On Wed, Jun 1, 2011 at 1:13 PM, Martin Sustrik <[email protected]> wrote:

> Yes. The idea is that the client will resend the request in case the reply
> haven't arrived in time.
>
> The only alternative to that is blocking on send in XREP socket which would
> mean one slow/misbehaving client can block all the other clients.

Right. The problem in our use case is that the server will send
messages much faster than the network can handle them, and certainly
faster than most clients, so full queues will be a normal situation.

So XREP currently drops on HWM, meaning we have to handle message loss
at the application level, meaning that we have to treat "queue full"
as an exceptional condition rather than normal back-pressure. Which
means we have to use flow-control that ensures our queues are never
full, and treat message loss as a sign the flow control has failed.

OK, makes sense. If you agree with my analysis, I can eventually
document this as a pattern in the Guide.

-Pieter
_______________________________________________
zeromq-dev mailing list
[email protected]
http://lists.zeromq.org/mailman/listinfo/zeromq-dev

Reply via email to