bmenkveld wrote on Thursday, November 27, 2008:
I ran into exactly the same thing in a recent application. I also started to think about a semaphore to trigger the waiting task into inspecting the various queues. However, I didn’t like the extra overhead of checking all the queues every time the semaphore is triggered.
Instead, I changed the design of my application. Since I was writing the data "producers" as well as the "consumer", I changed the design to a single queue containing a union of the different types of data produced by the various producers. The producer tasks all send to the one queue, and the consumer task simply waits on the one queue.
This worked for me because there was not a huge difference in data size from the different producer tasks. If you had a high frequency producer sending 1-byte data, and a low-frequency producer sending 100-byte data, this would not be a good plan since each 1-byte data message would be 100 bytes in size, so you’d have a lot of needless data copying going on.