unlock() {
sync.release(1);
}
@Override
public Condition newCondition() {
// sync.newCondition();
return sync.newCondition();
}
}
public final void acquire(int arg) {
if (!tryAcquire(arg) &&//尝试加锁
acquireQueued(addWaiter(Node.EXCLUSIVE), arg))//加锁失败线程进入阻塞队列
selfInterrupt();
}
public final boolean release(int arg) {
if (tryRelease(arg)) {
Node h = head;
if (h != null && h.waitStatus != 0)
unparkSuccessor(h);
return true;
}
return false;
}
ctrl+alt+左键可以查看接口方法
首先就是ReentrantLock中的Sync同步器继承了AQS而且有两个子类。
public void lock() {//ReentrantLock的方法
sync.lock();
}
final void lock() {//NonfairSync
if (compareAndSetState(0, 1))
setExclusiveOwnerThread(Thread.currentThread());
else
acquire(1);
}
总共尝试了4次获取锁。(作用?)
final void lock() {
if (compareAndSetState(0, 1))
setExclusiveOwnerThread(Thread.currentThread());
else
acquire(1);
}
public final void acquire(int arg) {
if (!tryAcquire(arg) &&
acquireQueued(addWaiter(Node.EXCLUSIVE), arg))
selfInterrupt();
}
final boolean acquireQueued(final Node node, int arg) {
boolean failed = true;
try {
boolean interrupted = false;
for (;;) {
final Node p = node.predecessor();
if (p == head && tryAcquire(arg)) {
setHead(node);
p.next = null; // help GC
failed = false;
return interrupted;
}
if (shouldParkAfterFailedAcquire(p, node) &&
parkAndCheckInterrupt())
interrupted = true;
}
} finally {
if (failed)
cancelAcquire(node);
}
}
流程总结ReentrantLock.unlock->AQS.release->Sync.tryRelease->state修改,owner设置为null->唤醒下一个队列节点->parkAndCheckInterrupt中重新往下面执行
如果这个时候有线程在外面进来,由于是非公平锁,所以还是会与唤醒的线程竞争。竞争失败再次进入队列
final boolean acquireQueued(final Node node, int arg) {
boolean failed = true;
try {
boolean interrupted = false;
for (;;) {
final Node p = node.predecessor();
if (p == head && tryAcquire(arg)) {
setHead(node);
p.next = null; // help GC
failed = false;
return interrupted;
}
if (shouldParkAfterFailedAcquire(p, node) &&
parkAndCheckInterrupt())//解锁后的位置
interrupted = true;
}
} finally {
if (failed)
cancelAcquire(node);
}
}
private void unparkSuccessor(Node node) {
/*
* If status is negative (i.e., possibly needing signal) try
* to clear in anticipation of signalling. It is OK if this
* fails or if status is changed by waiting thread.
*/
int ws = node.waitStatus;
if (ws < 0)
compareAndSetWaitStatus(node, ws, 0);
/*
* Thread to unpark is held in successor, which is normally
* just the next node. But if cancelled or apparently null,
* traverse backwards from tail to find the actual
* non-cancelled successor.
