[zeromq-dev] PUSH/PULL over inproc and nonblocking sends

Martin Sustrik sustrik at 250bpm.com
Thu Aug 4 07:18:58 CEST 2011


On 08/03/2011 04:15 PM, Peter Bourgon wrote:

In short, swapping inproc for tcp should not result
> in a change of behavior, in this instance.
>
> Comments? Counter-arguments?

I would say there's no change in the behaviour here, it's just a timing 
issue. There's a buffer on the push side (set to 1) and a buffer on the 
pull side (set to infinite).

If 0MQ is able to pass messages from the push-side buffer to the 
pull-side buffer fast enough, you'll never get EAGAIN (the pull-side 
buffer is infinite).

If, on the other hand, you are publishing messages faster than 0MQ 
passes them from push buffer to pull buffer, you'll get EAGAINs (the 
push side buffer is limited to 1 message).

Martin



More information about the zeromq-dev mailing list