The service control manager (SCM) controls a service by sending service control events to the service's control handler routine. The service must respond to control events in a timely manner so that the SCM can keep track of the state of the service. Also, the state of the service must match the description of its state that the SCM receives.
Due to this communication mechanism between a service and the SCM, you must be careful when using multiple threads in a service. When a service is instructed to stop by the SCM, it must wait for all the threads to exit before reporting to the SCM that the service is stopped. Otherwise, the SCM can become confused about the state of the service and might fail to shut down correctly.
The SCM needs to be notified that the service is responding to the stop control event and that progress is being made in stopping the service. The SCM will assume the service is making progress if the service responds (through SetServiceStatus) within the time (wait hint) specified in the previous call to SetServiceStatus, and the check point is updated to be greater than the checkpoint specified in the previous call to SetServiceStatus.
If the service reports to the SCM that the service has stopped before all threads have exited, it is possible that the SCM will interpret this as a contradiction. This might result in a state where the service cannot be stopped or restarted.
原来可能是多线程服务的代码设计的不周全，导致上文中红色部分描述的情况出现了。
那也就说明了，为什么有的时候重启之后就好了（操作系统重启，原来未退出的进程自然早没了，问题自然也就没了。）
因为，作为程序员，我想我也找到了相关的解决办法，用工具（Process Monitor或者ProcessExplorer）杀死线程，不过这可不是件容易干的活，对一般人来说，还是重启简单稳当点。