*/
Node s = node.next;//获取下一个节点
if (s == null || s.waitStatus > 0) {
s = null;
for (Node t = tail; t != null && t != node; t = t.prev)
if (t.waitStatus <= 0)
s = t;
}
if (s != null)
LockSupport.unpark(s.thread);//解锁
}
public final boolean release(int arg) {
if (tryRelease(arg)) {
Node h = head;
if (h != null && h.waitStatus != 0)
unparkSuccessor(h);
return true;
}
return false;
}
protected final boolean tryRelease(int releases) {
int c = getState() - releases;
if (Thread.currentThread() != getExclusiveOwnerThread())
throw new IllegalMonitorStateException();
boolean free = false;
if (c == 0) {
free = true;
setExclusiveOwnerThread(null);//设置owner为空
}
setState(c);//修改状态
return free;
}
本质就是state代表状态也代表锁重入次数,通过state来决定是不是可重入锁,并且对比owner来是不是当前线程来决定是否可以可重入还是说直接加锁。
final boolean nonfairTryAcquire(int acquires) {
final Thread current = Thread.currentThread();
int c = getState();
if (c == 0) {//如果锁没有锁定资源,那么就把锁交给当前线程
if (compareAndSetState(0, acquires)) {
setExclusiveOwnerThread(current);
return true;
}
}
else if (current == getExclusiveOwnerThread()) {//如果当前的资源是当前线程那么就可以修改状态,也就是锁被同1线程获取多少次
int nextc = c + acquires;
if (nextc < 0) // overflow
throw new Error("Maximum lock count exceeded");
setState(nextc);
return true;
}
return false;
}
protected final boolean tryRelease(int releases) {
//state-1相当于就是锁释放一次,如果发现不是0那么就是可重入锁。
int c = getState() - releases;
if (Thread.currentThread() != getExclusiveOwnerThread())
throw new IllegalMonitorStateException();
boolean free = false;
if (c == 0) {
free = true;
setExclusiveOwnerThread(null);
}
setState(c);
return free;
}
private final boolean parkAndCheckInterrupt() {
LockSupport.park(this);
return Thread.interrupted();
}
final boolean acquireQueued(final Node node, int arg) {
boolean failed = true;
try {
boolean interrupted = false;
for (;;) {
final Node p = node.predecessor();
if (p == head && tryAcquire(arg)) {
setHead(node);
p.next = null; // help GC
failed = false;
return interrupted;
}
if (shouldParkAfterFailedAcquire(p, node) &&
parkAndCheckInterrupt())//中断并且设置interrupted为true
interrupted = true;
}
} finally {
if (failed)
cancelAcquire(node);
}
}
public final void acquire(int arg) {
if (!tryAcquire(arg) &&
acquireQueued(addWaiter(Node.EXCLUSIVE), arg))
selfInterrupt();//调用中断
}
static void selfInterrupt() {//再次调用线程中断
Thread.currentThread().interrupt();
}
private void doAcquireInterruptibly(int arg)
throws InterruptedException {
final Node node = addWaiter(Node.EXCLUSIVE);
boolean failed = true;
try {
for (;;) {
final Node p = node.predecessor();
if (p == head && tryAcquire(arg)) {
setHead(node);
p.next = null; // help GC
failed = false;
return;
}
if (shouldParkAfterFailedAcquire(p, node) &&
parkAndCheckInterrupt())
//被打断后抛出异常
throw new InterruptedException();
}
} finally {
if (failed)
cancelAcquire(node);
}
}
public final void acquireInterruptibly(int arg)
throws InterruptedException {
if (Thread.interrupted())
throw new InterruptedException();
if (!tryAcquire(arg))
doAcquireInterruptibly(arg);
}
难点:可打断和不可打断的区别实质上就是被打断之后的处理方式。不可打断是通过变量interrupted来标记被打断,然后获取锁之后才能够打断线程。可打断直接抛出异常
公平这个地方会先判断是否有前驱节点,如果有那么就没办法直接去抢占,如果没有那么就可以直接抢。主要就是判断前驱节点是否有下一个节点,和下一个节点是不是就是当前的线程,如果不是那么就没办法锁定,只能够进入acquireQueue方法判断是不是需要阻塞。
protected final boolean tryAcquire(int acquires) {
final Thread current = Thread.currentThread();
int c = getState();
if (c == 0) {
if (!hasQueuedPredecessors() &&
compareAndSetState(0, acquires)) {
setExclusiveOwnerThread(current);
return true;
}
}
else if (current == getExclusiveOwnerThread()) {
int nextc = c + acquires;
if (nextc < 0)
throw new Error("Maximum lock count exceeded");
setState(nextc);
return true;
}
return false;
}
}
public final boolean hasQueuedPredecessors() {
// The correctness of this depends on head being initialized
// before tail and on head.next being accurate if the current
// thread is first in queue.
