[boost::thread] Problems with condition objects
Hi: I'm having a bit of trouble with the implementation of the monitor pattern by means of using mutex and condition classes. I have a scheduler thread and a group of worker threads and the scheduling algorithm waits until some worker notifies that it has done the job. It seems that the call to condition::notify_one() must be in a specific order because if the worker calls notify_one before the scheduler arrives to the waiting line it behaves as if the condition object has lost the previous call to notify_one. Is this behavior the one we must expect? and, if so, can any one share a workaround. Thanks in advance. René
Rene Tarantino wrote:
Hi:
I'm having a bit of trouble with the implementation of the monitor pattern by means of using mutex and condition classes.
I have a scheduler thread and a group of worker threads and the scheduling algorithm waits until some worker notifies that it has done the job.
It seems that the call to condition::notify_one() must be in a specific order because if the worker calls notify_one before the scheduler arrives to the waiting line it behaves as if the condition object has lost the previous call to notify_one.
You haven't posted any code, but from what you say above it seems you believe that notify_one is "sticky". It's not -- it only wakes a thread that is already waiting on a condition variable. So, 'notify_one' and 'notify_all' should be viewed as 'something changed' notification, and you should maintain all 'interesting' state yourself, and check it like this: while(!something_interesting) cond.wait() - Volodya
participants (2)
-
Rene Tarantino
-
Vladimir Prus