我对Java中多线程如何工作的了解很少。我知道当你运行两个线程时，Java会为每个线程分配“时间块”以供执行
例如：
public void test()
{
Thread testThread = new Thread(new TestThread());
testThread.start();
for (int i = 0; i < Integer.MAX_VALUE; ++i)
System.out.print("a");
}
private class TestThread extends Thread
{
public void run()
{
for (int i = 0; i < Integer.MAX_VALUE; ++i)
System.out.print("b");
}
}
会打印出类似的内容：
aaaaaaaaaaabbbbbbbbbbbbaaaaaaaaaaabbbbbbbbbbbbbaaaaaa ...
而不是：
abababababababababababababababab .....
我的问题是：是否可以减少“时间块”，以便我们更接近：
aabbaaabbaabbaaabbaabbbaabbaabbbaabbaaa ...
为什么？我正在尝试编写一个苹果推送通知服务器（只是为了好玩）。当您向Apple推送通知服务写入请求时，可能会发生一两件事：
1.如果请求有效，则不会返回任何内容
2.如果请求无效，它将返回错误代码并关闭连接，并且在无效请求之后和连接关闭之前发送的任何请求都将被丢弃。
因为读取套接字会阻塞，直到数据可供读取（如果我不写任何无效请求，这可能永远不会发生），我不能简单地在每次写入后读取，看看是否有错误而没有设置200-500毫秒的超时。如果我们有一百万个写入请求（非常可能），这个超时会增加55 - 138 HOURS，并且由于短暂的超时会导致请求永远不会被发送，我们可能会错过返回的错误。
所以我有两个线程，比如上面的例子，一个写入服务器，一个正在读取，等待查看是否返回错误。问题在于：如果请求＃4不好但我们在读取错误并且连接关闭之前编写＃5-＃10，则苹果服务会丢弃请求＃5-10。因此，一旦我们知道＃4是坏的并且我们知道我们写的最后一个请求是＃10，我们需要重新排队＃5-10再次发送。
我现在遇到的问题是由于大量的“时间卡盘”，我能够在读取线程读取之前写入＃1-＃400请求＃5出现错误，因此＃6-＃400被重新排队并发送再次。然后读取线程读取＃21的错误，因此＃22-＃400被重新排队并再次发送...等。理想情况下，读取线程将能够从写入的每5-10个请求中读取套接字。
来源：
private Object readWriteLock = new Object();
private volatile int lastWrittenIndex;
private volatile boolean doneWriting;
private List<PushNotificationRequest> pushNotificationRequestsResnedList = new ArrayList<PushNotificationRequest>();
public boolean write()
{
// get the requests read list
List<PushNotificationRequest> requests = getPushNotificationRequests(false);
// as long as there are more notifications to write...
while (requests.size() > 0)
{
lastWrittenIndex = -1;
doneWriting = false;
// create and start the read thread
Thread readThread = new Thread(new ReadThread(), "APNS Reader");
readThread.start();
for (int i = 0; i < requests.size(); ++i)
{
PushNotificationRequest request = requests.get(i);
// write
boolean success = false;
// attempt to send the notification a number of times
for (int j = 0; j < MAX_NUM_PN_WRITE_ATTEMPTS; ++j)
{
synchronized (readWriteLock)
{
try
{
// get the socket connection
SSLSocket socket = getAppleServiceSSLSockett();
OutputStream socketOutputStream = socket.getOutputStream();
socketOutputStream.write(request.binary);
socketOutputStream.flush();
success = true;
lastWrittenIndex = i;
break;
}
catch (IOException e)
{
e.printStackTrace();
}
catch (AppleServiceConnectionException e)
{
e.printStackTrace();
}
}
}
if (!success)
System.err.println("APNS Unable to send push notification:\n" + request);
}
// wait for some time so we can make sure the read thread can read everything
try
{
Thread.sleep(Config.APNS_READ_TIME_AFTER_DONE_WRITING_MILLISECONDS);
}
catch (InterruptedException e)
{
e.printStackTrace();
}
// let the read thread know we are done writing and close the connection so it unblocks
doneWriting = true;
closeAppleServiceSSLSockett();
// wait for the read thread to return
try
{
readThread.join();
}
catch (InterruptedException e)
{
e.printStackTrace();
}
// clear the reading list
requests.clear();
// add the requests from the re-send to the list
if (pushNotificationRequestsResnedList.size() > 0)
{
requests.addAll(pushNotificationRequestsResnedList);
// clear the re-send list
pushNotificationRequestsResnedList.clear();
}
}
}
private class ReadThread extends Thread
{
public void run()
{
byte[] readBuffer = new byte[1024];
int numBytesRead;
int totalNumBytesRead;
while (!doneWriting)
{
try
{
// get the socket connection
SSLSocket socket = getAppleServiceSSLSockett();
socket.setSoTimeout(Config.APNS_READ_TIMEOUT_MILLISECONDS);
InputStream socketInputStream = socket.getInputStream();
// read (blocking)
totalNumBytesRead = 0;
while ((numBytesRead = socketInputStream.read(readBuffer)) != -1)
totalNumBytesRead += numBytesRead;
// check for an error
if (totalNumBytesRead > 0)
{
synchronized (readWriteLock)
{
try
{
PushNotificationResponse response = new PushNotificationResponse(readBuffer, 0);
System.err.println("APNS Read got response with id: " + response.identifier);
// find the request with the given identifier
int i;
for (i = lastWrittenIndex; i > -1; --i)
{
if (pushNotificationRequestsReadingList.get(i).identifier == response.identifier)
break;
}
if (i == -1)
{
// something went wrong, we didn't find the identifier
System.err.println("APNS Read unable to find request with id: " + response.identifier);
}
else
{
System.err.println("APNS Read " + response.getErrorMessage(pushNotificationRequestsReadingList.get(i)));
// add the requests between the bad request and the last written (included)
for (++i; i <= lastWrittenIndex; ++i)
pushNotificationRequestsResnedList.add(pushNotificationRequestsReadingList.get(i));
}
}
catch (InvalidPushNotificationResponseException g)
{
g.printStackTrace();
}
}
// the socket will be closed, reopen it
try
{
reopenAppleServiceSSLSockett();
}
catch (AppleServiceConnectionException e)
{
e.printStackTrace();
}
}
}
catch (SocketException e)
{
// ignore a close, it is expected
if (!e.getMessage().equals("Socket closed"))
e.printStackTrace();
}
catch (IOException e)
{
e.printStackTrace();
}
catch (AppleServiceConnectionException e)
{
e.printStackTrace();
}
}
}
}
答案 0 :(得分：1)
您是否尝试过使用Thread.yield（）来请求线程调度程序移动到下一个线程？
// using Thread.yield() MIGHT give you the results you want
for (int i = 0; i < Integer.MAX_VALUE; ++i)
{
System.out.print("a");
Thread.yield();
}
请记住，线程调度取决于底层操作系统，因此以上只是一个有根据的猜测 - 我还没有尝试过运行它。有关示例，请参阅here。
编辑：here是关于不同平台如何实现收益的更多信息。
答案 1 :(得分：1)
这不是java，但实际上操作系统会调度线程的时间块。在生产者/消费者场景中，您希望在生产者方面提供公平性，以便一个生产者在所有其他生产者给出自己的产出后给出一个输出。为此，您可以使用一个线程来绕过几个懒惰的生成器。那就是：
N个生产者在生产者上公开getNextThing（）和1个消费者循环，并将结果传递给您消费的列表