Node t = tail; // Read fields in reverse initialization order
Node h = head;
Node s;
return h != t &&
((s = h.next) == null || s.thread != Thread.currentThread());//判断前驱节点的下一个是不是为空
}
直接就是判断owner是不是空,如果不是那么就直接切入,而不是先去判断AQS是不是有节点在阻塞,导致的非公平问题。
final boolean nonfairTryAcquire(int acquires) {
final Thread current = Thread.currentThread();
int c = getState();
if (c == 0) {//直接判断而不会去查询AQS是不是有节点在阻塞
if (compareAndSetState(0, acquires)) {
setExclusiveOwnerThread(current);
return true;
}
}
else if (current == getExclusiveOwnerThread()) {
int nextc = c + acquires;
if (nextc < 0) // overflow
throw new Error("Maximum lock count exceeded");
setState(nextc);
return true;
}
return false;
}
总结:进入condition队列,清空锁并且唤醒线程,最后就是使用park进行阻塞。
public final void await() throws InterruptedException {
if (Thread.interrupted())
throw new InterruptedException();
Node node = addConditionWaiter();//进入condition
int savedState = fullyRelease(node);//清空锁
int interruptMode = 0;
while (!isOnSyncQueue(node)) {
LockSupport.park(this);//进入阻塞
if ((interruptMode = checkInterruptWhileWaiting(node)) != 0)
break;
}
if (acquireQueued(node, savedState) && interruptMode != THROW_IE)
interruptMode = REINTERRUPT;
if (node.nextWaiter != null) // clean up if cancelled
unlinkCancelledWaiters();
if (interruptMode != 0)
reportInterruptAfterWait(interruptMode);
}
final int fullyRelease(Node node) {
boolean failed = true;
try {
int savedState = getState();
if (release(savedState)) {//清空锁实际上就是清空state
failed = false;
return savedState;
} else {
throw new IllegalMonitorStateException();
}
} finally {
if (failed)
node.waitStatus = Node.CANCELLED;
}
}
public final boolean release(int arg) {
if (tryRelease(arg)) {
Node h = head;
if (h != null && h.waitStatus != 0)
unparkSuccessor(h);//唤醒同步器队列的线程。
return true;
}
return false;
}
总结:signal完成了条件队列清除(单项链表清除),然后就是把对应的节点全部送去Sync队列。如果失败可能就是队列满了或者是超时了。最后就是取出前驱节点修改状态。
public final void signal() {
if (!isHeldExclusively())
throw new IllegalMonitorStateException();
Node first = firstWaiter;
if (first != null)
doSignal(first);
}
private void doSignal(Node first) {
do {
if ( (firstWaiter = first.nextWaiter) == null)
lastWaiter = null;
first.nextWaiter = null;
} while (!transferForSignal(first) &&
(first = firstWaiter) != null);
}
private Node addConditionWaiter() {
Node t = lastWaiter;
// If lastWaiter is cancelled, clean out.
if (t != null && t.waitStatus != Node.CONDITION) {
unlinkCancelledWaiters();
t = lastWaiter;
}
Node node = new Node(Thread.currentThread(), Node.CONDITION);
if (t == null)
firstWaiter = node;
else
t.nextWaiter = node;
lastWaiter = node;
return node;
}
final boolean transferForSignal(Node node) {
/*
* If cannot change waitStatus, the node has been cancelled.
*/
if (!compareAndSetWaitStatus(node, Node.CONDITION, 0))
return false;
/*
* Splice onto queue and try to set waitStatus of predecessor to
* indicate that thread is (probably) waiting. If cancelled or
* attempt to set waitStatus fails, wake up to resync (in which
* case the waitStatus can be transiently and harmlessly wrong).
*/
Node p = enq(node);
int ws = p.waitStatus;
if (ws > 0 || !compareAndSetWaitStatus(p, ws, Node.SIGNAL))
LockSupport.unpark(node.thread);
return true;
}
读的过程中大部分时间都是可以共享的因为不会修改资源,但是写的时候需要上锁。为了更高的并发和读多写少的问题,可以使用读写锁
public class MyReadAndWrite {
public static void main(String[] args) {
DataContainer dataContainer = new DataContainer("123");
new Thread(()->{
dataContainer.read();
},"t1").start();
Sleeper.sleep(0.1);
new Thread(()->{
dataContainer.write();
},"t2").start();
}
}
@Slf4j(topic = "c.data")
class DataContainer{
public DataContainer(String s) {
this.s = s;
}
private String s;
private ReentrantReadWriteLock rw=new ReentrantReadWriteLock();
private ReentrantReadWriteLock.ReadLock r=rw.readLock();
private ReentrantReadWriteLock.WriteLock w=rw.writeLock();
public void read(){
log.debug("上读锁");
r.lock();
try{
log.debug("读取{}",s);
Sleeper.sleep(1);
}finally {
log.debug("释放读锁");
r.unlock();
}
}
public void write(){
log.debug("上写锁");
w.lock();
try{
log.debug("读取{}",s);
}finally {
log.debug("释放写锁");
w.unlock();
}
}
}
问题就是线程B清空之后还没更新数据库切换到线程A发现缓存空了查询数据库,然后又把旧数据弄上去。B这个时候才修改数据库,如果没有线程修改数据库,那么缓存就会一直不会发生改变,也就是读取错误数据的时间很长
B去更新数据库,但是还没有缓存的时候,切换到A读取缓存中错误数据,再切换到B来存入新数据到缓存。但是A再次读取就是正确的数据,而不像先清除缓存导致,A获取旧数据库的错误数据,并且如果没有线程修改那么缓存就不会发生变化。
问题
缓存错误期间怎么进行处理?
