[zeromq-dev] Proposal for synchronous connect option
Pieter Hintjens
ph at imatix.com
Tue Feb 22 11:47:01 CET 2011
On Tue, Feb 22, 2011 at 11:19 AM, Martin Sustrik <sustrik at 250bpm.com> wrote:
> Well, I think you are fighting a losing battle here. Even if you wait for
> connect() confirmation, there's no guarantee that a disconnection won't
> happen immediately aftrwards. It won't work with transports with inherently
> async connect (pgm), it won't work if there are devices in the middle etc.
It's specifically for TCP and we already have a mechanism to handle
disconnections, namely durable subscribers (which are also only TCP).
As for devices, it's a bogey man, essentially you're saying
"everything becomes multicast" but that's only true for some (very
large case) scenarios, a small minority of real use cases.
One does not do synchronized, nor durable subscribers, with devices
and multicast.
So I'm not sure what the battle is, nor why it's a losing one. We
already do half the handshake at start up (and again that won't work
with PGM, and it won't work across devices, yet you implemented it).
I'm wondering if the other half can't be profitable as well.
-Pieter
More information about the zeromq-dev
mailing list