From one of GA’s prior messages I quote:
The final issue, and one I’m on the fence about, is the idea to make nn_bind()
synchronous — at least the “listen and bind to address” part, while the
underlying accept() would be asynchronous. I have ideas as to how to do this,
but it involves some larger surgery, and it isn’t 100% clear to me that we
should hold back 1.0 pending this change, which IMO would represent some risk.
Would this proposed change remove the need to pause between bind/connect and
the first use?
RC
On Sun, Jun 5, 2016 at 11:47 AM, Garrett D'Amore <garrett@xxxxxxxxxx
<mailto:garrett@xxxxxxxxxx>> wrote:
The first message the server sends is lost, because the clients haven’t
connected yet when the send takes place.
If you ask the server to wait for > 1 sec (perhaps 2 seconds) before sending
the first message, then that will give the clients a chance to connect, and
it will be fine.
- Garrett