使用的是读写锁,当读取query的时候就是读锁,但是可能会多线程判断value不存在缓存之后被切换,那么这个时候除了再给map设置查询缓存的时候加上写锁(避免多个进来写),同时需要进行双重检查value是不是空,因为前一个线程可能已经完成缓存的替换。如果是更新那么直接加上写锁就可以了。
@Override
public <T> T queryOne(Class<T> beanClass, String sql, Object... args) {
// 先从缓存中找,找到直接返回
SqlPair key = new SqlPair(sql, args);;
// rw.readLock().lock();
rw.readLock().lock();
try {
T value = (T) map.get(key);
if(value != null) {
return value;
}
} finally {
rw.readLock().unlock();
// rw.readLock().unlock();
}
// rw.writeLock().lock();
rw.writeLock().lock();
try {
// 多个线程
T value = (T) map.get(key);
// if(value == null) {
// 缓存中没有,查询数据库
if(value==null){
value = dao.queryOne(beanClass, sql, args);
map.put(key, value);
}
// }
return value;
} finally {
rw.writeLock().unlock();
// rw.writeLock().unlock();
}
}
@Override
public int update(String sql, Object... args) {
// rw.writeLock().lock();
rw.writeLock().lock();
try {
// 先更新库
int update = dao.update(sql, args);
// 清空缓存
map.clear();
return update;
} finally {
rw.writeLock().unlock();
// rw.writeLock().unlock();
}
}
可继续优化的地方
public void lock() {
sync.acquire(1);
}
protected final boolean tryAcquire(int acquires) {
Thread current = Thread.currentThread();
int c = getState();
int w = exclusiveCount(c);
if (c != 0) {
// (Note: if c != 0 and w == 0 then shared count != 0)
if (w == 0 || current != getExclusiveOwnerThread())//如果别人获取读锁或者是owner不是自己,说明了别人获取了写锁或者是读锁,我无法获取锁。假设都不成立说明写锁被获取,并且owner是自己,那么就可以执行写锁可重入。
return false;/
if (w + exclusiveCount(acquires) > MAX_COUNT)//大于可重入的最大值
throw new Error("Maximum lock count exceeded");
// Reentrant acquire
setState(c + acquires);//重入
return true;
}
if (writerShouldBlock() ||
!compareAndSetState(c, c + acquires))//判断是不是公平锁,判断之后就上锁
return false;
setExclusiveOwnerThread(current);
return true;
}
public final boolean release(int arg) {
if (tryRelease(arg)) {
Node h = head;
if (h != null && h.waitStatus != 0)
unparkSuccessor(h);
return true;
}
return false;
}
protected final boolean tryRelease(int releases) {
if (!isHeldExclusively())
throw new IllegalMonitorStateException();
int nextc = getState() - releases;
boolean free = exclusiveCount(nextc) == 0;
if (free)
setExclusiveOwnerThread(null);
setState(nextc);
return free;
}
private void setHeadAndPropagate(Node node, int propagate) {
Node h = head; // Record old head for check below
setHead(node);
if (propagate > 0 || h == null || h.waitStatus < 0 ||
(h = head) == null || h.waitStatus < 0) {
Node s = node.next;
if (s == null || s.isShared())//如果下一个节点还是读锁
//再次唤醒
doReleaseShared();
}
}
private void doReleaseShared() {
for (;;) {
Node h = head;
if (h != null && h != tail) {
int ws = h.waitStatus;
if (ws == Node.SIGNAL) {
if (!compareAndSetWaitStatus(h, Node.SIGNAL, 0))
continue; // loop to recheck cases
unparkSuccessor(h);
}
else if (ws == 0 &&
!compareAndSetWaitStatus(h, 0, Node.PROPAGATE))
continue; // loop on failed CAS
}
if (h == head) // loop if head changed
break;
}
}
protected final int tryAcquireShared(int unused) {
