EOFError with multiprocessing Manager

Question

I have a bunch of clients connecting to a server via 0MQ. I have a Manager queue used for a pool of workers to communicate back to the main process on each client machine.

On just one client machine having 250 worker processes, I see a bunch of EOFError's almost instantly. They occur at the point that the put() is being performed.

I would expect that a lot of communication might slow everything down, but that I should never see EOFError's in internal multiprocessing logic. I'm not using gevent or anything that might break standard socket functionality.

Any thoughts on what could make puts to a Manager queue start raising EOFError's?

Answer 1

For me the error was actually that my receiving process had thrown an exception and terminated, and so the sending process was receiving an EOFError, meaning that the interprocess communication pipeline had closed.