Why do reusing TcpChannel reduce redelivered messages

[mi-mi36]

I executed two RabbitMQ consumer programs. One(A) is a lot of redelivered message arise, and the other(B) is rarely arise. Why is it?

The difference between them is whether TcpChannel is reused or not. Why is it affect redelivered behavior?

[paolopas]

hi @mi-mi36,
I don't understand on which loop you are doing the ev_break in the AMQPConsumer::timeoutCb which if I read correctly is a static method, so it shouldn't be able to access the local variable _loop that you build in the AMQPConsumer(). I'm afraid using the magic macro EV_A_ is not suitable in this case, maybe the ev_break(EV_A_ EVBREAK_ONE) is breaking on the default loop EV_DEFAULT and not on the one you allocated in the constructor.
I'm not a libev expert, but I don't understand why you allocate a loop in the ctor when all the callbacks call the same static method. If you don't use multiple AMQPConsumer instances at the same time, then you don't even need multiple loops and you could use ev_default_loop(0) instead of ev_loop_new(0).
I would do it as an experiment, maybe you are the victim of a race condition and the dynamic allocation of the TCPChannel has little to do with your problem.