Thread current = Thread.currentThread();
int c = getState();
if (exclusiveCount(c) != 0 &&
getExclusiveOwnerThread() != current)//如果别人获取了锁c自然就不是0,然后如果也不是自己的线程那么就没办法获取锁,返回-1
return -1;
int r = sharedCount(c);
if (!readerShouldBlock() &&
r < MAX_COUNT &&
compareAndSetState(c, c + SHARED_UNIT)) {
if (r == 0) {
firstReader = current;
firstReaderHoldCount = 1;
} else if (firstReader == current) {
firstReaderHoldCount++;
} else {
HoldCounter rh = cachedHoldCounter;
if (rh == null || rh.tid != getThreadId(current))
cachedHoldCounter = rh = readHolds.get();
else if (rh.count == 0)
readHolds.set(rh);
rh.count++;
}
return 1;
}
return fullTryAcquireShared(current);
}
private void doAcquireShared(int arg) {
final Node node = addWaiter(Node.SHARED);//加入节点到队列
boolean failed = true;
try {
boolean interrupted = false;
for (;;) {
final Node p = node.predecessor();
if (p == head) {//如果是老二
int r = tryAcquireShared(arg);//再次尝试获取锁
if (r >= 0) {
setHeadAndPropagate(node, r);
p.next = null; // help GC
if (interrupted)
selfInterrupt();
failed = false;
return;
}
}
if (shouldParkAfterFailedAcquire(p, node) &&//标记前驱节点状态-1
parkAndCheckInterrupt())
interrupted = true;
}
} finally {
if (failed)
cancelAcquire(node);
}
}
public void unlock() {
sync.releaseShared(1);
}
protected final boolean tryReleaseShared(int unused) {
Thread current = Thread.currentThread();
if (firstReader == current) {
// assert firstReaderHoldCount > 0;
if (firstReaderHoldCount == 1)
firstReader = null;
else
firstReaderHoldCount--;
} else {
HoldCounter rh = cachedHoldCounter;
if (rh == null || rh.tid != getThreadId(current))
rh = readHolds.get();
int count = rh.count;
if (count <= 1) {
readHolds.remove();
if (count <= 0)
throw unmatchedUnlockException();
}
--rh.count;
}
//解锁的关键代码
for (;;) {
int c = getState();
int nextc = c - SHARED_UNIT;
if (compareAndSetState(c, nextc))
// Releasing the read lock has no effect on readers,
// but it may allow waiting writers to proceed if
// both read and write locks are now free.
return nextc == 0;
}
}
总结:
lock区别:基本流程是一样的,但是读锁的tryAcuireShare返回的是int整型。都是对比状态,然后进行加锁或者是阻塞。但是在doAcquireQueue的时候就不一样,对于写锁来说,直接删除节点即可,但是读锁阻塞结束之后还会去doReleaseShare能够唤醒其它读锁,一系列完成操作
unlock区别:读锁的unlock会等所有线程都退出,那么才会唤醒下一个线程,对于写锁来说除非同一线程可重入,其他都是直接去唤醒队列中的线程。
关键的两个tryAcquire加锁和doAcquireQueue加入阻塞队列
tryRelease尝试解锁,doRelease唤醒下一个线程(这里的-1状态就起作用了)
模板(可能不是很精准)
lock
if(tryAcquire){//尝试获取
doAcquireQueue()//如果无法获取锁进入阻塞队列
}
doAcquireQueue(){
取 node
node如果是老二,再次tryAcquire
标记前驱节点
再次tryAcquire
阻塞
阻塞结束之后
- 再次tryAcquire
- 如果是写锁删头,如果是读锁再次唤醒读锁线程
- 结束
}
其实就是每次都加锁的时候都会生成戳,并且解锁对比戳而不是CAS来修改状态进而让读速度达到极致
定义其实就是生成戳(不加锁),然后就是判断戳是否被改变,不被改变就继续,改变就升级为读锁
案例上面,只有写锁和读锁一起被使用的时候会导致stamp的改变。两个读锁并不会
StampedLock的问题
@Slf4j(topic = "c.TestStampedLock")
public class TestStampedLock {
public static void main(String[] args) {
DataContainerStamped dataContainer = new DataContainerStamped(1);
new Thread(() -> {
dataContainer.read(1);
}, "t1").start();
sleep(0.5);
