[zeromq-dev] zmq_ipc

[Martin Sustrik]

Hi Andrew,

> i am currently seeing the following scenario:
>
> one process opens a PUSH socket on an ipc address /x/y
> several processes open a PULL socket on that same /x/y
> traffc flows as expected for 24-48 hrs.
> then traffic stops.
> the sending process is stalled in send, and all the reading processes
> are stalled in a recv.
>
> it is not readily reproducible.
> traffic rate is not well known at this point, but an offpeak number
> is 1.2M messages/s (= 173MB/s). the high point might be 3x or 4x that.
>
> any hints on how i might proceed?
> shoudl i try another transport?

I think the important point here is that the sender is stuck in send. 
This should never happen with PUB socket. (It should drop messages if 
the internal pipe gets full.)

Thus, it would be useful to attach the stuck sender to the debugger and 
check out what exactly is happening in the sender thread.

Btw, you haven't mentioned your 0MQ version.

Martin