第一个进入的线程如何向其他并发线程发出相同方法结束的信号？
我有一个名为 PollDPRAM() 的方法。它必须通过网络访问一些慢速硬件并刷新对象私有数据。如果相同的方法被其他线程同时调用，它们一定不能执行此行程，而是等待第一个到来的线程完成工作并简单地退出，因为数据是新鲜的（比如 10-30 毫秒前没有区别） . 在不先进入第二、第三等线程的方法中很容易检测到。我使用联锁计数器来检测并发性。
问题：我通过观察计数器 (Interlocked.Read) 来检测第一个线程的退出是一个糟糕的选择，以便在计数器减少到小于在 n>1 线程入口处检测到的值之后进行观察。选择很糟糕，因为第一个线程在离开后几乎可以立即重新进入该方法。所以 n>1 线程永远不会检测到计数器的下降。
所以问题：如何正确检测第一个进入的线程已经退出方法，即使这个第一个线程可以立即再次进入它？
谢谢
PS 一段代码
private void pollMotorsData()
{
// Execute single poll with "foreground" handshaking
DateTime start = DateTime.Now;
byte retryCount = 0;
// Pick old data atomically to detect change
uint motorsDataTimeStampPrev = this.MotorsDataTimeStamp;
bool changeDetected = false;
// The design goal of DPRAM is to ease the bottleneck
// Here is a sensor if bottleneck is actually that tight
long parallelThreads = Interlocked.Increment(ref this.motorsPollThreadCount);
try
{
// For first thread entering the counter will be 1
if (parallelThreads <= 1)
{
do
{
// Handshake signal to DPRAM write process on controller side that host PC is reading
this.controller.deltaTauTcpClient.Pmac_SetBit(OFFSET_0x006A_BIT15_FOREGROUND_READ, 15, true);
try
{
bool canReadMotors = false;
byte[] canReadFrozenDataFlag = new byte[2];
do
{
this.controller.deltaTauTcpClient.Pmac_GetMem(OFFSET_0x006E_BIT15_FOREGROUND_DONE, canReadFrozenDataFlag);
canReadMotors = (canReadFrozenDataFlag[1] & 0x80) == 0x80;
if (canReadMotors) break;
retryCount++;
Thread.Sleep(1);
} while (retryCount < 10);
if (!canReadMotors)
{
throw new DeltaTauControllerException(this.controller, "Timeout waiting on DPRAM Foreground Handshaking Bit");
}
// The lock is meaningless in contructor as it is certainly single threaded
// but for practice sake the access to data should always be serialized
lock (motorsDataLock)
{
// Obtain fresh content of DPRAM
this.controller.deltaTauTcpClient.Pmac_GetMem(OFFSET_0x006A_394BYTES_8MOTORS_DATA, this.motorsData);
this.motorsDataBorn = DateTime.Now;
}
}
finally
{
// Handshake signal to DPRAM write process on controller side that host PC has finished reading
this.controller.deltaTauTcpClient.Pmac_SetBit(OFFSET_0x006A_BIT15_FOREGROUND_READ, 15, false);
}
// Check live change in a separate atom
changeDetected = this.MotorsDataTimeStamp != motorsDataTimeStampPrev;
} while ((!changeDetected) && ((DateTime.Now - start).TotalMilliseconds < 255));
// Assert that result is live
if (!changeDetected)
{
throw new DeltaTauControllerException(this.controller, "DPRAM Background Data timestamp is not updated. DPRAM forground handshaking failed.");
}
}
else
{
// OK. Bottleneck ! The concurrent polls have collided
// Give the controller a breathe by waiting for other thread do the job
// Avoid aggressive polling of stale data, which is not able to be written, locked by reader
// Just wait for other thread do whole polling job and return with no action
// because the data is milliseconds fresh
do
{
// Amount of parallel threads must eventually decrease
// But no thread will leave and decrease the counter until job is done
if (Interlocked.Read(ref this.motorsPollThreadCount) < parallelThreads)
{
// Return is possible because decreased value of concurrentThreads means that
// this very time other thread has finished the poll 1 millisecond ago at most
return;
}
Thread.Sleep(1);
retryCount++;
} while ((DateTime.Now - start).TotalMilliseconds < 255);
throw new DeltaTauControllerException(this.controller, "Timeout 255ms waiting on concurrent thread to complete DPRAM polling");
}
}
finally
{
// Signal to other threads that work is done
Interlocked.Decrement(ref this.motorsPollThreadCount);
// Trace the timing and bottleneck situations
TimeSpan duration = DateTime.Now - start;
if (duration.TotalMilliseconds > 50 || parallelThreads > 1 || retryCount > 0)
{
Trace.WriteLine(string.Format("Controller {0}, DPRAM poll {1:0} ms, threads {2}, retries {3}",
this.controller.number,
duration.TotalMilliseconds,
parallelThreads,
retryCount));
}
}
}