new Thread(() -> {
dataContainer.write(1);
}, "t2").start();
}
}
@Slf4j(topic = "c.DataContainerStamped")
class DataContainerStamped {
private int data;
private final StampedLock lock = new StampedLock();
public DataContainerStamped(int data) {
this.data = data;
}
public int read(int readTime) {
long stamp = lock.tryOptimisticRead();
log.debug("optimistic read locking...{}", stamp);
sleep(readTime);
if (lock.validate(stamp)) {
log.debug("read finish...{}, data:{}", stamp, data);
return data;
}
// 锁升级 - 读锁
log.debug("updating to read lock... {}", stamp);
try {
stamp = lock.readLock();
log.debug("read lock {}", stamp);
sleep(readTime);
log.debug("read finish...{}, data:{}", stamp, data);
return data;
} finally {
log.debug("read unlock {}", stamp);
lock.unlockRead(stamp);
}
}
public void write(int newData) {
long stamp = lock.writeLock();
log.debug("write lock {}", stamp);
try {
sleep(2);
this.data = newData;
} finally {
log.debug("write unlock {}", stamp);
lock.unlockWrite(stamp);
}
}
}
基本流程是一样的,但是读锁的tryAcuireShare返回的是int整型。都是对比状态,然后进行加锁或者是阻塞。但是在doAcquireQueue的时候就不一样,对于写锁来说,直接删除节点即可,但是读锁阻塞结束之后还会去doReleaseShare能够唤醒其它读锁,一系列完成操作
unlock区别:读锁的unlock会等所有线程都退出,那么才会唤醒下一个线程,对于写锁来说除非同一线程可重入,其他都是直接去唤醒队列中的线程。
关键的两个tryAcquire加锁和doAcquireQueue加入阻塞队列
tryRelease尝试解锁,doRelease唤醒下一个线程(这里的-1状态就起作用了)
模板(可能不是很精准)
lock
if(tryAcquire){//尝试获取
doAcquireQueue()//如果无法获取锁进入阻塞队列
}
doAcquireQueue(){
取 node
node如果是老二,再次tryAcquire
标记前驱节点
再次tryAcquire
阻塞
阻塞结束之后
- 再次tryAcquire
- 如果是写锁删头,如果是读锁再次唤醒读锁线程
- 结束
}
其实就是每次都加锁的时候都会生成戳,并且解锁对比戳而不是CAS来修改状态进而让读速度达到极致
定义其实就是生成戳(不加锁),然后就是判断戳是否被改变,不被改变就继续,改变就升级为读锁
案例上面,只有写锁和读锁一起被使用的时候会导致stamp的改变。两个读锁并不会
StampedLock的问题
@Slf4j(topic = "c.TestStampedLock")
public class TestStampedLock {
public static void main(String[] args) {
DataContainerStamped dataContainer = new DataContainerStamped(1);
new Thread(() -> {
dataContainer.read(1);
}, "t1").start();
sleep(0.5);
new Thread(() -> {
dataContainer.write(1);
}, "t2").start();
}
}
@Slf4j(topic = "c.DataContainerStamped")
class DataContainerStamped {
private int data;
private final StampedLock lock = new StampedLock();
public DataContainerStamped(int data) {
this.data = data;
}
public int read(int readTime) {
long stamp = lock.tryOptimisticRead();
log.debug("optimistic read locking...{}", stamp);
sleep(readTime);
if (lock.validate(stamp)) {
log.debug("read finish...{}, data:{}", stamp, data);
return data;
}
// 锁升级 - 读锁
log.debug("updating to read lock... {}", stamp);
try {
stamp = lock.readLock();
log.debug("read lock {}", stamp);
sleep(readTime);
log.debug("read finish...{}, data:{}", stamp, data);
return data;
} finally {
log.debug("read unlock {}", stamp);
lock.unlockRead(stamp);
}
}
public void write(int newData) {
long stamp = lock.writeLock();
log.debug("write lock {}", stamp);
try {
sleep(2);
this.data = newData;
} finally {
log.debug("write unlock {}", stamp);
lock.unlockWrite(stamp);
}
}
}
[外链图片转存中…(img-6OKhK5Ce-1634435379876)]